Content uploaded by Riad Balaghi
Author content
All content in this area was uploaded by Riad Balaghi
Content may be subject to copyright.
Agrometeorological
Cereal Yield Forecasting
in Morocco
Riad BALAGHI
Mohammed JLIBENE
Bernard TYCHON
Herman EERENS
National Institute for Agronomic Research
Morocco
2013
The right of publication in print, electronic and any other form and in any language is reserved by INRA -
Morocco. Short extracts from INRA publications may be reproduced without authorization, provided that the
complete source is clearly indicated. Editorial correspondence and requests to publish, reproduce or translate
this publication in part or in whole should be addressed to:
INRA Avenue Ennasr Rabat, Morocco
BP 415 RP Rabat, Morocco
www.inra.org.ma
Tel : +212 537 77 09 55
Fax : +212 537 77 00 49
National Institute for Agronomic Research (INRA)
Division of Information and Communication
2013 edition
ISBN: 978 - 9954 - 0 - 6683 - 6
A word from the Director
Food security, in Morocco as in many parts of the world, relies heavily on cereal production which
fluctuates depending on weather conditions. Cereal production in Morocco does not meet
consumption needs of the ever growing population, leading to massive import of grain to fill the
gap. The cost of imported grain is high and is expected to increase due to several factors: (1) the
continuous rising of prices in conjunction with increasing world population and inputs prices, (2) the
use of cereal grains for bio-fuel production, and (3) the negative impacts of climate change. In
Morocco cereals are highly exposed to climatic risks, since they are mainly produced in arid and
semi-arid lands, characterized by limited soil and water resources to satisfy crop growth
requirements.
The Green Morocco Plan which is the strategy of the Government of Morocco for the agricultural
sector, aims at insuring food security through a sustainable improvement of productivity while
saving water and soil resources. It is an ambitious but achievable objective, given the range of
available agricultural technologies and know-how developed at National Institute for
Agronomic Research (INRA), in the field of adaptation to drought and land valorization. These
include weather risk management tools for decision-making at both farm and policy levels, in
addition to improved cultivars, agronomic and plant protection practices. Weather risk
management refers to the drought issue in particular, for which INRA developed operational
approaches and tools to monitor the season and forecast cereal yields, so as to timely undertake
appropriate mitigation measures, and deal with the international cereal market. These
achievements are the result of INRA’s investment in the field of agrometeorology oriented toward
operational cereal yield forecasting systems.
The present work is the result of a long term research programme started in the 1990s, sustained
by our institution and involving committed human resources which are aware of the critical issues
challenging the agricultural sector in Morocco. The research was supported by international
cooperation with key European institutions (Ulg, JRC, VITO, Alterra, UNIMI). This research was
fruitful, since a national cereal yield forecasting system called “CGMS-MAROC” was developed and
implemented according to international quality standards. The system is currently operational, and
a dedicated Web viewer was developed (www.cgms-maroc.ma). It is managed autonomously by a
consortium composed of three national institutions (INRA, DMN, DSS) bound by a mutual strategic
agreement. Cereal yields can presently be forecasted three months ahead of harvests, offering a
large flexibility for decision making.
The document I have the pleasure to present, relates the history and findings of the operational
agrometeorology research carried out at INRA.
Pr. Mohamed BADRAOUI
Director of INRA
Preface
The science of agrometeorological crop forecasting is also an art, because different experts come
up with different solutions, where efficiency, accuracy and elegance all play a part. It is usually
practiced in collaboration by several institutions, under the pressure of fixed deadlines, for mostly
critical and demanding customers. Beyond the limitations that affect data, methods and models,
the practitioners of this art have learned to provide crop yield estimates on time for decision
making, and with acceptable errors.
Over the last fifteen or twenty years, due to significant advances in computer science and
hardware, remote sensing, geo-statistics, and Geographic Information Systems, there has been a
tendency to develop “industrial” or “brute-force” methods to monitoring crops and predicting
yields. Some people are satisfied with the approach; they can easily be recognized: they produce
semi-automatic graphs and maps of the current season based on satellite indices and compare
them with a “historical average” assumed to represent “normal crop conditions.”
The authors of the present publication are of a different calibre. While using modern knowledge
and techniques, they also know by experience that yield forecasting remains in essence an
agronomic application. Beyond describing crop conditions, they emphasize the need to understand
climate and its variability at all temporal and spatial scales relevant to agriculture, and the
variability brought about by farming activities and practices, which are often controlled by
economic or policy factors. This is what characterizes this little book: it explains the reasoning and
the methods, ranging from traditional ones like frequency analysis of rainfall, to more advanced
features like the use of satellite data and indices, and makes us understand why and how they can
be used to quantify the effect of factors that affect crop yields.
I have known the authors for a long time. Their art of predicting crop harvests stems from
experience in the real world. They do not neglect traditional techniques which are often ignored, or
unknown to the practitioners of recent “industrial” approaches. One can be convinced by
consulting the bibliography where classics like “De Martonne, Célérier and Charton (1924)” get
along well with recent ones like “Bénichou and Le Breton (1987)” and “de Wit, Duveiller and
Defourny (2012)”.
By their meticulous way of trying to understand, Riad Balaghi and his colleagues somehow remain
traditional natural scientists. This can be seen in some of their empirical rules, like the one
providing the level of accumulated rainfall above which the relation between yields and rainfall
fades out. It can also be seen in numerous original figures that literally dissect data; they are unique
and found nowhere else in the literature.
The work of Balaghi, Jlibene, Tychon and Eerens deserves ample diffusion, beyond Moroccan
boundaries. First, in schools that teach crop yield forecasting, at various levels, under one name or
another. Students will learn how to thoroughly observe data and extract all relevant information
they contain. Second, and more importantly, professionals and practitioners, even experienced
ones, will be surprised to find “know-how” and original analyzes that are overlooked in most texts.
As to me, I am delighted that such a publication of “advanced popular science” on crop yield
forecasting is now made widely available. I have no doubt that it will meet the success that it
deserves.
Dr. René GOMMES
Doctorate in bio-geochemistry (1977).
Agroclimatologist at FAO since 1980.
Coordinator of the Agrometeorology Group of FAO from 1994 to 2007,
and Leader of the “Climate impact team” of FAO, from 2008 to 2010.
Senior scientist in the FoodSec Unit at the EC Joint Research Center at
Ispra, from 2010 to 2012.
Since 2012, visiting professor at the Chinese Academy of Sciences.
Member of the European Society of Agronomy (ESA) and the
International Society of Biometeorology (ISB).
Acknowledgement
Research on agro-meteorology was initiated at the Regional Agronomic Research Center of Meknes
(INRA-Meknes), in 1992, as part of the research actions of the “High Rainfall Research Programme”
(Programme Bour Favorable) coordinated then by Dr Mohammed Jlibene, until 2002. It has been
later pursued under supervision of Dr Riad Balaghi, as head of the research unit “Agronomy and
Plant Physiology” at INRA-Meknes, from 2002 to 2008, and as head of “Department of Environment
and Natural Resources”, at INRA-Rabat, from 2008 to date.
Since 2008, research in agrometeorology has received moral support from the Director of INRA, Pr
Mohamed BADRAOUI, not just as a manager but also as a scientist convinced by the interest of
supporting this research project for the benefit of food security in Morocco.
This research project needed considerable amount of agro-meteorological and agronomical
databases, as well as scientific and technical help from national and international cooperation.
Institutions involved in the project were: “Direction de la Stratégie et des Statistiques" (DSS,
Morocco), “Direction de la Météorologie Nationale (DMN, Morocco), “University of Liege (Arlon
Campus Environnement, former Fondation Universitaire Luxembourgeoise, ULg - Belgium)", "Joint
Research Centre of the European Commission " (JRC, Italy), "Flemish institute for technological
research" (VITO, Belgium) and "Food and Agriculture Organization of the United Nations" (FAO,
Italy).
The project was partly financed by the Belgium Agency of Development” (CTB, Belgium) as well as
the European Union thanks to the Seventh Framework Programme (FP7 - Project « Crop Monitoring
as an E‐Agricultural Tool for Developing Countries », E‐AGRI).
The idea of this research project has matured with the help and perseverance of INRA researchers
who can identify themselves and find the expression of their common output. Researchers of great
scientific and technical quality have contributed to this achievement with varying degrees:
Mr. Moha MARGHI, former Director of “Direction Provinciale de l’Agriculture de Meknes” and later
former general secretary of the Ministry of Agriculture and Marine Fishery (MAPM), was the first
official to have requested INRA-Meknes to present scenarios of drought consequences on cereal
production in the Meknes region during the dry season of 1994-1995. By approaching us, he
inspired us with the idea of investing in research for forecasting cereal yields.
From the Joint Research Centre of the European Commission (JRC): Dr. Giampiero GENOVESE, Dr.
Bettina BARUTH, Dr. Mohamed EL AYDAM, Dr Giovanni NARCISO, have supported the research
agreement between JRC and INRA in the field of harvest forecasting as well as the E-AGRI research
project.
From the “Direction de la Stratégie et des Statistiques” (DSS, Rabat): Mr. Redouane ARRACH and
Mr. Mustafa TAHRI, who were committed to provide agricultural statistical data and helped with
the research collaboration agreement between DSS, INRA and DMN, as part of the E-AGRI project.
From the “Direction de la Météorologie Nationale” (DMN, Casablanca): Mr. Tarik EL HAIRECH and
Mr. Rachid SEBBARI have concretized the research agreement between DMN and INRA and were
formally committed to the E-AGRI project.
From the E-AGRI: Dr. Qinghan DONG, Dr. Allard DE WITT, Mr. Steven HOEX, helped in the
adaptation of “CGMS” to Moroccan context. Dr Mouanis LAHLOU, from Institute of Agronomy and
Veterinary Hassan II, developed the Web viewer of the “CGMS-MAROC” forecasting system.
From INRA: Dr. Rachid MRABET, Mr. Hassan BENAOUDA, Dr. Hamid MAHYOU, Dr. Sliman ELHANI,
Dr. Rachid DAHAN, Dr. Hassan OUABBOU and Mr. Mohamed BOUGHLALA have exchanged data and
information with us and participated in different research projects some of which were not directly
related to yield forecasting per se.
The authors would also like to thank all colleagues who directly or indirectly contributed in our
research work, either by providing agro-climatic data or contributing to the overall thinking process,
particularly: Mr. Abdelaziz EL OUALI (former Head Agrometeorology Service at DMN), for his advice
on climatology in Morocco; Mr. Wolfgang GÖBEL (former researcher at INRA) for exchange
information on agro-meteorology and for common publications on the “Atlas Agroclimatique du
Maroc”; Dr. Mohamed EL MOURID (former researcher and research manager at INRA) who initiated
the first research studies on crop modelling at INRA; Mr. Hamid FELLOUN and Ms. Fatiha SELOUANI
(MAPM) who provided us with climatic data of the Ministry of Agriculture and Marine Fishery.
Least but not last, thanks go also to Mr. Chafik KRADI, head of the “Division de l’Information et de la
Communication” at INRA and his staff, particularly Mr. Reddad TIRAZI who helped in the diffusion
of this document.
Our special thank goes to Dr. René GOMMES, ex senior scientist at FAO with whom we were
honored to work on several occasions on operational agro-meteorology and climate change and
who was kind to preface this document.
The authors
SUMMARY
The present document provides a summary of research work carried out, at National Institute for
Agronomic Research of Morocco (in French, Institut National de la Recherche Agronomique - INRA),
since early 1990s, in the area of operational agrometeorology oriented toward forecasting crop
harvests. Forecasting the production of crops early before harvest allows decision makers to be
prepared in advance for eventual consequences of abnormal deviations of the climate, particularly
for strategic commodity crops to food security like cereals. To our knowledge, to date there is no
official method to forecast cereal production in Morocco on the basis of agrometeorological data.
However, cereal productions are estimated based on a sampling method some weeks before
harvest, every year by the Ministry of Agriculture and Marine Fishery (MAPM) through the
Direction of Strategy and Statistics (in French, Direction de la Stratégie et des Statistiques - DSS). It
is a direct method, precise, and applied directly before harvest, but requires consequent human
and financial resources. The need to elaborate an indirect method to early forecast yields that is
fast and economical, has been understood at INRA as early as in 1995, triggered by the severe
drought of that particular season, described as the worst dry season of the 20th century in Morocco.
Neither the classical frequency analyses of the climate used to identify seasons of close similarity to
1994-1995 season, nor the available mechanistic models for crop forecasting used in developed
countries, have been able to monitor crop development during that season and a fortiori predict
the catastrophic harvest of 1995. Therefore, it became necessary to come up with a new approach
for forecasting cereal yields using an innovative methodology which combines empirical and
statistical approaches with agronomic and meteorological expertise. First we had to study the
interaction between the climate and the cereal crops behaviors, particularly climatic and crop
cycles were analyzed together in a series of long term data, initially for Meknes region where the
first two authors were posted, extended later to other regions of Morocco. Preliminary results
indicated for the first time in Morocco that inter-annual variation of cereals yields could be
explained by variation in the amount of rainfall cumulated during the crop cycle, with a relatively
high accuracy. The relationship could be enhanced by partitioning the season into three or more
phases. In collaboration with the University of Liège (ULg, Belgium) and later with the Joint
Research Centre of the European Commission (JRC), a new indicator was identified as highly
correlated to cereal yields, which is the Normalized Difference Vegetation Index (NDVI) derived
from satellite images. Unlike many European countries, this index was highly correlated to cereal
yields in Morocco, mainly due to the aridity of Moroccan climate and the predominating coverage
of cereals of agricultural areas. NDVI is correlated with cereal yields as long as cropping season
rainfall did not exceed 550 mm, which explains the irrelevance of NDVI to forecast crop yields in
Northern Europe. The combination of both rainfall and NDVI allowed forecasting of cereal yields as
early as March, three months before harvest, and at a low cost, with a level of accuracy similar to
the one of the direct sampling method used at crop maturity by DSS. These astonishing results have
led INRA to publish for the first time in Morocco three crop forecasting bulletins between 2009 and
2011, in collaboration with JRC. In these bulletins, an approach combining four individual
approaches was used: (1) similarity approach using rainfall and/or NDVI as criteria of comparison,
(2) regression models using rainfall and NDVI as predictors of cereal yields, and (3) the JRC
approach which is based on a simulation model of crop growth called WOFOST. The deterministic
model WOFOST is now being adapted to the Moroccan agro-climatic context and incorporated in
an operational forecasting system. To ensure durability of the system, a strategic partnership
between INRA, DSS and DMN was formalized in addition to that bounding INRA and JRC. This new
collaboration has allowed establishment of the first national cereal yields forecasting system
named “CGMS-MAROC”, based on the combined approach developed in the present document.
The system is carried out by the three national institutions (INRA, DSS and DMN), leading to the
edition of a fourth bulletin of cereal yields forecasts issued for the 2012 season. The combined
approach can be extended to forecast yields for other crops in morocco as well as in countries of
similar climatic pattern, provided some adjustments. In parallel to yield forecasting, a new field of
research can be explored, dealing with estimating cropped areas, using low resolution and
inexpensive satellite images.
Key words: Agrometeorology, yield forecasting, similarity analysis, NDVI, rainfall, temperature,
cereals, soft wheat, durum wheat, barley, drought, Morocco, INRA.
TABLE OF CONTENTS
I. INTRODUCTION .............................................................................................................................................. 1
II. APPROCHES OF ANALYSIS .............................................................................................................................. 9
1. Data base sets used ............................................................................................................................................... 9
1.1. Climatic data ............................................................................................................................................... 10
1.2. North Atlantic Oscillation Index .................................................................................................................. 11
1.3. Vegetation index from satellite imagery ..................................................................................................... 12
1.4. Administrative boundaries of Morocco in GIS format ................................................................................ 13
1.5. Land cover maps ......................................................................................................................................... 13
1.6. Agricultural statistics ................................................................................................................................... 14
2. Methods of Analyses ........................................................................................................................................... 15
2.1. Exploratory treatment of rainfall data ........................................................................................................ 15
2.2. Frequency analysis of rainfall ...................................................................................................................... 17
2.3. Ombrothermic index of bagnouls and gaussen .......................................................................................... 17
2.4. Length of the growing period ..................................................................................................................... 18
2.5. Analysis of breaking point in chronological series of rainfall ...................................................................... 19
2.6. Productivity of rain water ........................................................................................................................... 19
2.7. processing of satellite images ..................................................................................................................... 20
2.8. Cereal yield forecasting approach .............................................................................................................. 21
2.8.1. Technological trend .................................................................................................................................... 23
2.8.2. Similarity analysis ........................................................................................................................................ 24
2.8.3. Linear regression analysis ........................................................................................................................... 25
III. ANALYSES OF MOROCCAN CLIMATE ............................................................................................................ 27
1. Influence of North Atlantic Oscillation................................................................................................................ 28
2. Climate change..................................................................................................................................................... 29
3. Temperature ........................................................................................................................................................ 31
3.1. Spatial variation of temperature................................................................................................................. 31
3.2. Inter-annual variation of temperature ....................................................................................................... 33
3.3. Seasonal variation of temperature ............................................................................................................. 35
4. Rainfall ................................................................................................................................................................. 36
4.1. Spatial Variation of rainfall ......................................................................................................................... 36
4.2. Interannual variation of rainfall .................................................................................................................. 38
4.3. Seasonal variation of rainfall....................................................................................................................... 40
4.3.1. Frequency analysis of dekadal rainfall ........................................................................................................ 41
4.3.2. Linear approximation of dekadal rainfall .................................................................................................... 43
4.4. Seasonal variation of rainfall....................................................................................................................... 45
5. Growing season from Ombrothermic diagram ................................................................................................... 48
6. Length of the growing Period .............................................................................................................................. 52
IV. ANALYSYS OF CEREAL PRODUCTION ............................................................................................................ 54
1. End uses of cereals ............................................................................................................................................... 54
2. Geographical distribution of cereals ................................................................................................................... 55
3. Technological trend ............................................................................................................................................. 58
4. Cereal crop development .................................................................................................................................... 63
V. AGROMETEOROLOGICAL ANALYZES ............................................................................................................ 65
1. Relationship between transpiration and growth ............................................................................................... 65
1.1. Relationship between rainfall and cereal yields ......................................................................................... 67
1.2. Relationship between rainfall and cereal area ........................................................................................... 68
1.3. Relationship between rainfall and rainwater productivity ......................................................................... 71
1.3.1. At national scale .......................................................................................................................................... 71
1.3.2. At agro-ecological zone scale ...................................................................................................................... 74
2. Agricultural drought ............................................................................................................................................ 74
3. Relationship between growth and rainfall ......................................................................................................... 76
4. Relationship between NDVI and rainfall ............................................................................................................. 77
5. NDVI profile.......................................................................................................................................................... 79
5.1. Intra and inter-annual Variation of NDVI .................................................................................................... 79
5.2. Regional variation of NDVI .......................................................................................................................... 80
VI. CEREAL YIELD FORECASTING APPROACHES ................................................................................................. 82
1. Non-parametric approach ................................................................................................................................... 83
2. Similarity approach .............................................................................................................................................. 84
2.1. Similarity analysis using rainfall .................................................................................................................. 84
2.2. Similarity analysis using NDVI ..................................................................................................................... 90
2.3. Similarity analysis using NDVI and rainfall combined ................................................................................. 94
3. Regression models approach............................................................................................................................... 97
3.1. Forecasting using rainfall as predictor ........................................................................................................ 98
3.2. Forecasting using agro-climatic indices .................................................................................................... 104
3.3. Forecasting using remote sensing............................................................................................................. 105
3.4. Forecasting using weather data and NDVI ................................................................................................ 111
4. Combined approach ........................................................................................................................................... 113
VII. INSTITUTIONALIZATION AND OPERATIONALIZATION OF THE FORECASTING SYSTEM .............................. 115
VIII. CONCLUSIONS AND OUTLOOK ................................................................................................................... 118
IX. REFERENCES ............................................................................................................................................... 121
LIST OF TABLES
Table 1: Classes of rainfall and corresponding aridity levels with administrative provinces
included in each class. Rainfall data are averages of 1988-2005. ....................................... 4
Table 2: Change point analysis in chronological series of cumulated rainfall between
September and May, at the provincial level (Source: Balaghi, 2006). ............................... 30
Table 3: Linear regression models applied to cumulated dekadal rainfall over the growing
season (from September to May) at the country level. ................................................... 44
Table 4 : Ombrothermic Index, as a ratio between precipitation (P) and temperature (T), by
synoptic station. Unit of P/T is in mm/°C. ...................................................................... 50
Table 5 : Main uses of cereal grains in Morocco (Source: ONICL, 2012). ......................... 55
Table 6: Partitioning of cereal production, area and yield, by agro-ecological zones
(Average of 1990 to 2011). .......................................................................................... 56
Table 7: Yield trend over years (Quintal/ha.year), due to technology improvement of the
three main cereals, by agro-ecological zone, and coefficients of determination (R2) of the
regression lines. (Data series of 1979 till 2006; Source: Balaghi and Jlibene, 2009). ......... 59
Table 8 : Average and Potential Rain Water Productivity (RWP) of cereals, at national level.
RWP is equal to the slope of the relationship between yield and rainfall. ......................... 73
Table 9: Rain Water Productivity (gram/liter), for the three winter cereals, at the national
and agro-ecological zones levels (average from 1988 to 2011). ...................................... 74
Table 10 : Non-parametric relationship between wheat grain yield (Q/ha) and drought
indicators, during critical season’s phases (beginning, middle and end of cycle). (0: No
drought; 1:Drought). .................................................................................................... 83
Table 11 : Coefficient of determination (R2) of the linear regression models between cereals
yields (soft wheat, durum wheat and barley) and cumulated rainfall over the cropping
season (from October till March), at the level of the agro-ecological zones (Data from 1988
to 2008). ................................................................................................................... 102
Table 12: Coefficient de determination (R2) of the regression models between grain yields
of the three winter cereals (soft wheat, durum wheat and barley) and average NDVI
(SPOT-VEGETATION) from February till March and from February till April, at the level of
the agro-ecological zones of Morocco (Data of 1999 to 2011). ...................................... 109
Table 13: Schematic table of the combined approach and predictors developed for cereal
yield forecasting in Morocco. Shaded areas indicate the predictors used for each method.
................................................................................................................................. 114
LIST OF FIGURES
Figure 1: Positive phase (left) and negative one (right) of the North Atlantic Oscillation. At
the positive phase, drought reigns over the Mediterranean region, while storms are
frequent over Europe. On the contrary, at negative phase, the Mediterranean region enjoys
a humid weather while Europe is less humid.
(Source:
http://www.ldeo.columbia.edu/res/pi/NAO/).
................................................................. 12
Figure 2: Country average dekadal (10-days) rainfall, of a series of crop seasons from 1988
till 2011. ...................................................................................................................... 16
Figure 3 : Crop Growth Monitoring System (CGMS) for crop harvests forecasting. ............ 22
Figure 4: Hypothetical example showing how crop yields (red curve) can be influenced by
weather, technology innovation, policy, extreme factors and general trend (Source:
Gommes
et al
., 2010). .................................................................................................. 24
Figure 5: Correlation between the North Atlantic Oscillation (NAO) index, averaged from
September to February and cumulated precipitation from September to May in Morocco
(data of 1979 to 2011) ................................................................................................. 29
Figure 6: Maximum temperature of the warmest month (A), minimal temperature of the
coldest month (B), annual mean temperature (C) and temperature annual range (D) in
Morocco (these maps were created from data taken from www.worldclim.org at the spatial
resolution of 1 km, Series from 1950 to 2000, Hijmans
et al
., 2005) ............................... 32
Figure 7 : Sum of daily average temperatures, in growing degree days, along cereal
cropping season, at the country level. Growing cycles from 1999 to 2009 (year of harvest)
are sorted out in descending order of cumulated temperature. ....................................... 34
Figure 8 : Evolution of coefficient of variation of average inter-annual dekadal temperatures
at the country and Meknes levels (data of 1999 to 2009). .............................................. 35
Figure 9 : Average maximal (red curve), mean (yellow curve) and minimal dekadal
temperature (blue curve), along the cereal growing season. Red dots correspond to
dekadal values of maximal temperature and blue dots to minimal temperature, from 1999
until 2009. ................................................................................................................... 36
Figure 10 : Country average rainfall cumulated between September and April. The map
created from data of www.worldclim.org (Hijmans
et al
., 2005). Red dots refer to main
cities of Morocco. ......................................................................................................... 37
Figure 11 : Inter-annual variation of cumulated rainfall between September and May, at the
country level, for 1988 to2011. ..................................................................................... 38
Figure 12 : Geographic localization of the 35 synoptic weather stations used to generate
weather data of Morocco, on a background map of elevation provided from radar data of
«
Shuttle Radar Topography Mission
» (http://www2.jpl.nasa.gov/srtm/). ....................... 39
Figure 13 : Examples of dekadal rainfall distribution along the cropping cycle, for specified
contrasting synoptic weathers stations (averages of 1988 to 2010). ................................ 40
Figure 14 : Distribution of dekadal rainfall, at the country level, for 5 frequency levels (0.1;
0.3; 0.5; 0.7 and 0.9). Data series from 1988 to 2011. ................................................... 41
Figure 15 : Annual rainfall distribution (cumulated from September to June) at Meknes
(1932-2004), Khemisset (1986-2004), Settat (1910-2004) and Safi (1901-2004). The red
curve represents adjusted theoretical distribution curve to observed one (Balaghi
et al
.,
2005). ......................................................................................................................... 42
Figure 16 : Dekadal cumulated rainfall from September till May, at the country level.
Cropping cycles of 1988 to 2011 are arranged in ascending order of cumulated rainfall.
... 43
Figure 17 : Average increment rate of rainfall per dekad (x-axis, in mm/dekad), for 4
classes of cumulated rainfall (200 to 300, 300 to 400, 400 to 500 and > 500 mm). Data are
country averages of 1988 – 2011 series. ....................................................................... 45
Figure 18 : Monthly rainfall distribution, as influenced by the overall volume of rainfall in
Morocco (data of 1988 to 2005). ................................................................................... 46
Figure 19 : Monthly rainfall distribution for 15 contrasting synoptic weather stations
scattered along the Atlantic coast (data from 1988 till 2005). Stations are arranged in
ascending order of latitude. Red lines represent points of equal monthly rainfall, of 20 mm,
40 mm, 60 mm and 80 mm. ......................................................................................... 47
Figure 20 : Ombrothermic diagram, at the country level.
(Average rainfall and temperature,
from a series of 1999 to 2009).
..................................................................................... 48
Figure 21: Average Ombrothermic index (from September to May) as influenced by latitude.
Monthly data series of 1999-2009 for 25 different synoptic stations were used. ............... 49
Figure 22 : Geographic areas where cereal growing seasons extend from September until
March (areas in blue, green and brown colors), or until April (areas in green and brown
colors) or until May (areas in brown color), according to Ombrothermic index of Bagnouls
and Gaussen (1953). The map was generated from data of www.worldclim.org (Hijmans
et
al
., 2005). ................................................................................................................... 52
Figure 23 : Lengths and time limits of different periods within the cereal crop cycle: (1) soil
preparation (W in purple), (2) period of growth (green) which includes sowing period (S),
(3) humid period (light green), flowering (F) and grain filling (G); and (4) grain maturity
and ripening (M in pink). The humid period includes times of emergence (L), tillering (T),
and heading (H) of cereal crops in the north-western Morocco (Jlibene and Chafai, 2002).
Weather data are a series of 70 years from Arbaoua Meteorological station (Data source:
ORMVAL). .................................................................................................................... 53
Figure 24 : Map of average production (in Mega tons) of autumn cereals (soft wheat,
durum wheat and barley confounded) by province (data series from 1990 to 2010). ........ 57
Figure 25 : Country soft wheat yield variation and trend in Morocco (in green) (Source of
data: FAOSTAT and DSS). ............................................................................................ 60
Figure 26 : Curves of cumulative probability of soft wheat yields at farm level (real
situation) and at potential situations at Meknes (Source: Boughlala
et al
., 1994). ............ 61
Figure 27 : Technology trend or improvement achieved in 28 years for soft wheat in
Morocco. ..................................................................................................................... 62
Figure 28 : Technological trend or improvement achieved in 28 years for durum wheat in
Morocco. ..................................................................................................................... 62
Figure 29: Evolution of daily actual evapotranspiration of soft wheat, simulated on the basis
of temperature alone (Eta_T°) and of temperature in association with relative humidity
(RH), wind speed and global radiation, at Meknes (Balaghi, 2000). ................................. 66
Figure 30: Schematic figure illustrating the relationship between transpiration and biomass
production (Jlibene and Balaghi, 2009). ......................................................................... 67
Figure 31: Deviation (%) from mean values of cumulated rainfall over September to May
period, and main cereal yield (soft wheat, durum wheat and barley), at the country level
(data from 1988 to 2011); indicating the co-variation of rainfall and cereal yields. ........... 68
Figure 32: Deviation (%) from mean value of cumulated rainfall over September to May
period, and of main cereal area (soft wheat, durum wheat and barley), at the country level
(data from 1988 to 2011). ............................................................................................ 69
Figure 33: Coefficient of determination (R2) of the relationship between main winter cereal
area (soft wheat, durum wheat and barley) and cumulated rainfall during the cropping
season (starting in October in average), at the country level (Data series from 1988 to
2011). Average dekadal rainfall is plotted in second axis (blue bars), for illustration. ........ 70
Figure 34: Relationship between soft wheat country yield (Kg/ha) and cumulated rainfall
during the cropping season (mm) (data from 1988 to 2011). .......................................... 71
Figure 35 : Relationship between durum wheat country yield (Kg/ha) and rainfall during the
cropping season (mm) (data from 1988 to 2011). .......................................................... 72
Figure 36: Relationship between barley country yield (Kg/ha) and rainfall during the
cropping season (mm) (data from 1988 to 2011). .......................................................... 72
Figure 37 : Dekadal cumulated rainfall between September and May, of dry cropping
seasons. Periods of drought are recognized as flat segments of the curves (in bold).
Drought can occur at diverse periods of the crop cycle. .................................................. 75
Figure 38: Spatial distribution of cumulated dekadal rainfall (from September till May) for
the dry season of 1996-1997. Locations used cover 23 provinces of Morocco, scattered all
over the country (Source: Jlibene, 2011). ...................................................................... 76
Figure 39: Typical weather conditions, during wheat growing season. Long term average of
dekadal rainfall (series from 1988 to 2011) and dekadal temperature (series from 1998 to
2009) were used. Major development phases of wheat crop are also displayed (green
arrow). ........................................................................................................................ 77
Figure 40: Relationship between NDVI (NOAA-AVHRR), averaged over the period from
February till April and over agricultural zones, and cumulated rainfall over the cropping
season from September till April, in Morocco. The 345 dots in this figure are data from 23
weather stations, over the period from 1990 to 2004. .................................................... 78
Figure 41: Comparison of the vegetation status in Morocco, between a wet season (2005-
2006) and a dry season (1999-2000), using NDVI (SPOT-VEGETATION) of the 3rd dekad of
March. Dark green colored pixels indicate high NDVI values. .......................................... 79
Figure 42: NDVI profile, during the cropping season for all croplands of the country (SPOT-
VEGETATION) and for 12 cropping seasons (1999-2000 to 2010-2011), arranged in
descending order of observed cereal yields. ................................................................... 80
Figure 43: Average NDVI profile (SPOT-VEGETATION), during the cropping season and
over all agricultural lands, for the six agro-ecological zones of Morocco (Data from 1999-
2000 to 2010-2011). .................................................................................................... 81
Figure 44 : Cropping seasons, from 1987-1988 to 2011-2012, arranged in ascending order
of cumulated rainfall between 1st of September and 10th of April of the following calendar
year (Source: Balaghi
et al
., 2012). Numbers in x-axis correspond to years of harvest.
Green line indicates average cumulated rainfall over the 25 years. The red bar points to
cumulated rainfall of the latest season i.e. 2011-2012. ................................................... 85
Figure 45: Dendrogram of cluster analysis, discriminating cropping seasons from 1960-1961
to 1999-2000, with regard to monthly rainfall of 8 months, from September to April, for
Meknes region (Source: Balaghi, 2000). ........................................................................ 86
Figure 46: Cumulated rainfall over the cropping season (September till May) for the 1994-
1995 cropping season, at Meknes, compared to historical data from 1960-1961 to 1993-
1994. Seasons are arranged in ascending order of cumulated rainfall. ............................. 87
Figure 47: Similarity analysis of 2007-2008 cropping season, using cumulated dekadal
rainfall, at the country level. ......................................................................................... 88
Figure 48: Forecasted country yields of soft wheat, for cropping seasons of 2007-2008,
2008-2009, 2009-2010 and 2010-2011, based on similarity analysis of cumulated rainfall
since October. ............................................................................................................. 89
Figure 49: "MARSOP3" Web viewer, for agro-climatic analysis of the cropping season.
Deviation of NDVI (SPOT-VEGETATION), at the first dekad of April for agricultural lands,
from long term average NDVI is provided by this application. Comparison with other
seasons can also be performed. Spatial resolution of grids is 25x25 km (Source: Balaghi
et
al
., 2012). ................................................................................................................... 91
Figure 50: Similar seasons to the 2011-2012 cropping season, based on dekadal NDVI
(SPOT- VEGETATION) over agricultural lands, at the country level. The 2011-2012 cropping
season (in orange) is compared to historical seasons. The 2000-2001 cropping season is
the most similar to 2011-2012 with regard to NDVI profile from 1st of February to 1st of
April, followed by 2007-2008 and 2004-2005 (Source: Balaghi
et al
., 2012). .................... 92
Figure 51: Forecasted soft wheat grain yields at the country level, for cropping seasons of
2007-2008, 2008-2009, 2009-2010 and 2010-2011, based on similarity analysis and using
dekadal NDVI (NOAA-AVHRR) since 1st of February. ....................................................... 93
Figure 52: The "CGMS Statistical Toolbox" software, for cereal yield forecasting, adapted to
Morocco. ..................................................................................................................... 94
Figure 53: Similarity analysis of 2010-2011 cropping season, based on Principal Component
Analysis, using in "CGMS Statistical Toolbox" software CST (http://e-
agri.wikispaces.com/CGMSStatTool). ............................................................................. 95
Figure 54: Observed and forecasted durum wheat grain yield at the country level, based on
Principal Component Analysis, for the 2008-2009, 2009-2010 and 2010-2011 cropping
seasons, using NDVI (NOAA-AVHRR) and cumulated rainfall (Data from 1988 to 2011), at
dekadal time step. ....................................................................................................... 96
Figure 55: Coefficient of determination (R2), by step of one dekad, of the regression line
between grain yields of the three main cereals (soft wheat, durum wheat and barley) in
Morocco and cumulated rainfall since September till May, at the country level. Average
(1988-2011) cumulated rainfall is displayed in blue bars, for illustrating the steady increase
of R2 with rainfall during the season, from the 2nd dekad of October to the 2nd dekad of
March. ......................................................................................................................... 99
Figure 56 : Relationship between country yields of soft wheat, durum wheat and barley,
and cumulated rainfall over the cropping season (from September to March), for data series
of 1988 to 2008. ........................................................................................................ 101
Figure 57: Coefficient of determination (R2) of linear regression models between grain
yields of the three main cereals (soft wheat, durum wheat and barley) in Morocco and
rainfall cumulated over significant periods during the cropping season, at the country level
(data from 1988 to 2011).
R
2 is higher when using cumulated rainfall over 6 months, and in
partitioning the cropping season’s rainfall into two groups of three months and three
groups of 2 months each (in these two latter cases, lengths of colored bars correspond to
partial R
2 of multiple regressions).
.............................................................................. 103
Figure 58: Coefficient of determination (R2) of regression models between soft wheat grain
yields and agro-meteorological indices derived from AgroMetShell program of the FAO, in
Meknes province (Source: Balaghi, 2006). ................................................................... 104
Figure 59: Coefficient of determination (R²) of regression models between grain yields of
the three main cereals (soft wheat, durum wheat and barley) in Morocco, and dekadal
NDVI (SPOT-VEGETATION), at the country level (data from 1999 to 2011), by step of one
dekad, since 1st dekad of January. .............................................................................. 105
Figure 60: Coefficient of determination (R2) of regression models between grain yields of
the three main cereals (soft wheat, durum wheat and barley) in Morocco, and average
NDVI (SPOT-VEGETATION), at the country level (data from 1999 to 2011), by step of one
dekad, since the 1st dekad of February. ....................................................................... 107
Figure 61: Linear regression between country grain yields of the three main cereals (soft
wheat, durum wheat and barley) and average dekadal NDVI (SPOT-VEGETATION) between
1st of February and end of March, and between 1st of February and end of April (data of
1999 to 2011). ........................................................................................................... 108
Figure 62 : Mapping of four classes of coefficient of determination (R2) of the linear
regression models between yields of soft wheat, durum wheat and barley, and average
dekadal NDVI (SPOT-VEGETATION) from February till March and from February till April.
................................................................................................................................. 110
Figure 63 : Relationship between observed and forecasted yields of soft wheat at Meknes
and Settat, based on multiple linear regression models, involving NDVI, rainfall and
temperature (Source: Balaghi
et al.
, 2008). ................................................................. 112
Figure 64 : Increase of precision and accuracy of the country soft wheat yield forecast,
based linear regression models, using NDVI (SPOT-VEGETATION), as the cropping seasons
reach completion. Examples of the 2008-2009, 2009-2010 and 2010-2011 cropping seasons
are shown. ................................................................................................................ 113
Figure 65: The Web viewer of CGMS-MAROC (www.cgms-maroc.ma) for bio-climatic
monitoring of cereals, and preliminary statistical analysis of the cropping season. On the
left side window is displayed the cumulated rainfall over the season on agricultural lands
(10x10 km grid). On the right window is shown, the similarity analysis using cumulated
rainfall, in the district of Ben Ahmed (province of Settat). ............................................. 117
ACRONYMES
°C
Degree Celsius
AFI
Area Fraction Image
AGRIMA
Agriculture Maroc
AgroMetShell
FAO tool for monitoring crops ftp://ext‐
ftp.fao.org/sd/reserved/agromet/AgroMetShell/
Alterra
Research Institute at the University of Wageningen, Netherlands
http://www.alterra.wur.nl
AURELHY
Analysis Using RELief for Hydro‐meteorology
AVHRR
Advanced Very High Resolution Radiometer
CGMS
Crop Growth Monitoring System
www.marsop.info/marsopdoc/cgms92/1_en.htm
CRTS
Centre Royal de Télédétection Spatiale www.crts.gov.ma
CST
CGMS Statistical Toolbox http://e-agri.wikispaces.com/CGMSStatTool
CV
Coefficient of Variation (%)
DMN
Direction de la Météorologie Nationale www.marocmeteo.ma
DMP
Dry Matter Productivity www.geoland2.eu
DPA
Direction Provinciale de l’Agriculture
DPAE
Direction de la Programmation et des Affaires Économiques, currently
DSS
DSS
Direction de la Stratégie et des Statistiques www.agriculture.gov.ma
E‐AGRI
A research Project on Crop Monitoring as an E‐agriculture tool in
Developing Countries www.e‐agri.info
ENSO
El Niño–Southern Oscillation or El Niño/La Niña–Southern Oscillation
ENVISAT
Environment Satellite, European satellite launched in 2002 to observe
Earth
ETa
Actual Evapotranspiration
ETP
Potential Evapotranspiration
EU
European Union
EVI
Enhanced Vegetation Index
FAO
Food and Agriculture Organization of the United Nations:
http://www.fao.org
FAOCLIM
FAO database of monthly agro-climatic observed and calculated
parameters http://www.fao.org/nr/climpag/pub/EN1102_en.asp
fAPAR
Fraction of solar radiation absorbed by plants
g
Gram, unit of weight
GHCN
Global Historical Climatology Network
GIS
Geographic Information System
GLC2000
Global Land Cover for the year 2000
GlCropV2
Land cover map made of in collaboration with VITO
ha
Hectare (1 hectare = 10.000 m²)
INRA
National Institute for Agronomic Research: www.inra.org.ma
JRC
Joint Research Center: http://ec.europa.eu/dgs/jrc/index.cfm
Kg
Kilogram
Km
Kilometer = 1000 meters
l
Liter (1 l = 1 dm3)
LAI
Leaf Area Index
MAPM
Ministry of Agriculture and Marine Fishery www.agriculture.gov.ma
MARS
Monitoring Agricultural ResourceS mars.jrc.ec.europa.eu
MERIS
Medium Resolution Imaging Spectrometer
MIAC
Mid America International Agricultural Consortium
mm
Millimeter rainfall (1mm=1 liter per square meter)
MODIS
Moderate Resolution Imaging Spectro-radiometer
http://modis.gsfc.nasa.gov/
NAO
North Atlantic Oscillation
NDVI
Normalized Difference Vegetation Index
NOAA‐AVHRR
The Advanced Very High Resolution Radiometer sensor carried by the
National Oceanic and Atmospheric Administration satellite
ONICL
Office National Interprofessionnel des Céréales et
Légumineuses www.onicl.org.ma
ONSSA
Office National de la Sécurité Sanitaire des produits Alimentaires
www.onssa.gov.ma
ORMVA
Office Régional de Mise en Valeur Agricole
PCA
Principal Component Analysis
PROBA‐V
Project for On‐Board Autonomy – VEGETATION sensor to be launched
in 2013 http://www.vgt.vito.be/
PSDA
Projet de Soutien au Développement Agricole dans les ORMVA
Q
Quintal (1 quintal = 100 Kg)
R²
Coefficient of determination
RWP
Rain Water Productivity
RWUE
Rain Water Use Efficiency
SEEE
Secrétariat d’État chargé de l’Eau et de l’Environnement
http://www.minenv.gov.ma
SPOT-
VEGETATION
Programme conceived to allow daily monitoring of terrestrial
vegetation cover through remote sensing, at regional to global scales
http://www.vgt.vito.be/pages/mission.htm
SRTM
Shuttle Radar Topography Mission: www2.jpl.nasa.gov/srtm
ULg
University of Liege (Arlon Campus Environnement)
http://www.facsc.ulg.ac.be/cms/c_636656/en/arlon-campus-
environnement-home
UNDP
United Nations Development Programme
USAID
United States Agency for International Development
VITO
Flemish institute for technological research, Belgium www.vito.be
WMO
World Meteorological Organization http://www.wmo.int
WOFOST
WOrld FOod STudies is a simulation model for the quantitative analysis
of the growth and production of annual field crops
http://www.wageningenur.nl/en/Expertise-Services/Research-
Institutes/alterra/Facilities-Products/Software/WOFOST.htm
Agrometeorological Cereal Yield Forecasting in Morocco
Introduction
Balaghi R., Jlibene M., Tychon B., Eerens H.
1
I. INTRODUCTION
The main authors’ interest, in studying the climate and its impact on cereal production with the aim
of yield forecasting was sparked by the major drought that occurred during the 1994-1995 growing
season in Meknes region. Scenarios of the prospective consequences of drought on cereal yields
were analyzed by identifying, from the historical climatic and crop database, similar seasons to
1994-1995 in terms of cumulative rainfall over the period of September-January. While similar
cropping seasons could be identified from early rainfall distribution, it turned out later as the season
evolved, that the season of 1994-1995 was exceptional, far away from recorded past seasons. The
severe drought during this unexpected season was about to cause an economic disaster in Morocco,
narrowly averted by the bumper harvests of the following season (1995-1996). Analysis of historical
data since 1960 indicated that the climate in Morocco was relatively stable until 1980 from where a
change had occurred, expressed in lower rainfall and higher temperatures during the next three
decades. Also, higher intra and inter annual variations tended to increase. Frequency and severity of
climatic hazards were amplified with negative consequences on water resources and agriculture,
and ultimately on the whole economy of the country. It then became necessary to invest in
agrometeorological research science in order to develop effective methods and rapid tools capable
of forecasting cereal yields early in the season. The present document reports the thinking process
and research work done at National Institute for Agronomic Research that led to the development
of a global approach of cereal yields forecasting in Morocco. The document provides also some
ideas worth exploring to improve precision and accuracy of the forecasts.
Crops evolve under the direct influence of agrometeorological factors such as temperature,
moisture, sunlight and radiation, or hygrometry. The development of methods for crop yield
forecasting, requires a thorough understanding of the interaction between these factors and the
crops. The study of these interactions is the result of a maturing process of an emerging science
termed “Agrometeorology” stemming from “Biogeography” and later from “Bioclimatology”. The
term agrometeorology
1
, meaning meteorology applied to agriculture, emerged during the 1920s
(WMO, 2006), and developed as a recognized and established science in the 1950s (Seemann et al.,
1979). The objective of operational agrometeorology is to predict the response of crops to external
conditions of natural (climate, soil, diseases, parasites, etc.) or human origins (cultural practices,
prices, etc.).
Research on operational agrometeorology at National Institute for Agronomic Research of Morocco
(INRA), precisely at its Regional Agronomic Research Centre of Meknes, was triggered by the
1
Agricultural meteorology can be defined in large terms as « a scientific discipline that is concerned with the study on
heat, air and biomass inside and above ground, in areas devoted to agricultural production, in addition to the incidence
of parasites and diseases on crops and animals which also depend on these factors for their expression” (WMO, 2006).
Agrometeorological Cereal Yield Forecasting in Morocco
Introduction
Balaghi R., Jlibene M., Tychon B., Eerens H.
2
extreme drought that happened during the cropping season of 1994-1995. This drought sparked
the awareness of the need to forecast crop yields. Cumulated rainfall over the cereal cropping cycle
(September till May) in Meknes region, as example, was less than half (43%) the amount of an
average season (530 mm). The season started with a 33% deficit in rainfall (September till
November) as compared to the long term average (126.5 mm) and stayed dry until February,
receiving practically no rainfall during three months (7.6 mm). Between the months of February and
May, rainfall was low and insufficient (134.8 mm) to help crop rebound.
Alarmed by the persistent drought during December 1994 and January 1995, the representative of
the Department of Agriculture in the province of Meknes (Direction Provinciale de l’Agriculture de
Meknes) requested the Regional Research Centre of INRA in Meknes to provide scenarios of
prospective developments of the season. Due to lack of established methodologies and forecasting
tools for cereal forecasting at that time, distribution of rainfall of the season of 1994-1995 was
compared to historical weather data available at the National Meteorological Service since 1960,
using frequency analysis method
2
. Based on this type of analysis, it appeared that season of 1974-
1975 was displaying a rainfall pattern similar to that of 1994-1995. During season of 1974-1975,
despite early drought the total amount of rainfall was close to average in the region. Hence,
frequency analysis of rainfall indicated that expected total season rainfall would be greater than
445 mm, with 90% probability, and greater than long term average with 50% probability. On the
basis of these two scenarios (50% and 90% probabilities), two figures of expected yields were
proposed. Considering the scenario of 90% probability, expected yields of soft wheat would be 11
Quintals per hectare (Q/ha) at the national level and 22 Q/ha at Meknes. Based on the scenario of
50%, expected yields would be 14 Q/ha at the national level and 27 Q/ha at Meknes. Finally at the
end of the season, none of the two scenarios has helped to provide yield forecasts for the season of
1994-1995. The reason is that this season was particular, with no similarity in both rainfall amount
and distribution with past recorded seasons since 1960. The drought period lasted long, covering
two thirds of the cereal cropping cycle. Final yields of soft wheat (official statistics of the MAPM),
were 4.79 and 5.70 Q/ha at the national and regional levels, respectively, which was an
unprecedented situation.
From the unexpected results, it was understood that the hypothesis of a stationary climate in
Morocco was no longer holding, that is historical rainfall frequencies have changed as a result of a
change in the climate since 1980. Other methods and tools for cereal yield forecasting had to be
invented based on an agro-meteorological approach resulting from a fine study of the climate and
its interaction with cereals development.
2
Frequency analysis is a statistical methodology that quantifies probabilities of a hazardous event like climate or
hydrology, based on past recorded events. It relies on the hypothesis of stationary or homogeneous distribution of the
time series.
Agrometeorological Cereal Yield Forecasting in Morocco
Introduction
Balaghi R., Jlibene M., Tychon B., Eerens H.
3
Preliminary analysis indicates that, in general characteristics, Moroccan climate is of
Mediterranean
3
type with major influences of the ocean, the desert and the mountains,
determined basically by its extended latitude (between 21°N and 36°N), sea shore, Sahara desert
and Atlas mountains. The latitude determines temperature due to the curving of the globe and the
inclination of solar rays, as compared to the tropics, resulting in a decrease of temperature from
south to north. The Mediterranean Sea in the North, with 512 Km of coast from Saïdia East to Cap
Spartel West, and the Atlantic Ocean to the west, with a longer coast of 2934 Km, from Cap Spartel
north to Lagouira South, attenuate temperature variation and temper the seasons. During summer,
temperatures are mild, similar to those on the Mountain, and moderate in the winter, creating an
environment of low temperature amplitude. The climate becomes continental from the Ocean
toward East, and temperature becomes cooler with altitude. The Sahara desert south of the
country, which has an arid climate (< 150 mm) influences inland climate through the movement of
tropical dry and hot air masses moving from south to north and from East to West. In the Atlas and
Rif mountains temperature is cooler due to elevation. Due to low temperature in mountains,
moisture precipitates as snow during winter and sometimes during spring.
The Azores anticyclone located in the Atlantic Ocean near Portugal, and the Saharan depression in
the south, exert antagonistic actions on the Moroccan climate. Humid and cold air masses reaching
Morocco from the Atlantic Ocean are accompanied with rainfall and snow in high elevation, When
the Azores anticyclone moves to the west or south west. During spring and summer, the Azores
moves up to high latitudes, pushing perturbations to the 45th parallel North. At the same time,
tropical dry and hot air moves up from south, leading to a net decrease in precipitation. Other
anticyclones of less importance influence Moroccan climate as well, particularly the Atlantic one
and the Mediterranean one.
Total annual precipitations in Morocco increase along latitude from south to north and along
longitude from east to west. Water supply in Morocco is entirely dependent on precipitations,
unlike countries of the Middle East, Eastern and central Africa. There are no supplied water sources
outside the boundaries of the country, unlike some other countries such as Egypt whose water
supply is dependent on the Nile River or Syria or Iraq who depend on the Euphrates River, or the
central African counties who are sharing the great lakes. Therefore, because of the total
dependence of Moroccan agriculture on precipitation, any sudden deficit has immediate negative
impact on agriculture and water resources, and consequently on the economy of the country as a
whole. This great dependence of Moroccan economy on rainfall was understood early by Marchal
Louis Hubert Lyautey whose famous statement “Governing Morocco is raining” is still at date.
3
The Mediterranean climate is characterized by a rainy season in autumn and winter seasons, hot and dry summers,
and cool winter temperatures.
Agrometeorological Cereal Yield Forecasting in Morocco
Introduction
Balaghi R., Jlibene M., Tychon B., Eerens H.
4
The gradient of moisture from south to north is so large that six relatively homogeneous agro-
ecological zones could be distinguished (Table 1):
Saharan desert zone, with less than 150 mm of annual rainfall ;
Pre- Saharan zone, with annual rainfall in the range of 151-250 mm ;
Arid zone, where annual rainfall in the range of 251 and 350 mm ;
Semi-arid zone, with annual rainfall in the range of 351 and 450 mm ;
Sub-humid zone, with annual rainfall in the range of 451-550 mm ;
Humid zone, with annual rainfall above 550 mm.
Table 1: Classes of rainfall and corresponding aridity levels with administrative provinces included
in each class. Rainfall data are averages of 1988-2005.
The Atlas Mountains lanyard the country diagonally, along the axis north-east to south-west. They
act as a natural barrier against desert influence from its Southern side and against moisture
originating from the Ocean from its Western side. The blocked oceanic moisture by the Atlas
Mountain chains precipitates as snow or rainfall depending on the harshness of winter
temperatures (Figure 10). Because of its natural geography, the Atlas Mountains are assimilated to
the national water reservoir. Elevation peaks at 4,165 meters above sea level at Jbel Toubkal
Mountain. Beyond this natural barrier, climate is therefore arid and pre-Saharan. Water masses,
coming from runoff of rain water and melting of the snow, in the mountainous chains, feed the
major rivers of the country, which flow to the ocean for most of them (Loukkos, Bouregreg, Sebou,
Oum Er-Rbia, Souss), to the Mediterranean Sea (Moulouya) or to the Sahara (Ziz and Draâ). The
Middle Atlas, a mountainous range stretching from south-east to north-west along 350 km, is the
North African Mountain most covered with moist areas, mainly natural lakes, rivers and fresh
springs.
Due to its geographic configuration, agriculture in Morocco, is confined within the borders of the
Mountains and the Seas, and is highly influenced by climatic factors, mainly rainfall. Availability of
Annual rainfall (mm)
Aridity level
Province
<150
Sahara
Dakhla, Lâayoune, Tantan, Errachidia, Ouarzazate, Bouarfa,
Tiznit, Sidi Ifni
151-250
Pre Sahara
Midelt, Taroudante, Marrakech, Oujda, Agadir
251-350
Arid
Settat, Nador, Al Hoceima, Essaouira, Beni Mellal,
Nouasser, Khouribga, Kasba Tadla
351-450
Semi-arid
El Jadida, Safi, Casablanca, Sidi Slimane
451-550
Sub-humid
Rabat-Sale, Kenitra, Tounate, Meknes, Fes, Taza
> 551
Humid
Larache, Tetouan, Tangier, Chefchaouen, Ifrane
Agrometeorological Cereal Yield Forecasting in Morocco
Introduction
Balaghi R., Jlibene M., Tychon B., Eerens H.
5
water, which is the most vital factor to agriculture, is shrinking due to the growing pressure of
demand for water for urban, industrial and touristic development, and to the negative impact of
climate change on rainfall. Major expected expressions of climate change in Morocco include
reduced precipitations, increased temperatures, and intensified extreme hazards like drought, heat
waves, frost, flooding, etc. (SEEE, 2010).
The economic weight of agriculture on the Moroccan economy (15 to 20% of GDP and 40%
employment) is so high that any temporal or seasonal variation of the climate will immediately
affect agricultural production, particularly that involving crops used as the basis of food security like
cereals. Rainfall variation between the successive cropping seasons of 1994-1995 and 1995-1996, in
a ratio of 1 to 3, affected cereal yields in a ratio of 1 to 3.61 and production in 1 to 5.74.
Comparison between the seasons of 1994-1995 and 1995-1996, indicates a cumulated rainfall over
the crop cycle (September till May) of 198 mm vs. 591 mm, and yield a of 4.79 Q/ha vs. 17.27 Q/ha,
and a production of 17 vs. 98 million quintals. The unpredicted harvest of 1994-1995 could have led
to an economic disaster if the next season didn’t happen to be rainy and productive.
Cereal imports were consistent since 1980, representing nearly half (48.7%) of the cereal
production and most of imported food products and import cost. Annual cereal imports amounts to
2.6 million tons on average for the period of 1980-1981 till 2010-2011 (ONICL, 2012), most of it
composed of soft wheat which accounts for 77%, followed by durum wheat (12%) and barley (11%).
Cereal imports are in constant progression since early 1990s, fluctuating over time and ranging
from 10% of average cereal production (during season of 1994-1995) following the good harvest of
1993-1994 to 244% during 2000-2001 following the dry season of 1999-2000. However, cereals are
imported even during record productive seasons like during 2008-2009 (10.2 million tons of
production), where significant quantity was imported during the next season (2.56 million tons),
that is 25% of the 2008-2009 total cereal production. Total cereal supply, as the sum of production
and import, without accounting for stocks, which may represents total needs for food and feed,
increases over years with a rate of 0.16 million ton per year since the 1990s. This high dependence
on imported cereals is associated with risks of short supplies and high prices in the international
market which may result from the variation in global production, embargos on imports and
speculation.
The 5.3 million hectares of cereal land have been, by commodity, subdivided by the MAPM into six
agro-ecological zones: Favorable, Intermediate, Unfavorable South, Unfavorable East, Mountainous
and Saharan. The favorable zone is located in the northern parts of Morocco, starting at its
southern limit at the Kenitra-Taza line. The Intermediate zone is located in the center of the country
north of the Casa-Benslimane line. The Unfavorable South zone is located between the Casablanca-
Benslimane line and Agadir province, while the Unfavorable East zone is located in the eastern
parts of the country. The Saharan zone is located South of Agadir province. The Mountainous zone
is mainly located in the Atlas Mountains. The Favorable, Intermediate and Unfavorable South
account for 75% of total cereal production and 75% of total area. The Mountainous zone
contributes to total cereal production and total cereal area by 12% and 10%, respectively. Yields are
Agrometeorological Cereal Yield Forecasting in Morocco
Introduction
Balaghi R., Jlibene M., Tychon B., Eerens H.
6
higher in the Mountain zone with an average of 1.41 ton per hectare for the period 1991-2011
(DSS), followed by the Favorable zone (1.39 tons per hectare), Intermediate zone (1.19 t/ha) and
Unfavorable South (0.76 t/ha).
Yields at the country level are low, half of those obtained in research experiments at the INRA
experiment stations (Jlibene, 2009), indicating that there is still important room for cereal yields
improvement in Morocco. National yields have slightly improved over the years. For example, soft
wheat yields evolved from 0.7 t/ha in the 1940s, to 0.9 t/ha in the 1950s and remained unchanged
during the 1960s and the 1970s, despite large scale state development programs: “plowing”,
“fertilizers”, and “seed” (Jlibene, 2009). During the next three decennials (1980s, 1990s, 2000s)
country average yields improved to 1.4 t/ha despite a decrease in rainfall and frequent drought
episodes. However, this improvement was yet insufficient to cover the fast growing population
needs. Imports rocketed from 0.9 million ton in the 1970s to more than 2.0 in the 2010s, with peaks
of 5.0 million tons in extremely dry year.
Episodes of drought can occur early in the cropping season, in the middle or at the end of the
season, or in combination of two out of three periods, in association with variation of temperature
and biotic stresses (insects and fungi). Rainfall variation, in amount and distribution within the
season, in addition to temperature variation and biotic stresses, creates a multitude of agro-
climatic situations which appear difficult to model.
Studies on the behavior of cereals in interaction with the climate in Morocco are scattered and lack
precise data and consistent monitoring of both the climate and plant parameters along the cycle.
Grain yield is the most often studied variable in relation to rainfall which is considered as by far the
most important climatic variable. The influence of climate on cereal growth and production is a
reality traditionally known to Moroccan farmers and students, becoming almost a faith not worth
investigating. Farmers usually wait for the first rain storms of September and October to prepare
the soil for planting before the rainy season in November and December, and entertain the crop
during spring before harvest in the summer. This is an ancestral practice that was perpetuated over
centuries. As a consequence, research studies have focused mainly on means to improve yields
instead of understanding their formation in relation to climate. Without understanding the
interaction between crops and the climate, yields forecasting would be unrealistic.
Crop harvest forecasting
4
provides the opportunity to become prepared for consequences of any
shortage in production through actions to reduce vulnerability to climatic risks. It is hence a
valuable tool for decision making in agriculture, allowing for planning in advance actions like aids to
farmers, or cereal imports. It also allows quantifying drought impacts needed by institutions like
agricultural insurance company. Harvest forecasting evolved from applied research status to an
operational one, due to INRA research efforts in collaboration with partner institutions national and
international.
4
Crop forecasting is a science that permits to foresee crop yields using mathematical models.
Agrometeorological Cereal Yield Forecasting in Morocco
Introduction
Balaghi R., Jlibene M., Tychon B., Eerens H.
7
From a practical point of view, harvest forecasting can be realized at different spatial scales, ranging
from farmer’s level to country level. Approaches to use for crop forecasting depend on the scale,
available basic data, and the required precision. From a scientific point of view, approaches can be
grouped into three categories depending on the level of conceptualization (Gommes et al., 2010):
Expert approach is based either on experience or opinions of the investigator or farmers
when economic factors are at stake. This is the case of Delphi approach for predicting
harvests (Moricochi et al., 1994) ;
Extrapolation approach is based on diverse statistical analyses (simple or multiple
regression, principal component analysis, neuronal net, etc.) of agricultural production and
environmental factors (climate, fertilizers, prices, etc.) or semi-empirical models and
simulation
5
models. This approach is often used in operational agrometeorology ;
Intermediary approach, which is a combination of expertise and extrapolation approaches.
Empirical approaches are often based on statistical relations between the climate and vegetation,
whether it is natural or grown. These relations can also be studied using directly measured climatic
parameters by the parametric approach, or by subjective appreciations of the cropping seasons by
an expert agricultural scientist. Relations that use empirical approaches are valid only for contexts
similar to the one that led to their development. Statistical approaches are quite important because
they are practical; they can rapidly recognize relevant climatic element that effect crops. However,
they are valid only for stationary phenomena in time, that is, when climatic phenomena stay
unchanged over the period under consideration. In Morocco, we have observed since early 1980s a
change in climate leading to more frequent droughts and increased aridity. Statistical approaches
can be used only for series of climatic data recorded after the change point.
Agrometeorological forecasting of crop harvests, at the country or the region levels, which is a
branch of operational agrometeorology, refers to two main schools of thought: the school of
modelling approach of interactions between crops and their environment (water balance,
physiology processes, energy absorption, etc.) and the school that can be qualified as “pragmatic”
which relies on methods using statistical models linking crop production to agronomic factors,
climatic, environmental, or economical indices
6
.
The pragmatic approach has been adopted by FAO for forecasting crop harvests worldwide due to
its effectiveness and relative simplicity to implement (Gommes, 2001). Agro-meteorological
forecasting approach of crop harvests adopted by the World Meteorology Organization is a
combination of simulation models and statistical models (WMO, 2010). Likewise, the approach
5
Agro-meteorological simulations models of growth and crop development allow understand the response of plants to
variations of the environment.
6
Indices are useful for modelling because they constitute an integrative or a combination of a range of environmental
parameters (rainfall, temperature, moisture, hygrometry of soil, etc.) which explain the behavior of plants. An index is a
practical way of simplifying the plant environment.
Agrometeorological Cereal Yield Forecasting in Morocco
Introduction
Balaghi R., Jlibene M., Tychon B., Eerens H.
8
used by FAO to estimate crop yields, relies on linear regressions between official statistics and
outputs of a model named AgroMetShell
7
which calculates water balance of cultivated soils.
Outputs of this model are regressed linearly to agricultural statistics, which accounts for technology
improvement, real field conditions of farmers, and resolves the problem of spatial resolution
8
, and
the bias of simulation models. The pragmatic approach has also been used to propose indicial
systems to manage agricultural insurance for cereals (Skees et al., 2001; Stoppa and Hess, 2003)
and sugar beet (Koch, 2011) in Morocco.
Operational forecasting of cereal yields has been attempted for the first time in Morocco in 1994
(Bazza and Tayaa, 1998), for the province of Settat, as a part of AGRIMA (Agriculture Maroc)
project, launched jointly by the MAPM, the Royal Centre of Spatial Remote Sensing (CRTS) with
assistance from the United Nation Development Program (UNDP). In this project, proposed
statistical models for forecasting cereal yields used actual evapotranspiration (ETa) as the
predicting variable for Settat province. To generalize this model to other regions, a simulation
model of evapotranspiration was tested. However, this experience was carried out for only one
year, and could not be conclusive.
At the research level, numerous studies have tested simulation models developed in other
countries in varietal selection to identify criteria of selection for wheat (Confalonieri et al., 2012), in
agronomy research to simulate grain yields of cereals at research experiment plots (El Mourid,
1991; Bennani et al., 1993), in risk analysis of climatic risk in relation with choice of barley cultivar
and planting date (Hanchane, 1998 and 2009), in improving wheat productivity through
management of genotype and irrigation (Debaeke and Aboudrare, 2004), in irrigation management
of wheat (Hadria et al., 2006) or sugar beet productions (Taky, 2008), etc.
Now a day, the only public institutions that realize operational agrometeorological cereal yield
forecasting at the country level are from one side the consortium composed of INRA, DSS and DMN,
and the other the central bank of Morocco (Bank Al Maghrib http://www.bkam.ma/). Both
institutions are using pragmatic approach with different methods and tools.
The objective of the present document is to summarize research works on agrometeorological
cereal yield forecasting in Morocco, including both the thinking process and the results elaborated
by the authors.
7
AgroMetShell is a tool for monitoring and forecasting crops, developed by FAO (ftp://ext-
ftp.fao.org/sd/reserved/agromet/AgroMetShell/)
8
Major problem concerning agrometeorological simulation models is the scaling up from the experiment station field
to the administrative province or region. Most variables used in modelling are obtained in small plots controlled
experiments, and are difficult to find or measure on larger scales.
Agrometeorological Cereal Yield Forecasting in Morocco
Approaches of analysis
Balaghi R., Jlibene M., Tychon B., Eerens H.
9
II. APPROCHES OF ANALYSIS
To develop a cereal yields forecasting approach, two things were needed: (1) assembling and
managing a data base on meteorological, biological, agricultural, geographical, satellite and
administrative variables, and (2) identifying agro-climatic indicators that correlate to yields. Satellite
data were provided by our European partners, while other types of data were available in Morocco,
but raw and often fragmented and discontinuous, requiring preliminary treatments. Search of yield
correlating agro-meteorological indicators was first limited to indicators derived from temperature
and rainfall data, independently from the cereal crops, particularly statistics of means and variances
of inter region, inter and intra season variations, or else probability of occurrence of a fixed rainfall
amount. Similar to climatic analyses, indicators of cereal crop production were also studied
independent from climate variation. Likewise, means and variances across regions and years were
computed. These analyses remained descriptive, suggesting search and validation of other
indicators which integrate both the climate and the crop. New indicators of cereal yields forecasting
highly correlated to yields, were explored, mainly water balance indices and vegetation indices
derived from satellite imagery. Normalized Difference Vegetation index (NDVI) is delivered as an
image in a “raster” format (pixel), with no distinction among the numerous land covers. Therefore, a
specific land cover map was developed for Morocco, by compiling various maps available at the
global level, and which were improved by adding extra information reflecting the importance of
agricultural land in each pixel. Recent development of computer science tools, particularly the
geographic information system, and the availability of satellite images of high resolution at low
price, have opened new direction toward analyzing interactions of crops with the climate, in general
and toward forecasting cereal yields in particular.
1. DATA BASE SETS USED
Agrometeorology requires compilation of data sets of different types: meteorology, soil, hydrology,
biology, agronomy, satellite, geography, etc. These data sets have to be satisfactory in quality and
quantity, so as to detect interrelations among agricultural environment and crop behavior. The
different data sets must inter cross, spatially and temporally, so as to be able to be cross analyzed.
Compilation of data, its storage, data updating including quality control, are all part of required
upstream work
9
.
Data sets used in this document include:
9
Gathering and managing agro-meteorological data bases (climatic, satellite imagery, geographical, soil, etc.) need an
inter-institutional effort to develop operational agro-meteorology.
Agrometeorological Cereal Yield Forecasting in Morocco
Approaches of analysis
Balaghi R., Jlibene M., Tychon B., Eerens H.
10
Meteorological data, mainly rainfall, solar radiation, air temperature, air moisture, wind
speed and direction ;
Satellite imagery data provided by Flemish Institute of Technology Research (VITO,
Belgium) ;
Historical data on North Atlantic Oscillation Index
10
(Hurrell, 1995) ;
Administrative boundaries in GIS vector format of the country, the regions, the agro-
ecological zones and provinces ;
Agricultural land cover map “GICropV2” developed in collaboration with VITO in 2012 ;
Agricultural statistical data, involving historical series of yields and areas of cereals by
province.
All these data sets were compiled in a new data set entertained at the Environment and Natural
Resources Department of INRA.
1.1. CLIMATIC DATA
Data sets on climate used in this document, are of different sources:
Historical set of 35 synoptic stations, available at the Ministry of Agriculture and Marine
Fishery (MAPM). These stations are part of the 44 synoptic stations of the “Direction de la
Meteorologie Nationale” (DMN). Data include rainfall data from 1987 till 2011, and
decadal temperature data (minimum and maximum) for the period of 1999-2009. The 35
stations are localized each in one of the provinces, the reason why their names are the
same as those of the provinces. The synoptic stations of DMN operate day long for most
of them (24h/24h) and produce hourly observations on main meteorological variables:
atmospheric pressure, temperature, relative humidity, strength and direction of the wind,
cloudiness, quantity and intensity of precipitations, insulation time, and radiation.
However, these stations cover mostly costal zones, and less mountainous, Eastern and
Saharan regions ;
Long term daily historical data on rainfall and temperature from some synoptic stations,
provided by DMN. Some of these data date back to the start of last century ;
Data of Global Historical Climatology Network (GHCN,
http://www.ncdc.noaa.gov/ghcnm/) (Peterson and Vose, 1997). This network provides
daily temperature, precipitation and air pressure. It is managed by the National Climatic
Data Center at the University of Arizona (USA). Data are collected continuously from a
10
North Atlantic Oscillation index is generally calculated as the difference in air pressure at sea level, between two
meteorological stations localized respectively at the Island depression and Azores anticyclone.
Agrometeorological Cereal Yield Forecasting in Morocco
Approaches of analysis
Balaghi R., Jlibene M., Tychon B., Eerens H.
11
great number of fixed stations on the surface of the globe (approximately 6,000 stations
for temperature, 7,500 stations for rainfall, and 2,000 stations for air pressure) ;
Daily climatic data on rainfall, temperature max and min, sped of wind, and relative
humidity, provided by the web site www.tutiempo.net which reuse data of the GHCN
network ;
FAO climatic data “FAOCLIM 2.0” (http://www.fao.org/nr/climpag/pub/en1102_en.asp;
FAO, 2001). This website contain daily climatic data worldwide, including Morocco, over
the period of 1960-1990 ;
Dekadal meteorological data over a 70 years period for the rural district of Arbaoua,
kindly supplied by the “Office Regional de Mise en Valeur Agricole de Loukkos” (ORMVA-L)
for the PSDA project (Jlibene and Chafai, 2002). These data were used to determine
periods of cultural interventions for the Loukkos region using the FAO method (FAO,
1978) ;
World climatic data (www.worldclim.org/; Hijmans et al., 2005), containing among others,
monthly precipitation and temperature (mean, max and min) for the 1950-2000 period,
interpolated at a spatial resolution of 1x1 km ;
Meteorological data provided by INRA automated stations, particularly the one of Meknes
regional research center, for the period of 1995-2000. Hourly data were recorded in this
station on rainfall, temperature, relative humidity, wind speed and direction, dew point
and solar radiation.
1.2. NORTH ATLANT IC OSCILLATION INDEX
History of the North Atlantic Oscillation Index (NAO) is available for download at the website of the
University of East Anglia, England (http://www.cru.uea.ac.uk/ ~timo/datapages/naoi.htm). Data on
this index are available for all months of the year without interruption since 1821, which allows for
reconstitution of past climatic events. It was used to demonstrate the influence of NAO index on
precipitation pattern in Morocco. NAO index links the intensity of Island depression to that of the
Azores anticyclone. NAO fluctuations have direct consequences on Moroccan climate (Figure 1).
Agrometeorological Cereal Yield Forecasting in Morocco
Approaches of analysis
Balaghi R., Jlibene M., Tychon B., Eerens H.
12
Figure 1: Positive phase (left) and negative one (right) of the North Atlantic Oscillation. At the
positive phase, drought reigns over the Mediterranean region, while storms are frequent over
Europe. On the contrary, at negative phase, the Mediterranean region enjoys a humid weather
while Europe is less humid. (Source: http://www.ldeo.columbia.edu/res/pi/NAO/).
1.3. VEGETATION I NDEX FROM SATELLITE IMAGERY
With the development of satellite imagery, it was possible to develop agro-meteorological indices
from the spectral reflectance of the vegetation. These indices can be used to forecast crop harvests,
either directly as predicting factors in regression equations (Kogan, 2000; Maselli et al., 2000;
Balaghi et al., 2008), or indirectly to estimate biophysical variables like LAI
11
or fAPAR
12
used as
input variables to simulate growth of crops (Duchemin et al., 2006; De Wit et al., 2012). However,
one of the major obstacles of using such indices in simulation models lies in the discrepancy
between the spatial resolution of the geographic information and that of the physiological
processes of photosynthesis. Physiological parameters are obtained in small experiment plots,
while satellite imagery derived indices are obtained for large areas and in high frequency (dekadal)
to allow agro-meteorological monitoring (Balaghi et al., 2010).
Normalized Difference Vegetation Index or NDVI, as derived from NOAA-AVHRR sensor since 1980,
SPOT-VEGETATION since 1988 or MODIS
13
since 2001, is one of the indices most used to measure
the vitality of vegetation. NDVI has been largely used to monitor vegetation and forecast crop yields
all over the world. It is computed as follows:
11
Leaf Area Index (LAI) is the projected area of leaves in a unit area of soil surface (Watson, 1947).
12
fAPAR is the fraction of solar radiation absorbed by plants, in the spectral range of photosynthesis.
13
Moderate Resolution Imaging Spectroradiometry (MODIS) sensor is installed on board of satellite Terra and Aqua of
the National Aeronautics and Space Administration (NASA), used for the follow up of vegetation (NDVI and EVI) at 250
meters spatial resolution.
Agrometeorological Cereal Yield Forecasting in Morocco
Approaches of analysis
Balaghi R., Jlibene M., Tychon B., Eerens H.
13
REDNIR REDNIR
NDVI
Where, NIR and RED are respectively measures of reflected radiation in the near infra-red, the red.
NDVI increases progressively with increased vegetal density; from a value of 0.15 (average value)
for bare soils, to a value of 0.75 (average value) for dense plant covers. One of the main benefits of
using NDVI is the integration of environmental factors, in a sense that it reflects the state of global
environmental stress of the vegetation, more than separate climatic variables or simulation models
can do (Balaghi et al., 2010). For example, water stress resulting from a prolonged deficient water
balance is reflected by low NDVIs. Decreases of NDVI values can also be caused by abiotic stresses
like mineral deficiency or toxicity or biotic stresses like epidemics of diseases and insects.
NDVI values are delivered on a 10-days basis for NOAA-AVHRR images since 1982 and SPOT-
VEGETATION since 1999, and a 15-days basis for MODIS images since 2001. These images are
pretreated by VITO before delivery, correcting for radiometric, geometric and atmospheric
variations.
1.4. ADMINISTRAT IVE BOUNDARIES OF MOROCCO IN GIS FORMAT
Official agricultural statistics are delivered on administrative province basis; which makes the
province, the smallest territorial area for forecasting. Polygons delimiting provinces are available in
a GIS format. The rural districts boundaries are also available in GIS format and can be used to
forecast yields at this level when statistical yield data will made available.
1.5. LAND COVER MAPS
Different land cover maps issued from spatial remote sensing, are available for free use at the
global level, with varied quality and precision. Global Land Cover-2000 (GLC2000 version 5.0,
Mayaux et al., 2004), CORINE-2000 and GlobCover (Tchuente et al., 2010; Neumann et al., 2011),
were of particular interest to us. The European program CORINE Land Cover (CORINE-2000) is an
inventory of 29 European states land cover from satellite images. It covers part of Morocco as well.
The CORINE-2000 land cover map has a geometric precision greater than 100 m which allows for
elaborating maps of less than 30 meters spatial resolution, has been updated in 2006 by the Global
Monitoring for Environment and Security initiative (GMES) (http://sia.eionet.europa.eu/CLC2000).
GlobCover initiative of the European Spatial Agency aims at producing a global map of land cover,
using data of 300 meters of spatial resolution of MERIS sensor embarked on board of ENVISAT
satellite platform (http://postel.mediasfrance.org/fr/PROJETS/Pre-operationnels-
MES/GLOBCOVER/). Digital GlobCover land cover map elaborated in 2008 is the unique reference of
Agrometeorological Cereal Yield Forecasting in Morocco
Approaches of analysis
Balaghi R., Jlibene M., Tychon B., Eerens H.
14
intermediate resolution which covers Morocco. A land cover map for Africa is also made available
for free use by the « Southern African Development Community (SADC) » (http://www.sadc.int/).
To develop a specific land cover map for Morocco, GlobCover V2.2, CORINE-2000, AfriCover
14
, and
SADC maps were all grouped in one map covering Moroccan agricultural territory and improved by
the superposition of another map developed by USGS which provides data on the proportion of
agricultural land for each pixel, reducing there after intra pixel variation. Agricultural zones of this
map were extracted to serve as mask for NDVI images. Therefore, only values of NDVI of
agricultural zones are saved for possible use in establishing relationship between NDVI and cereal
yields.
1.6. AGRICULTURAL STATISTICS
Area and yield data of the three main cereal crops
15
, soft wheat, durum wheat and barley, were
graciously provided by “la Direction de la Stratégie et des Statistiques
16
” (DSS). They are available
for 40 provinces of the country for the period of 1978-2011. Production at a province level is
obtained by multiplying the yield value with the area estimated by DSS.
Area estimation for cereal crops in Morocco is made every year by DSS between February 10th and
March 30th, using a sampling method of 3,000 unit areas representing 19 million hectares. Starting
since 2008, DSS has renewed the sampling procedure to integrate modern techniques of satellite
remote sensing and GIS which improved precision of estimators. A GIS application has been
specifically developed for this purpose with capability of automated steps of the sampling
procedure.
Within the sampled areas, sub-sampled plots are harvested and their yields directly measured.
Production of a sampled area is the product of measured yield on sub-sampled plots and area
represented by the sample. Data on production and area is then aggregated by province.
Monitoring of vegetation, area and yield estimations are carried out by DSS along the cereal crop
cycle in three phases:
Phase 1: A survey on the evolution of harvest, is done in February, to evaluate crop
growth stages and vegetative stand of crops ;
14
The objective of Africover project is to establish a numerized data base geo-referenced for global vegetation cover
and geographic referential for all Africa, including: Geodesic referential, toponymic, roads, hydrographic. The polyvalent
data base Africover for environmental resources is produced at a scale of 1:200.000 (1:100.000 for small countries and
specific zones). www.africover.org/.
15
These cereals are sown in the fall (autumn) and are sometimes called autumnal cereals as compared to spring cereals
like corn or sorghum.
16
Previously called « Direction de la Programmation et des Affaires Économiques » (DPAE).
Agrometeorological Cereal Yield Forecasting in Morocco
Approaches of analysis
Balaghi R., Jlibene M., Tychon B., Eerens H.
15
Phase 2: A survey on land cover, done between February and June, to estimate cereal
areas ;
Phase 3: A survey on expected production, done in April (1 to 2 months before harvest) to
estimate production of the three main cereals: soft wheat, durum wheat and barley.
2. METHODS OF ANALYSE S
2.1. EXPLORATORY TREATMENT OF RAINFALL DATA
In operational agrometeorology, the first modelling steps consist in computing basic statistics of
agro-climatic variables (rainfall, temperature, yield, etc.). Simple statistics like averages, minima and
maxima, or deviations, are calculated from daily or 10-days raw data of a series of crop cycles or
meteorological stations. For temperature, daily average is obtained as the mean of minimum and
maximum values. Monthly average temperatures are obtained as the mean of daily averages over a
period of the month. The mean of 12 monthly averages represents year average temperatures. The
mean temperature of a series of year’s averages represents long term average. Difference between
maximum and minimum of each average temperature represents thermal amplitude.
For rainfall, daily total is the sum of hourly rainfall records over a day. Likewise, 10-days, monthly or
yearly averages are cumulated rainfall over each respective period. The sum of daily or 10-days
rainfall data is used for better graphic visualization and interpretation, since punctual daily or 10-
days
17
data graph representations for a number of years or stations are overlapping and hence
difficult to interpret. An example of 10-days rainfall graphics of a series of years is presented in
Figure 2.
17
Only 10-days data on rainfall at the province level and satellite imagery were available and therefore used in all
computations. 10-days rainfall data has generally proved to be sufficient for crop yield forecasting.
Agrometeorological Cereal Yield Forecasting in Morocco
Approaches of analysis
Balaghi R., Jlibene M., Tychon B., Eerens H.
16
Figure 2: Country average dekadal (10-days) rainfall, of a series of crop seasons from 1988 till
2011.
The first rainfall data treatment consists simply in analyzing a graphic representation of average
climatic data in the form of climograms, which indicate, for each dekad or month of the cropping
cycle, the heights of average precipitations (or medians).
The next rainfall data treatment consists in cumulating rainfall over the cropping cycle for 10-days
data points; reproducing clear graphs easy to interpret and even model statistically. Cumulus of
rainfall has many advantages, it allows:
Representation in a same curve, of the annual amount of rainfall as well as its distribution
along the cropping season ;
Modelling the obtained curve of rainfall by using linear, logarithmic or sigmoid equations;
Identify and determine periods of droughts; they will appear as flat horizontal segment in
the cumulative rainfall curve ;
Comparison of different crop cycles and regions, with regard to amount and distribution
of rainfall.
Agrometeorological Cereal Yield Forecasting in Morocco
Approaches of analysis
Balaghi R., Jlibene M., Tychon B., Eerens H.
17
2.2. FREQUENCY ANALYSIS OF RAINFALL
Frequency analysis is justified for variables displaying Gaussian distribution (Nicholson, 1986),
which is the case for 10-days or monthly rainfall in semi-arid regions. Frequency analysis method
(or probability analysis) of rainfall has been used by Gibbs and Maher (1967) to study drought in
Australia. In this analysis, values in series of rainfall data are arranged in ascending order, then
limits of deciles of the distribution are calculated from a frequency curve or a table. The 1st decile
represents the rainfall value that is not exceeded by one tenth of the values of the data series. Two
tenths of the rainfall values are less than the 2nd decile; three tenths are less than the 3rd decile and
so forth. The 5th decile or median is the rainfall value which could be received once in two seasons,
dividing the data series into two groups of equal number.
Frequency distribution is one way of showing the general form of intra-annual rainfall distribution.
A given season can be compared with median rainfall, which could be considered as reference
value. For example, rainfall values smaller than the 1st decile indicate situations of severe drought,
and those greater than the 9th decile indicate situations of high moisture. A major disadvantage of
this method is that the deciles are calculated for a given dekad or a month of the season
independently from other dekads or months.
2.3. OMBROTHERMIC INDEX OF BAGNOULS AND GAUSSEN
The Ombrothermic
18
index developed by Bagnouls and Gaussen (1953) is a climatic index which is
used to identify dry and humid months during the season. It is simply calculated as the ratio of
monthly rainfall (mm) over average monthly temperature (°C). This index was developed by these
authors, for the purpose of comparing different stations over the world where periods of drought
occur. Whenever the ratio is less or equal to 2, the month is considered as dry for plants,
considering that evaporative
19
demand of the air is greater than rainfall supply. On the reverse,
periods of the season where the index is greater than 2, are considered as wet. Dry and wet months
can be identified graphically from an Ombrothermic diagram where variations of temperatures and
precipitations are plotted in standardized grading: precipitation grade is “mm” and temperature
grade is “°C”. Using this diagram, climates of different stations all over the world can be compared
(Gaussen, 1956; Bagnouls and Gaussen, 1957). This diagram is however best adapted to the
Mediterranean climate where variations of rainfall and temperature result in periods of drought
and moisture. The Ombrothermic diagram was used in this document for the period of 1999-2009,
since the 10-days data of both rainfall and temperature were available for only this same time
period.
18
Ombrothermic diagram includes both rainfall (Ombros in Greek) and temperature.
19
Capacity of the atmospheric air to extract water vapor from the soil-plant system.
Agrometeorological Cereal Yield Forecasting in Morocco
Approaches of analysis
Balaghi R., Jlibene M., Tychon B., Eerens H.
18
2.4. LENGTH OF THE G ROWING PERIOD
The concept of Length of Growing Period
20
(LGP) was developed by FAO (1978) for the purpose of
identifying periods of the year where moisture and temperature conditions are adequate for crop
growth and development. It was later used by FAO for classifying agro-ecological zones to evaluate
potential and resources of global agriculture (FAO, 1996; IIASA/FAO, 2012). It is a renewed version
of Bagnouls and Gaussen (1953) ombrothermic diagram, taking into account water holding capacity
of the soil and evaporative capacity of the atmospheric air.
In Morocco, the same concept was used (Jlibene and Chafai, 2002; Jlibene and Balaghi, 2009) to
optimize cultural practices and placement of the crop cycle within the season, in order to minimize
risks of shortage of water (drought) or excess and improve agricultural productivity in.
LGP is determined by producing classes of rainfall on the basis of potential evapotranspiration (ET0,
i.e. atmospheric demand), under the following assumptions:
An amount of rainfall above one tenth and half of ET0 is sufficient for land
preparation for sowings ;
Conditions of rainfall between half and 100% of ET0 would be favorable to plant growth
and development ;
Conditions of rainfall greater than ET0 correspond to moist periods ;
The soil can hold a moisture equivalent to 100 mm of rainfall, which may extend the
growing period longer depending on ET0.
Based on these assumptions, different periods of intervention can be determined, mainly: land
preparation for sowings, date of sowings, growth period, moist period, start and end of each
period.
Using this concept in Morocco, it was possible to make recommendations in terms of optimal crop
cycle for cereals in the north-west region of Morocco (Jlibene and Chafai, 2002). A long series of
data, from the meteorological station of Arbaoua, pertaining to the “Office Régional de Mise en
Valeur Agricole du Loukkos”, and consisted of 10-days rainfall and temperature.
20
The concept of Length of the Growing Period (LGP), as defined by FAO (1978), is the period (in days) of the year
where amount of precipitations is above half of potential evapotranspiration. A time period necessary to evaporate 100
mm of soil water is sometimes added, assuming that the soil water reserve is 100 mm.
Agrometeorological Cereal Yield Forecasting in Morocco
Approaches of analysis
Balaghi R., Jlibene M., Tychon B., Eerens H.
19
2.5. ANALYSIS OF BREAKING POINT IN CHRONOLOGICAL SERIE S OF RAINFALL
Long term series of monthly rainfall data were available for 15 meteorological stations at the Global
Historical Climatology Network of the National Climatic Data Center of the United States of America
(www.ncdc.noaa.gov). Global climatic data, including data from Morocco, are compiled and
entertained by this institution. The long term climatic series for Morocco cover as much as 53 years
period, variable depending on stations.
A homogeneity test was performed to identify break points in the long term series of rainfall
cumulated between September and May (Balaghi, 2006). Three different tests were used: Mann
and Pettitt test (Pettitt, 1979), Buishand test (Buishand, 1982) and Lee and Heghinian test (Lee and
Heghinian, 1977).
2.6. PRODUCTIVITY OF RAIN WATER
Productivity of Rain Water (PRW) expressed by crops is defined as the quantity of grain produced
per unit of rain water of a cultivated land area (Grams*Liter-1*m-2). It is equal to grain yield weight
per hectare or meter squared (grams/m2) divided by the quantity of rainfall (in liters or millimeters)
received during crop cycle (October till April). PRW refers to productivity of green water
21
in
agriculture which concerns non-irrigated agricultural lands representing 87% of total land in
Morocco (MAPM, 2011). PRW differs depending on cultivation method, species, regions and
seasons. It is higher for crops grown under optimal conditions of fertilization, phytosanitary
protection, soil tillage, preceding crop, full or complementary
22
irrigation, etc. PRW is also higher
for species and cultivars that have been genetically improved.
In arid and semi-arid environments, in particular, wheat genetic improvement consisted in
developing cultivar resistant or tolerant to drought, efficient in using the limited water resources
available. When the actual evapotransipated part of rainfall by the crop during the cycle is
considered, RWP is called Rain Water Use Efficiency (RWUE). Potential RWUE of improved wheat
cultivars for agro-ecological zone « Bour Defavorable
23
» (unfavorable rainfed lands), has been
estimated at 2.2 Grams/liter, which represents a saving of water equivalent to 0.77 mm per year
and per hectare in comparison with previous cultivar Nasma (Jlibene, 2011). After three decades of
genetic improvement, water savings amount to 25 mm per hectare through the use of improved
new cultivars, as compared to Nasma.
21
The concept of « green water », introduced by Falkenmark (1995), is the part of rain water that is evapo-transpired
from the soil by non-irrigated crops. It is either consumed by plants (transpired) or evaporated by the planted soils. The
concept of « green water » is used as opposed to « blue water », which is water of surface runoffs and underground
tables used for agriculture and non-agriculture needs.
22
Irrigation that is generally supplied as complementary to rainfall during dry periods of the crop cycle, to meet the
crop needs.
23
The word « bour » means rainfed agriculture, which concerns about 7.9 million hectares in Morocco.
Agrometeorological Cereal Yield Forecasting in Morocco
Approaches of analysis
Balaghi R., Jlibene M., Tychon B., Eerens H.
20
RWP differs according to regions due to soil type’s variation, topography and mainly changes in
temperature. Hence, deep flat soils, with cool temperatures, use water more efficiently. RWP
differs also depending on years due to internal variation of rainfall and temperature, being more
important in uniformly distributed rainfall within the cropping season. It is also greater in
environments where temperatures across the season are relatively cool, because of reduced
evaporation losses. Soil can be classified with respect to RWP into three main categories of water
holding capacity (Roux, 1938):
1. Sandy soils, rich in sand and gravel fine or gross, poor in loam and clay, low in hygroscopic
capacity, have a limited stock of water ;
2. Clay soils, with high proportion of colloidal and fine elements, high in hygroscopic
capacity, have large stock of water, but not all readily available to plants ;
3. Soil of intermediate texture, with a balanced physical composition and intermediate
hygroscopic capacity, have high stock of water most of it made available for plant
consumption.
These three categories of soil can be further divided with regard to soil depth, permeability of
underground subsoil, and the existence of near and durable underground water table.
RWP was computed for autumn cereals at the country level, and agro-ecological zones, based on
provincial agricultural statistics from DSS and rainfall statistics from DMN, for the period of 1987-
2011.
2.7. PROCESSING OF SATELLITE IMAGES
Cereal yields forecasting on the basis of NDVI has been improved by two main preliminary actions
on raw NDVI data:
1. First, radiometric, atmospheric, and geometric corrections have been automatically
performed by the center producing NDVI images ;
2. Second, non-agricultural zones were removed from the geo-referenced map of NDVI. The
layer delimiting agricultural lands is called “agricultural mask”. This is a work of agro-
climatologists. The objective is to produce a new NDVI map which is free of NDVI from
non-agricultural origin.
The « agricultural mask» has been extracted from numerical maps of land cover, by eliminating
areas of the maps that do not correspond to agricultural land, like forests, pastures, cities, lakes,
rocky areas, etc. Agricultural masks available for Morocco at the moment have low spatial
resolution of 1x1 Km.
For this size, corresponding to 100 hectares, pixels identified as originating from agricultural land,
Agrometeorological Cereal Yield Forecasting in Morocco
Approaches of analysis
Balaghi R., Jlibene M., Tychon B., Eerens H.
21
may not be entirely covered by agriculture. Therefore, another precision is added, consisting of a
new layer, i.e. the proportion of agricultural land in each pixel, a variable called "Area Fraction
Image" (AFI). This map is available in higher resolution than existing agricultural masks.
By using pixels that are predominantly covered by agriculture, the relationships between NDVI and
yields of the three cereal crops (soft wheat, durum wheat and barley), have been improved; despite
the fact that a pixel’s NDVI is a reflectance from more than one crop. Three main reasons for these
improved relations: (1) cereals occupy most agricultural lands (5 of the 8.7 million hectares); (2) the
three species react almost simultaneously to variation of rainfall; and (3) spontaneous vegetation
(or weeds) responds to rainfall in a similar way. However, the agricultural mask can further be
improved, if the area specifically covered by cereals could be identified. This work is under
investigation by INRA, DSS, and DMN, as part of the project « Crop Monitoring as an E-agriculture
tool in Developing Countries » (E-AGRI, http://www.e-agri.info/), financed by the European Union in
its 7th Framework Programme for Research and Technological Development.
Images of NDVI corresponding to corrected agricultural mask are then subdivided into polygons to
fit the administrative boundaries of the country, the provinces or the agro-ecological zones, using
respective specific GIS layers. Spatial resolution used depend on the importance of the area for
which yield is predicted. Average NDVIs of all pixels within administrative boundaries are then
calculated to generate yield predictors for these geographic and administrative unit areas.
2.8. CEREAL YIELD FORECASTING APPROACH
The cereal yield forecasting approach for Morocco has been developed by INRA, in collaboration
with the University of Liège, VITO and JRC, using a combined approach which relied on several
methods of yield forecasting, mainly:
Analogy analysis, using rainfall or NDVI ;
Linear
24
regression, simple or multiple, using rainfall and NDVI data ;
Models of simulation of growth and development of crops. The simulation model
WOFOST
25
, which quantifies growth and production of annual crops, was used in the
forecasting of cereal yields.
Unlike agrometeorological modelling which is considered as basic research and oriented toward
scientific publications, operational forecasting of harvests has economic, social and political
implications on the society, leaving little room for error. In high level decision making, an error of
24
A very interesting lecture on linear regression models is given by Kutner et al. (2005).
25
WOFOST (WOrld FOod STudies) is a simulation model for the quantitative analysis of the growth and production of
annual field crops. It is a simulation model from “Wageningen family” (van Diepen et al. 1989; Supit et al. 1994; Hijmans
et al. 1994; Van Kraalingen et al. 1991).
Agrometeorological Cereal Yield Forecasting in Morocco
Approaches of analysis
Balaghi R., Jlibene M., Tychon B., Eerens H.
22
forecasting can ruin the reputation and even the carrier of whoever is making this error. Estimating
the error of forecasting is hence necessary for quality systems of crop harvests forecasting. For
example, JRC considers an error of forecasting of 3% as low giving high precision, and as acceptable
as long as it is less than 6% (Genovese and Bettio, 2004). The European crop forecasting system
“Crop Growth Monitoring System” (CGMS) predicts yields of main crops in Europe with an error of
3% to 5%; higher for durum wheat (8.6%). In absolute terms, this error is equivalent to 2 to 4 Q/ha
for wheat yields at the level of European countries (Genovese, 2001). The CGMS system is an
output of the European project « Monitoring Agricultural ResourceS » (MARS) carried out by the
Joint Research Centre of the European Commission (JRC) of General Agriculture and Rural
Development of the European Commission. The CGMS system provides, in an operational way, crop
harvest forecasts over Europe, in due time, at the regional and country levels. It monitors crops
development along the season, based on meteorological conditions, soil characteristics and crops
parameters. The heart of CGMS system is based on the deterministic crop growth model called
WOFOST. The CGMS system has three levels (Figure 3):
1. Level 1: Monitoring meteorological data and interpolating them over a square grid ;
2. Level 2: Simulation of growth for major crops ;
3. Level 3: Yield forecasting of crop harvests.
An organizational scheme of the CGMS model is presented in Figure 3.
Figure 3 : Crop Growth Monitoring System (CGMS) for crop harvests forecasting.
http://mars.jrc.ec.europa.eu/mars/About-us/AGRI4CAST/Crop-yield-forecast/The-Crop-Growth-
Monitoring-System-CGMS
Agrometeorological Cereal Yield Forecasting in Morocco
Approaches of analysis
Balaghi R., Jlibene M., Tychon B., Eerens H.
23
The combined approach offers a wide range of options for crop yield forecasting, as different
independent methods can be used simultaneously. It is used when none of the empirical or
simulation approaches available can predict crop yields with a satisfactory accuracy. This is the case
in operational forecasting where factors that affect crops development are complex and difficult to
measure and model. Combined approach is based on the use of expertise in agronomy which offers
flexibility to adapt to variations in database and limiting environmental factors (rainfall,
temperature, soils, pests and diseases, etc.). The combined approach is an alternative to simulation
approaches which cannot estimate forecasting
26
error. Also, simulation approaches are not entirely
adapted to procedures of calibration and validation because they generally produce no replicated
estimates and hence no variability of the estimates to use in usual statistical tests of comparison
with observed samples (Sinclair and Seligman, 2000; Van Oijen, 2002).
2.8.1. TECHNOLOGICAL TREND
One of the major problems in operational agrometeorology, in general, and crop yield forecasting
in particular, is the confounding effect of environmental factors with technology (fertilization,
irrigation, variety, phytosanitary treatment, etc.) and economic ones, which have a tendency to
influence yields over a long period of time (Gommes et al., 2010). A hypothetical example is
illustrated in Figure 4:
Technological trend (curve F1) has a direct influence on yields over long term (curve F3) ;
Economic policies (such as price) also influence the trend in yields (curve F2), since they
can encourage or discourage the use of inputs ;
Extreme weather factors, such as floods, hail, etc., can instantly influence crop yields, as
well (curve F4).
In general, technological trend is difficult to quantify as agricultural statistics in Morocco are
aggregated by province, region or dry vs. irrigated areas, therefore encompassing a multitude of
various inextricable agronomic situations. Based on agricultural statistics, it is generally difficult to
distinguish between the effect of the technological trend of price effects, inputs and extreme
factors. By product, the technological trend is considered to crop yield trend from a long period of
time.
26
A major disadvantage of deterministic simulation models is that they produce predictions with no error associated
with the estimates. A solution can be found using a holistic approach, where running a simulation model over climatic
variables and parameters associated with a probabilistic distribution as to obtain all probable estimates and their
statistical distribution (Allard et al., 2009). It is an indirect way of measuring prediction error in deterministic models.
Agrometeorological Cereal Yield Forecasting in Morocco
Approaches of analysis
Balaghi R., Jlibene M., Tychon B., Eerens H.
24
Figure 4: Hypothetical example showing how crop yields (red curve) can be influenced by
weather, technology innovation, policy, extreme factors and general trend (Source: Gommes et
al., 2010).
2.8.2. SIMILARITY ANALYSIS
Similarity analysis
27
is an effective, fast, and easy means for forecasting cereal yields. It consists in
identifying, among past cropping seasons, those that are agro-climatically similar to the one
concerned with crop yield forecasting. It is a statistical analysis that assumes that a cropping season
of similar agro-climatic conditions to past ones would result in similar crop yields, all other factors
assumed to be equal. This approach can be used only if a cropping season with similar climatic
conditions can be found in the past seasons. Consequently, a long chronological series of weather
data is necessary to encompass large agro-climatic variation, and to include the particular situation
under study. Obviously, specific copping seasons, like extremely moist or extremely cold ones, have
little chance of being encountered within historical seasons, and will be forecasted with higher
error. While this approach refers to real past agro-climatic situations, crop yields forecasted on the
basis of similarity analysis have to be adjusted for technological trend (cultivar, sowing, fertilization,
machinery, irrigation, crop protection, harvest etc.) achieved so far during the time difference
among the similar cropping seasons. For this type of analysis, it is also possible to associate a
forecasting error, considering that similar seasons, if any, are a sample of all similar seasons.
27
The concept of similarity analysis is used by meteorologists to quantitative forecast precipitation. It assumes that the
evolution of atmosphere is not random, but is governed by physical laws. Therefore, future evolution of precipitation
for example can be predicted, based on the precipitation pattern early in the season, using deterministic, probabilistic
or similarity models. Refer to Bontron (2004) for further readings on this subject.
Agrometeorological Cereal Yield Forecasting in Morocco
Approaches of analysis
Balaghi R., Jlibene M., Tychon B., Eerens H.
25
Similarity analysis can accommodate different kinds of analyses: visual graphics, simple statistical
techniques, Euclidian distance, principal components, or cluster analysis. One or more agro-climatic
factors, such as rainfall, or temperature, or indicators of vegetation such as NDVI or DMP
28
, can be
easily used for analysis, using simple computing facilities like spread sheet or usual statistical
procedures. Similarity analysis has been used at INRA, to forecast cereal yields as early as the
month of February, using rainfall or NDVI. As a matter of facts, while climatic conditions of the
months of March and April are determinant for final cereal yields, premises of a bad of good cereal
harvests are perceivable as early as end of February. The method, simple as it is, consists of
computing for each dekad (10 days), the absolute difference between respective values of the agro-
climatic variable used (rainfall or NDVI) of the cropping cycle of interest and those of historical
cycles, integrated over the cropping season. When the rainfall variable is used, the cropping cycle
extends from the first dekad of September to the 3rd dekad of April. For NDVI, means of dekadal
NDVI, over the period of first dekad of February to last dekad of April, are computed. A table of
absolute differences is then obtained, with dekads in columns and previous cropping seasons in
rows. From this table of absolute differences, averages across seasons by dekad (column average),
between February and April, are computed. Cropping cycles similar to the season of interest, are
those that deviate little from average values across seasons of absolute differences. Several
cropping seasons can be identified as similar to the season of interest, with however differences
among dekads within the cropping cycle. Yields of similar seasons are averaged to estimate yield of
the cropping season of interest. Yield of each cropping season can also be weighed according to the
relative distance from that of the season of interest. Due to inter-dekads variation, a confidence
interval of the weighed mean can be calculated. Similarity analysis relies on the principle of likely
neighborhood, with no distinct cutting point. Choice of how far a cropping seasons is similar is
hence subject to the forecaster’s appreciation, depending on his/her experience and knowledge of
past agro-climatic situations.
2.8.3. LINEAR REGRESSION AN ALYSIS
Linear regression analysis using the ordinary least squares method is a way to identify relationships
between crop yields and agronomic, environmental or economic variables used as predictors.
Agronomic predictors include inputs (water supplied, fertilizers, herbicides, pesticides, sowing and
harvest times) and methods of their application. Major climatic factors include rainfall, and
temperature, while NDVI and water balance are among environmental indicators, and price of
harvested crop, production cost are among major economic indicators. Indicators other than those
representing direct agronomic and climatic inputs can advantageously be used to improve yield
28
Dry Matter Productivity (DMP, http://www.geoland2.eu/) is an indicator of plant biomass growth rate derived from
remote sensing images of SPOT-VEGETATION and meteorological data, on a 10-dayss basis and a 1km spatial resolution.
It is expressed in kilogram per hectare and per day. Preliminary results of using this index in Morocco in E-AGRI project,
indicate a promising potential use in cereal yields forecasting.
Agrometeorological Cereal Yield Forecasting in Morocco
Approaches of analysis
Balaghi R., Jlibene M., Tychon B., Eerens H.
26
forecasting. For example, NDVI, which is an indirect measure of biomass dynamics, integrates
conditions of rainfall, temperature, soil water balance, agronomic practices, and any other factor or
condition affecting biomass production. However, contribution of any of these factors to NDVI is
hard to quantify. Predictors can be a quantitative variable like temperature, rainfall or NDVI, as well
as qualitative like presence or absence of drought, phenomena of alternance of some fruit trees,
and can be used separately or in combination. In Morocco, relevant predictors worth considering in
cereal yield forecasting include: rainfall, temperature and vegetation indices (NDVI and NDVI
derived). They can be computed for the entire cropping cycle or for parts of the cycle, and used in
simple linear regression models with yield or in combination in multiple regressions (Balaghi et al.,
2008). In a multiple regression model, variables must be carefully chosen to avoid collinearity
29
between variables, using statistical procedures of variables selection like « Stepwise », « Forward »
or « Backward » (Kutner et al., 2005).
29
Collinearity problems have been often encountered in agrometeorology because climatic or environmental variables
are never completely independent from each other. These variables can be considered as measures of a same but
different local, regional or global climate.
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
27
III. ANALYSES OF MOROCCAN CLIMATE
Exploration of climatic indicators, correlated to cereal yields to be used as predictors, focused on
characteristics of rainfall, temperature and derivatives. First, long term yearly averages were used.
Rainfall average over the country appeared to be highly influenced by North Atlantic Oscillation.
This influence has been cited in the literature for winter precipitations and verified also for total
season precipitations, as well as winter ones using historical rainfall data at the country level.
Season rainfall is variable from year to year, but relatively homogeneous on the long run, until 1980
where a change point in rainfall series has been statistically identified in many meteorological
stations, resulting in a clear rainfall decrease. Due to this change, forecasting the behavior of crops
in relation with the climate should be done using recent series, starting from 1980. Detailed
analyses of the within cropping season variation of rainfall and temperature have been undertaken,
using climatic data of series after this change point (i.e. 1980). Rainfall increases with increased
latitude, and increased elevation in agricultural zones. Temperature, on the opposite, decreases
with increased latitude and elevation. Low temperatures are encountered in mountains, high
temperatures in Saharan regions of Morocco, and cool temperatures are encountered in coastal
areas. Average temperature of the country varies little from year to year. Intra cropping season
variation of temperature is also low from year to year, with more stable temperatures during the
spring season than the winter. Rainfall, however, varies enormously from one cropping season to
another. In general, rainfall of the cropping season does not follow a Gaussian distribution. Unlike
normal distributions, means and medians of annual rainfall do not correspond, with medians being
on the dry season side. Annual distribution of rainfall is, for example, of Gamma type in the Meknes
region where average rainfall is close to median, of Log-Logistic type in the provinces south of
Meknes (Khemisset, Settat and Safi). Despite this high year to year variation, distribution of intra-
season rainfall is relatively stable, with a rate of accumulation of dekadal rainfall fitting a simple
linear regression, and where the slope of the regression is generally an indicator of the whole
season rainfall. Ombrothermic index and Length of the Growing Period (LGP) has been studied,
using Ombrothermic index (ratio of rainfall over temperature) and FAO method (ratio of real
evapotranspiration over rainfall) both based on rainfall and temperature. The Ombrothermic index
shows, a growing period of six months long on average, situated between the months of October
and March, but variable from 2 months in Saharan regions to 8 months in the mountains. The FAO
method provided an estimate of LGP in addition to periods of cultural interventions, taking into
account the water capacity holding of soils. However, due to climate change, LGP and its position in
the cycle may be modified.
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
28
1. INFLUENC E OF NORT H ATLAN TIC OSC ILLATION
The climate in Morocco is under the influence of extra-tropical circulation modes (Ward et al.,
1999; Glueck and Stockton, 2001; Herrera et al., 2001; Knippertz et al., 2003). Geographically, there
are three relatively homogeneous regions (in addition to Moroccan Sahara) with regard to
precipitations (Knippertz et al., 2003):
1. Atlantic region which includes northern and western parts of Morocco ;
2. Mediterranean region which includes north-east of Morocco, north-west of Algeria on the
Mediterranean coast ;
3. Southern region, which includes all Moroccan or Algerian regions, south of the Atlas
Mountains.
Differences in precipitation pattern between the three regions are probably due to orography, as
they are separated by the Mountain chains (High Atlas, Middle Atlas, Anti-Atlas, and Rif). In
addition to the North Atlantic Oscillation (NAO), rainfall during the winter season in the Atlantic
region is influenced by the southward shift of medium latitude disturbances, by local depressions,
and by advection from the west bringing wet air. Winter precipitation in the Mediterranean region
is resulting from the western Mediterranean depressions and moist air advection from the
northwest.
The region south of the Atlas Mountains has a more complex winter rainfall, resulting from the
transport of moisture from the Atlantic along the southern flank of the Atlas Mountains via a
stream south, enhanced cyclone activity at the Canary Islands and the occurrence of cyclones in the
southwest of the Iberian Peninsula (Driouech, 2010).
The North Atlantic Oscillation (NAO) has a great influence on precipitation in Morocco, as shown in
Figure 5 by the strong negative correlation between the NAO index and rainfall. The average index
of the NAO, during the months of September to February, explain three-quarters (R2 = 74%) of the
variation of the national average rainfall (cumulated from September till May) (Figure 5).
El Niño (ENSO) austral oscillation exerts also some influence on precipitations in Morocco. The
warm phase (positive) of the ENSO phenomenon is associated with a reduction in precipitations,
particularly during the spring season (Nicholson and Kim, 1997; Ward et al., 1999).
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
29
Figure 5: Correlation between the North Atlantic Oscillation (NAO) index, averaged from
September to February and cumulated precipitation from September to May in Morocco (data of
1979 to 2011)
2. CLIMATE CHANGE
A change point analysis performed on 15 Moroccan meteorological stations, detected a break in
the time series of rainfall around 1978 and 1981, for five of the 15 meteorological stations (Balaghi,
2006) (Table 2). For some stations (Tangier, Kenitra and Ouarzazate) change points was detected at
earlier dates. Annual rainfall decreased drastically after 1980, by 151 mm in Meknes and Oujda, and
by 82 mm in Fes, corresponding to a reduction of the annual rainfall of more than 25%. The change
point analysis also showed that the climate remained steady, from the 1980 onward, from a
statistical point of view. Of course, three very wet seasons of 2008-2009, 2009-2010 and 2010-2011
were exceptional. These results are also consistent with the work of Hurrell and Van Loon (1997) in
their analysis of anomalies caused by North Atlantic Oscillation on the southern part of Europe and
the Mediterranean. Both authors have detected particularly dry periods between 1981 and 1994 as
compared to the period of 1951-1980. The rainfall change was also observed, during the 1970s in
the West and Central Africa, but more significantly in the northwest of this region (Paturel et al.,
1998). Morocco seems to have been very sensitive to global warming since the drastic reduction of
rainfall that occurred in Morocco, has coincided with the rapid increase in air temperature across
the globe (Brohan et al., 2006). Since 1979, the rate of warming has been twice as fast on land than
on the ocean and in the last century, the Arctic has warmed by almost twice the global average rate
(IPCC, 2007). In addition to the decline in rainfall, the climate warmed significantly in Morocco,
during the period 1961-2008, from 0.2 to 0.4 °C per dekad (Driouech, 2010).
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
30
Table 2: Change point analysis in chronological series of cumulated rainfall between September
and May, at the provincial level (Source: Balaghi, 2006).
Province
Series
Rainfall
Change point Test
Average (mm)
Mann and Pettitt
Buishand
Lee and Heghinian
Tangier
1932-2004
770
1948 (0,0262)
(0,0001)
1948 (0.3214)
Tetouan
1938-2004
671
ns
ns
1971 (0.1669)
Kenitra
1951-2004
571
1972 (0,0375)
(0,1000)
1972 (0.1697)
Meknes
1932-2004
535
1980 (0.0009)
(0.0001)
1980 (0.2618)
Rabat
1931-2004
510
ns
ns
2002 (0.1435)
Fes
1915-2004
483
1978 (0.0846)
(0.1000)
1978 (0.0970)
Casablanca
1903-2004
399
ns
ns
1978 (0.0271)
El Jadida
1932-2004
371
ns
ns
2002 (0.0555)
Safi
1901-2004
353
ns
ns
2003 (0.0403)
Settat
1910-2004
353
1942 (0.0861)
(0.0500)
1980 (0.0781)
Essaouira
1894-2004
305
ns
(0.1000)
1898 (0.1362)
Oujda
1932-2004
297
1981 (0.0045)
(0.0001)
1981 (0.2376)
Agadir
1922-2004
232
ns
ns
Ns
Marrakech
1919-2004
229
ns
ns
1919 (0.2433)
Ouarzazate
1932-2004
91
1950 (0.0842)
(0.1000)
1950 (0.1564)
Ns: non-significant test.
This rainfall change in Morocco has also been observed by Knippertz et al. (2003), and Chaponniere
and Smakhtin (2006), Driouech (2010) and Sebbar et al. (2011). The decline was mainly severe in
areas of the Middle Atlas Mountains and arid areas of the Oriental zone. Knippertz et al. (2003)
explained this change by disturbances of the North Atlantic Oscillation, but no explanation of the
underlying causes was given.
Rainfall reduction that occurred around 1980 in Morocco had a direct impact on agricultural
research strategy at INRA. The institute indeed launched in 1981 an ambitious research programme
for arid and semi-arid environments, named « Aridoculture
30
Research Programme», in
collaboration with the Mid-America International Agricultural Consortium composed of five
American universities of the Middle West (MIAC), and with financial support from the Agency of
International Development of the United States (USAID) and the government of Morocco. The
objective of the Aridoculture Research Programme was to increase and stabilize the production of
cereals, food legumes and forages in arid and semi-arid environments of Morocco (250 to 450
mm/season), while developing profitable production systems to improve the rain use efficiency. In
terms of agrometeorology in particular, this change point in rainfall led INRA revise its climate
analysis methodologies toward a research programme on cereal yield forecasting in the objective of
30
The term « aridoculture » is not recent, dating back at least to the 1950s, having been quoted by Varaldi-Conia in
1953. It originates from the word "dry farming" imported from the United States of America. Already at the beginning
of last century, Augustine (1910) launched an advocacy in favor of this type of agriculture in North Africa.
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
31
country’s food security. A new research opportunity in the field of agrometeorology has led its way
to agricultural research during the recurrent droughts between 1980 and 2008, during which
agricultural production became more than ever dependent on rainfall conditions.
A major conclusion of all this work is that the analysis of the climate in Morocco could now be
carried out, on the homogeneous series of rainfall after the change point which occurred around
the year of 1980. Revealed break also meant that it became hazardous to forecast future climate
behavior based on past experience prior to 1980, including the use of frequency analysis. The
recommended series for analyzing Moroccan climate must take account this change point, and
therefore include data series posterior to 1980.
3. TE MPE RATURE
3.1. SPATIAL VARIATION OF TEMPERATURE
The long term average of mean temperature in Morocco (average over 1950 to 2000) varies
geographically from 22 to 24 °C in summer and 12 to 14 °C in winter, a variation range from 1 to
2.25 (Figure 6A) . The lowest temperatures are encountered in the mountains and the highest in
the desert. The daily cumulated average temperatures between September and May is maximum in
the Sahara site Laâyoune, (7,860 °C) and minimum in the Mountain site Ifrane, (5,110 °C). The
variation in average temperatures between stations remains low, ranging from 11% at the
beginning of the season to 9% at the end of the season.
The long term average of minimum temperature (1950-2000) of the coldest month varies from -
15.0 to -13.6 °C (range of variation of 1 to 2) and decreases from the Atlantic coast to the
mountains (Figure 6B). In general, the highest minimum temperatures occur in desert areas. Some
varieties of cereals, of winter type, fail to flower and produce seed if they do not accumulate
enough cold early in their cycle, and remain in a vegetative state. This kind of event has happened
several times in Morocco, when imported barley grain from Europe and distributed to farmers for
animal feed was planted. It is therefore necessary to be aware of the cooling requirements of
introduced varieties from cool regions like those in the northern hemisphere, before growing them.
The long-term average of maximum temperature of the hottest month, which is usually July (1950-
2000), varies from 22 to 46 °C, a wide range. In general, the highest temperatures occur in the
desert areas and the lowest maximum temperatures occur in the mountains, but also on the sea
coast (Figure 6C). The average daily temperatures accumulated between September and May is
highest in the Saharan site Laayoune, (7,860 °C x days) and lowest in Mountain site Ifrane (5,110 °C
x days). The variation in average maximum temperatures between stations remains low, ranging
from 11% at the beginning of the season to 9% at the end of the season. Biologically, high
temperatures can have negative effects on grain, at flowering, photosynthetic process and
mobilization of carbohydrates to grain. At the metabolic level, proteins, especially enzymes
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
32
responsible for the synthesis, denature at high temperatures and lose their primary functions.
Temperature range or amplitude, as a difference between maximum and minimum temperatures,
varies on average from 5.7 to 17.5 °C (a range of 1 to 3) with the highest amplitude obtained in the
Mountains and surroundings (Figure 6D). The lowest amplitude is encountered on the coastal
areas. Geographical distribution of crops species depends largely on this amplitude. For example,
tropical crops like Banana tree thrive on the coast where temperature amplitudes are lowest.
A
B
C
D
Figure 6: Maximum temperature of the warmest month (A), minimal temperature of the coldest
month (B), annual mean temperature (C) and temperature annual range (D) in Morocco (these
maps were created from data taken from www.worldclim.org at the spatial resolution of 1 km,
Series from 1950 to 2000, Hijmans et al., 2005)
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
33
Temperature affects plant development as a whole; it is particularly a major determinant of plant
phenology. Effects of cumulated temperature on the development of phenological stages of crops
have been largely documented in the scientific literature, mainly on wheat which requires a specific
sum of temperatures. In Morocco, soft wheat reaches flowering stage at approximately a
cumulated sum of daily temperatures of 1,750 °C x days an physiological maturity at 2,750 °C x
days, variable according to water and mineral conditions of the crops. Obviously, variation from
station to station in phenological stages development is associated with variation of cumulated
temperatures. For wheat, length of the biological cycle increases with reduced temperatures, from
south to north, and from the plains to the Mountains. Range of variation in wheat flowering and
maturity between extreme cereal regions (south vs. Mountain) can be as high as three months.
Wheat harvest starts mid-May in the Southern provinces of Rhamna and Abda and ends three
months later in the Atlas Mountain regions.
3.2. INTER-ANNUAL VA RIATION OF TEMPERATURE
Similar to rainfall, inter-annual variation of temperature, at the country level, can be apprehended
using cumulated daily temperatures
31
. Cumulated daily temperatures between September and May
(in °C), averaged over 35 synoptic stations used to compute country average, is relatively stable
among years, with a low variation of 10.47%. Range of variation amounts to 475 °C x days at the
end of May, ranging from a minimum of 4,300 °C x days during the 2008-2009 cropping season to a
maximum of 4,775 °C x days during the 2002-2003 cropping season (Figure 7).
Differences between cropping seasons in cumulated temperature are expressed between
September and mid-February. After mid-February rate of increase in temperature is similar for all
seasons, as indicated by paralleled curves in Figure 7. Spring temperatures seem to be stable across
seasons, as opposed to autumnal and winter temperatures.
While variation of cumulated temperature until the end of May is low, it is maximal at the 3rd dekad
of February (CV=4.9%), and minimal (CV=2.9%) at the 3rd dekad of May, indicating that most of the
temperature variation occurs during the winter. However, data series used was limited to the
period of 1999-2009. The 2008-2009 cropping season had an exceptionally high and regularly
distributed rainfall, and cool temperatures.
31
The concept of sum of temperatures (or degree x days) is used to express the plant requirements for an accumulated
quantity of heat. It is influenced by genetic variability of cultivated plants, extreme heat spells, need for low
temperatures, water constraint and photoperiodism. This concept is used in numerous models of agrometeorological
simulations of crops and pests development.
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
34
Figure 7 : Sum of daily average temperatures, in growing degree days, along cereal cropping
season, at the country level. Growing cycles from 1999 to 2009 (year of harvest) are sorted out in
descending order of cumulated temperature.
Unlike, year to year variation of cumulated temperature at the country level, it is higher at the level
of meteorological stations. For Meknes station, for example, the variation increases starting from
September, to a maximum in February, and decreases there after (Figure 8). Coefficient of variation
fluctuates between a minimum of 7.5% at the start of the season (September) to a maximum of
42.5% in the winter (February). This increase in temperature variation between September and
February explains increased divergence among curves of cumulated temperatures shown in Figure
7.
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
35
Figure 8 : Evolution of coefficient of variation of average inter-annual dekadal temperatures at
the country and Meknes levels (data of 1999 to 2009).
3.3. SEASONAL VARIATION OF TEMPERATURE
Temperature in Morocco is high in summer (July, August and September), diminishes progressively
to attain a minimum in winter (between December and January) and increases in spring (starting
from February). Along the cropping season, dekadal maximum temperature decreases on average
from 32.3 °C at the first dekad of September to 18.5 °C at the last dekad of December (data series
of 1999 to 2009), a drop of 13.8 °C (Figure 9); minimal dekadal temperature also decreases on
average from 17.2 °C at the first dekad of September to 3.8 °C for the last dekad of January, a drop
of 13.4 °C, similar to maximum temperature.
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
36
Figure 9 : Average maximal (red curve), mean (yellow curve) and minimal dekadal temperature
(blue curve), along the cereal growing season. Red dots correspond to dekadal values of maximal
temperature and blue dots to minimal temperature, from 1999 until 2009.
4. RAINF ALL
4.1. SPATIAL VARIATION OF RAINFALL
The first rainfall map of Morocco was elaborated by the German geographer Theobald Fischer early
1900. It contains some errors due to lack of meteorological observations and to meager technical
resources at that time. Many rainfall maps have been developed since then, with varied degree of
accuracy (Augustin, 1921; Jury and Dedebant, 1924; Roux et al., 1949). All these maps reveals high
spatial heterogeneity of total annual rainfall in Morocco like in Figure 10, varying from 13 mm in the
Sahara to above 1000 mm at Tangier. This heterogeneity had already been illustrated by the first
global work on rainfall in Morocco done by Augustin Bernard (1921). This author made a
classification of vegetation in Morocco, on the basis of total rainfall received during the cropping
season, into four classes: (1) regions of spontaneous vegetation where crops can be grown under
favorable rainfall conditions (> 400 mm), (2) intermediate regions where crops can also be grown
under conditions of deep soils and well distributed rainfall (300 < Rainfall < 400 mm), Steppic
regions where annual rainfall lies between 200 and 300 mm, and Saharan regions of less than 200
mm annual rainfall.
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
37
Figure 10 : Country average rainfall cumulated between September and April. The map created
from data of www.worldclim.org (Hijmans et al., 2005). Red dots refer to main cities of Morocco.
Annual rainfall data were first interpolated to cover all points of the map by Hijmans et al. (2005).
The map presented in Figure 10, shows a spatial distribution of annual rainfall heavily concentrated
over the Middle Atlas and north-west. The map illustrates once more, increased rainfall gradient
with latitude and elevation. In the same Figure 10, rainfall seems to be overestimated in the High
Plateaus of Eastern Morocco, between Oujda and Bouarfa. In this region, as in any other steep
region of the world not satisfactorily disserted with climatic stations, interpolation errors of rainfall
are high (Hijmans et al., 2005).
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
38
4.2. INTERANNUAL VARIATION OF RAINFALL
Long term rainfall average in Morocco is around 364 mm, varying from a minimum of 198 mm
recorded in 1994-1995 to a maximum of 610 mm recorded in 2009-2010 season. Rainfall data of 35
synoptic stations and 24 years (1988 to 2011), were used to arithmetically compute long term
average of cumulated rainfall over the cropping season, between September and May. Inter-annual
variation of rainfall at the country level is extremely high, in a ratio of 1 to 3; that is 300%
variation (Figure 11).
Figure 11 : Inter-annual variation of cumulated rainfall between September and May, at the
country level, for 1988 to2011.
The cropping season starts on the first day of September and ends the last day of May of next year.
The cropping seasons are arranged in ascending order of cumulated rainfall, and are grouped into
four classes of rainfall (200 to 300 mm, 300 to 400 mm, 400 to 500 and > 500 mm).
Cropping seasons can be grouped into four classes of rainfall with the following frequencies:
1. 200 to 300 mm: 8 seasons (1992, 1993, 1995, 1999, 2000, 2005, 2007 and 2008) ;
2. 300 to 400 mm: 9 seasons (1988, 1989, 1990, 1991, 1994, 1998, 2001, 2002 and 2006) ;
3. 400 to 500 mm: 4 seasons (1997, 2003, 2004 and 2011) ;
4. Above 500 mm: 3 seasons (1996, 2009 and 2010).
Annual rainfall cumulated over the cropping seasons of the 24 seasons are plotted in Figure 11, in
ascending order of rainfall, with a different color assigned to each class of rainfall.
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
39
Figure 12 : Geographic localization of the 35 synoptic weather stations used to generate weather
data of Morocco, on a background map of elevation provided from radar data of « Shuttle Radar
Topography Mission » (http://www2.jpl.nasa.gov/srtm/).
However, for agricultural purposes, the arithmetic mean of the 35 synoptic stations may yield a
biased estimate of the average rainfall, because used stations are not all localized in agricultural
areas of Morocco; some of them are localized in mountainous, pastoral or Saharan regions (Figure
12), not representing climatic conditions prevailing in cereal areas of Morocco. In addition, these
stations are not uniformly distributed across the country; most of them being scattered along the
coastal areas. A better coverage would definitely improve the understanding of Moroccan climate.
To reduce the bias, DMN has adapted the AURELHY
32
method (Analysis Using RELief for Hydro-
32
The AURELHY method utilizes terrain to improve the cartography of precipitations. The method is based on 3 points:
(1) automatic recognition of the statistic link existing between precipitation and surrounding relief; (2) optimal
utilization of this statistical relationship to points where no measured values are available; and (3) generation of a
regional map of precipitations, integrating effects of relief.
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
40
meteorology) of interpolation and mapping of precipitation data provided by synoptic stations to
cover inter stations
33
spaces. Initially developed by Meteo-France (Bénichou and Le Breton, 1987),
DMN has adapted it for Moroccan conditions, to produce a map of rainfall, on a dekadal data basis
for the whole territory. This method is being integrated into the National Crop Growth Monitoring
System to predict cereal harvests, named « CGMS-MAROC » (see chapter VII). It is then possible to
extract rainfall data for agricultural zones, using the agricultural mask described in section II1.5.
4.3. SEASONAL VARIATION OF RAINFALL
Most of the rainfall in Morocco is received between the months of September and May, the
months of June, July and August being generally dry (Figure 13). Rainfall distribution within the
season is random, with a great disparity between the south and the north, or between the west
and the east of the country. Rainfall increases proportionally from arid to humid regions, with more
accentuated increases in the winter than in any other time period during the season.
Figure 13 : Examples of dekadal rainfall distribution along the cropping cycle, for specified
contrasting synoptic weathers stations (averages of 1988 to 2010).
33
It is a priori possible to interpolate data of rainfall provided by climatic stations using NDVI images, due to the high
correlation between rainfall and NDVI, which is the principle of « Satellite enhanced data interpolation » (SEDI) method
developed by FAO and integrated in WinDisp (version 4) software. http://www.fao.org/giews/french/windisp/hist.htm.
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
41
4.3.1. FREQUENCY ANALY SIS OF DEKADAL RA INFALL
Seasonal variation of rainfall can be statistically modelled, using frequency analysis of its dekadal
values
34
. Frequency analysis is a statistical approach
35
which allows modelling or caching most of
rainfall distribution pattern, irrelevant of the crop under consideration. It determines probabilities
or chances of a rainfall value (dekadal or monthly) to occur based on a long time series of rainfall
data. It is therefore a pure statistical analysis method which nonetheless helps making decision in
agriculture (Jlibene and Balaghi, 2009).
Frequency analysis of dekadal rainfall, for recent series of cropping seasons after the climate
change (i.e. 1980), shows that within season distribution of rainfall is one-mode, concentrated in
winter, with a peak during the months of November, December or January (Figure 14). It also shows
that rainfall is rather concentrated in the fall, indicating the importance of the success of planting
operations for improved water efficiency, and ultimately improved crop yields. Rainfall received in
the fall helps reconstituting water reserves in the soil after the usually hot summer season.
Figure 14 : Distribution of dekadal rainfall, at the country level, for 5 frequency levels (0.1; 0.3;
0.5; 0.7 and 0.9). Data series from 1988 to 2011.
34
Worth keeping in mind that the frequencies associated with a rainfall value of a given dekad, are independent from
frequencies of other dekad’s rainfall.
35
In frequency analysis, observations are considered as a random sample representative of the climate under study,
assuming that time series is stationary (Péguy, 1983).
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
42
When using rainfall time series before the change point, the distribution of rainfall appears to be a
two-mode (Balaghi, 2000). Less important peaks were observed for the months of October and
March. During the summer, rainfall is infrequent and falls in the form of storms, especially near the
Atlas Mountains. The usefulness of heavy rains is proportional to the soil water holding capacity.
High rainfalls are more beneficial to crops in deep soil of heavy to intermediate texture, than light
soils or soils that poorly drain (Roux, 1938). In these last types of soils, agricultural production is
better only if rainfall is low but uniformly distributed within the cropping season. During dry
seasons, plants use water more easily in soils with low water holding capacity than in soils with high
capacity. Usually rainfall received during the fall allows development of a good plant stand in both
types of soils. However, under water stress, plant keep growing on residual water in soils with high
water holding capacity, until depletion and death of plant before seed set. In soils with low water
holding capacity, plants sense drought early due to low water reserves and thus tend to mature
earlier.
Meknes Gamma (19,912; 33,269)
Khemisset Log-Logistic (-166,57; 611,07; 7,5342)
Settat Log-Logistic (40,703; 291.71; 4.3505)
Safi Log-Logistic (-98,385; 437,78; 6,1625)
Figure 15 : Annual rainfall distribution (cumulated from September to June) at Meknes (1932-
2004), Khemisset (1986-2004), Settat (1910-2004) and Safi (1901-2004). The red curve represents
adjusted theoretical distribution curve to observed one (Balaghi et al., 2005).
2=4,945 n.s
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
43
Rainfall is distributed asymmetrically with respect to the median, with low rainfall seasons
outnumbering wet years, and therefore departing from statistical Normal (Gaussian) distribution.
Examples of distribution of annual rainfall of the stations Meknes, Khemisset, Settat and Safi are
given in Figure 15 (Balaghi et al., 2005). Annual rainfall during the growing season (September to
June), follows a theoretical non-Gaussian distribution, according to the χ2 test of Snedecor and
Cochran (1989). The distribution is of Gamma type for Meknes (1932 to 2004 data series) and Log-
Logistics for Khemisset (1986 to 2000 data series), Settat (1910 to 2004 data series) and Safi (1901
to 2004 data series). The annual rainfall is distributed more evenly in Meknes than in the other
three provinces, with years of low rainfall being more frequent further south.
4.3.2. LINEAR APPROXIMATION OF DEKADAL RAINFALL
Graphical representation of cumulated dekadal rainfall, from September to May, indicates that
such curve can be approximated by simple linear regression. Indeed, the dekadal rainfall
accumulated during the growing season (from September to May) rose almost linearly (Figure 16),
with high and significant coefficients of determination (R2) greater than 91%.
Figure 16 : Dekadal cumulated rainfall from September till May, at the country level. Cropping
cycles of 1988 to 2011 are arranged in ascending order of cumulated rainfall.
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
44
Given the quasi-linearity of the cumulative dekadal rainfall, it is possible to fit a linear regression
model. The R² of these models are very high, exceeding 90% (Table 3). For example, the 1988-1989
growing season recorded a rainfall that regularly accumulates at a rate of 14.94 mm/dekad, with
very high R² equal to 99%. Relatively lower R², which may fall below 95%, indicates the occurence of
drought. The increase rate of cumulated dekadal rainfall starting from September is measured by
the slope of the linear regression function over dekads. It varies from 7.64 mm/dekad in the case of
the historic droughty season 1994-1995 to 28.63 mm/dekad 2009-2010 in the case of the rainiest
season in the post 1980 series of data; 1980 was the year of breaking point in chronological series
and announcement of climate change in Morocco.
Table 3: Linear regression models applied to cumulated dekadal rainfall over the growing season
(from September to May) at the country level.
Growing seasons (Years of harvest) are classified in ascending order of rainfall.
Season
Rainfall (mm)
Intercept (mm)
Slope (mm/dekad)
R2
1995
198.2
-4.89
7.64
0.96
1999
226.9
-34.77
10.50
0.95
2005
228.4
-12.22
10.30
0.95
1993
236.9
-31.75
9.73
0.97
2007
237.1
-18.29
9.29
0.98
2000
263.6
13.36
9.50
0.93
2008
268.1
-16.59
11.48
0.97
1992
280.3
7.73
10.28
0.96
2001
320.6
-13.33
14.61
0.91
2002
332.2
-47.55
14.20
0.96
1994
344.4
-2.77
14.91
0.94
1989
350.9
-36.57
14.94
0.99
1998
377.3
-5.67
16.07
0.95
2006
377.5
-45.51
16.80
0.98
1988
387.7
-17.09
17.04
0.96
1990
388.5
-20.02
16.75
0.95
1991
395.5
-60.24
18.36
0.97
2003
411.3
-12.26
17.80
0.95
2004
419.2
-10.44
16.84
0.95
1997
466.4
-37.47
21.02
0.91
2011
492.9
-34.67
19.85
0.98
2009
547.9
6.93
23.61
0.95
1996
591.3
-109.58
28.13
0.94
2010
604.8
-88.54
28.63
0.93
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
45
The rate of dekadal rainfall accumulation determines the total rainfall in the growing season, with a
coefficient of determination close to unity (R2 = 97%).
The rate of dekadal rainfall increase is on average de (Figure 17):
9.82 mm/dekad for the growing seasons with total rainfall between 200 and 300 mm
(years of harvest: 1992, 1993, 1995, 1999, 2000, 2005, 2007 and 2008) ;
15.97 mm/dekad for the growing seasons with total rainfall between 300 and 400 mm
(years of harvest: 1988, 1989, 1990, 1991, 1994, 1998, 2001, 2002 and 2006) ;
18.88 mm/dekad for the growing seasons with total rainfall between 400 and 500 mm
(years of harvest: 1997, 2003, 2004 and 2011) ;
26.79 mm/dekad for the growing seasons with total rainfall above 500 mm (years of
harvest: 1996, 2009 and 2010).
Figure 17 : Average increment rate of rainfall per dekad (x-axis, in mm/dekad), for 4 classes of
cumulated rainfall (200 to 300, 300 to 400, 400 to 500 and > 500 mm). Data are country averages
of 1988 – 2011 series.
4.4. SEASONAL VARIATION OF RAINFALL
Similar to inter-annual variation, intra-annual rainfall distribution can also be modelled, based on
total rainfall or the latitude of the station, as total rainfall is proportional to the latitude. Monthly
rainfall from 1988 till 2005 is shown in Figure 18, sorted out in increasing order. This figure was
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
46
produced based on spatial interpolation of monthly rainfall using a Geographic Information System
(GIS). In spatial terms, annual rainfall increases with latitude, from south to north. Four classes of
rainfall can be distinguished rainfall: an arid class with less than 250 mm, a semi-arid class with
rainfall between 250 and 350 mm, a sub-humid class with rainfall between 350 and 450 mm, and a
wet class with rainfall over 450 mm (Jlibene, 2011).
Total season rainfall is generally concentrated within the period of September to May, for all
seasons, regardless of the amount of total rainfall. However, with increased volumes, rainfall
becomes more concentrated in winter season showing higher peaks that seem to disappear during
dry seasons (Figure 18).
Figure 18 : Monthly rainfall distribution, as influenced by the overall volume of rainfall in
Morocco (data of 1988 to 2005).
Cropping cycles are arranged from high to low rainfall. Areas of similar rainfall are represented by a
range of blue colors, from clear (low rainfall) to dark blue (high rainfall).
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
47
Intra-annual rainfall distribution also differs, depending on latitude which is correlated to total
amount of rainfall cumulated during the growing season. Figure 19 shows intra-annual rainfall
distribution for meteorological stations located along the Atlantic coast. In general, rainfall
distribution appears to be similar in all latitudes, but peaks of rainfall during winter flatten with
decreased latitude, from north to south of the country.
Figure 19 : Monthly rainfall distribution for 15 contrasting synoptic weather stations scattered
along the Atlantic coast (data from 1988 till 2005). Stations are arranged in ascending order of
latitude. Red lines represent points of equal monthly rainfall, of 20 mm, 40 mm, 60 mm and 80
mm.
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
48
5. GROWI NG SEASON FROM OM BRO THERMIC DIAGRAM
The Ombrothermic diagram of Bagnouls and Gaussen (1953) assumes that crop growth is possible
as long as monthly rainfalls are greater than two times the monthly temperatures. Used to estimate
the length of the growing period, this diagram shows that the country average growing season for
cereal crops extends from October to March (Figure 20). During these months, monthly rainfall
values are greater than two time average monthly temperatures. This diagram integrates the two
most important factors for growth and development of crops: rainfall and temperature.
Figure 20 : Ombrothermic diagram, at the country level. (Average rainfall and temperature, from
a series of 1999 to 2009).
Length of the growing season of crops varies greatly with latitude (Figure 21). It increases with
latitude from south to north, in the Atlantic regions of the country, situated between the Atlas
Mountains and the Atlantic Ocean. The increase in Ombrothermic index with latitude is
exponential, changing from less than 1 at the Saharan site Dakhla to more than 6 at the northern
site Chefchaouen. This result was expected since there are simultaneous increase in rainfall and
decrease in temperature with higher elevation, increasing the ratio of rainfall to temperature more
than proportional.
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
49
Figure 21: Average Ombrothermic index (from September to May) as influenced by latitude.
Monthly data series of 1999-2009 for 25 different synoptic stations were used.
The growing season of crops becomes shorter and shorter with lower elevation south, until
inexistent in Saharan provinces (Table 4). In other parts of the country, the growing season starts
with the first rains of October (Table 4). For most cereal regions, the length is limited only by the
end of the season, which could be in March for Khouribga in the center, April for Chefchaouen in
the north, and May for Ifrane in the Atlas Mountains. In general, the growing season ends in April in
regions above 33° 30’ latitude.
Ombrothermic index can be spatially presented, to map the end of growing season dates
throughout the country. This map presented in Figure 22 shows the season end, which varies across
latitude and elevation, between March and May. The season ends in March throughout the area
between Agadir and Bouarfa, south of the 32° 30’ parallel excluding areas of medium and high
elevation. It ends in April in the coastal area north of Safi, the central plains and plateaus of the
country in addition to Anti Atlas area. The season continues through May, in high elevation area of
Rif, Middle and High Atlas Mountains, and some high elevation area in eastern plateaus.
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
50
Table 4 : Ombrothermic Index, as a ratio between precipitation (P) and temperature (T), by
synoptic station. Unit of P/T is in mm/°C.
Indices lower than 2 indicate dry months not suitable for growth (in orange) and indices greater
than 2 indicate humid month suitable for growth (in green). Stations are arranged in descending
order of index. (Monthly rainfall and temperature series of 1999 to 2009).
Sep
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Latitude
(North)
Longitude
(West)
Elevation
(meter)
Ifrane
1.3
6.8
15.7
33.1
26.3
20.3
10.5
9.1
4.1
33 ° 30'
5 ° 10'
1663
Chefchaouen
1.2
5.3
7.8
14.3
8.0
9.6
5.5
5.6
1.3
35 ° 05'
5 ° 18'
300
Tetouan
1.8
5.2
6.4
7.9
5.5
6.7
4.6
4.0
1.5
35 ° 35'
5 ° 20'
5
Larache
1.0
4.9
6.5
9.4
6.5
3.8
4.4
3.5
1.5
35 ° 11'
6 ° 08'
47
Tangier
1.4
5.0
5.8
8.1
6.4
4.7
4.4
3.5
1.6
35 ° 43'
5 ° 54'
15
Taza
0.8
3.1
4.7
7.2
6.3
5.7
3.0
3.3
1.3
34 ° 13'
4 ° 00'
509
Kenitra
0.7
3.3
6.4
7.1
6.1
4.3
3.3
2.3
1.1
34 ° 18'
6 ° 36'
5
Rabat-Salé
0.5
3.4
6.6
6.9
5.7
4.6
3.3
2.3
1.2
34 ° 03'
6 ° 46'
75
Fes
0.8
3.6
4.6
6.5
6.5
4.8
2.7
3.5
1.4
33 ° 58'
4 ° 59'
571
Meknes
0.5
2.8
4.0
6.9
5.4
4.6
2.7
2.7
1.2
33 ° 53'
5 ° 32'
548
El Jadida
0.3
2.7
4.1
6.3
3.7
3.3
2.4
1.5
0.7
33 ° 14'
8 ° 31'
27
Casablanca
0.4
2.6
4.5
4.3
4.5
3.2
2.3
2.0
0.6
33 ° 34'
7 ° 40'
56
Nador
0.6
3.2
4.3
3.1
2.6
3.7
2.9
2.5
1.0
35 ° 09'
2 ° 55'
7
Sidi Slimane
0.3
2.3
4.1
5.2
4.4
2.5
2.3
1.9
1.0
34 ° 14'
6 ° 03'
52
Al Hoceima
0.5
3.8
4.6
2.8
3.0
3.0
2.2
2.2
1.0
35 ° 11'
3 ° 51'
12
Kasba Tadla
0.2
1.6
3.5
4.2
4.6
3.8
2.4
1.8
1.1
32 ° 52'
6 ° 16'
507
Béni Mellal
0.2
1.8
3.6
4.3
3.7
3.3
2.5
1.8
0.9
32 ° 22'
6 ° 24'
468
Safi
0.1
2.1
3.6
5.1
3.0
3.7
2.2
1.3
0.5
32 ° 17'
9 ° 14'
43
Khouribga
0.2
1.3
3.3
5.2
4.0
3.6
2.0
1.2
0.7
32 ° 52'
6 ° 58'
770
Oujda
0.4
1.8
3.2
2.9
3.2
2.7
2.1
1.6
1.1
34 ° 47'
1 ° 56'
465
Essaouira
0.2
1.6
2.2
3.5
2.5
3.1
1.9
1.1
0.3
31 ° 31'
9 ° 47'
7
Midelt
0.8
2.1
1.7
2.7
1.2
1.9
1.0
1.0
0.9
32 ° 41'
4 ° 44'
1508
Marrakech
0.2
1.1
1.7
1.9
1.8
1.7
1.4
1.0
0.5
31 ° 37'
8 ° 02'
464
Agadir
0.1
0.9
1.8
2.6
1.2
2.1
1.5
0.6
0.1
30 ° 23'
9 ° 34'
23
Taroudant
0.1
0.8
1.8
2.7
0.9
2.2
1.3
0.7
0.1
30 ° 30'
8 ° 49’
264
Bouarfa
0.7
2.0
1.5
1.4
1.4
1.4
1.0
0.8
0.4
32 ° 34’
1 ° 57'
1142
Tiznit
0.1
0.6
1.2
1.6
1.0
1.5
1.1
0.6
0.1
29 ° 41'
9 ° 44'
260
Errachidia
0.4
1.6
1.1
0.6
1.2
1.2
0.6
0.5
0.4
31 ° 56'
4 ° 24'
1037
Sidi Ifni
0.1
0.7
1.4
1.1
0.8
1.1
0.8
0.3
0.1
29 ° 22'
10 ° 11'
50
Tantan
0.0
0.8
0.5
0.8
0.9
1.1
0.6
0.1
0.0
28 ° 10'
10 ° 56'
45
Laâyoune
0.0
0.1
0.2
0.6
0.6
0.8
0.1
0.1
0.0
27 ° 10'
13 ° 13'
64
Dakhla
0.4
0.2
0.0
0.1
0.2
0.1
0.0
0.1
0.0
23 ° 43'
15 ° 56'
11
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
51
While the end of the growing season is in general influenced by latitude and elevation, a particular
spot in the map between Safi west and Beni Mellal east, and between El Jadida north and
Marrakech south (in blue, Figure 22), seems to be an exception. This spot includes arid zones of
Rehamna
36
and Chemaia (bordering the provinces of Marrakech, Safi and Chichaoua),
corresponding to the plateau of phosphates in Morocco. The same spot has been reported in maps
of rainfall as early as 1921 (Augustine, 1921) or agro-climatic maps developed for Morocco (Goebel
et al., 2007; Sebbar et al., 2011).
The Ombrothermic index does not take into account the capacity of soils to hold water, which could
further extend length of the growing season due to soil moisture. Soils with high water holding
capacity maintain humidity for root system to use for a time longer than that expressed in the
Ombrothermic map in Figure 22.
The length of plant growing season determined by the Ombrothermic index, matches that of most
annual fall sown crop cycles, particularly cycles of autumn cereals. Crop cycle may be referred to as
cropping cycle or cropping season in the rest of the text. According to Augustin (1921) the growing
season can be divided into three periods of 4 months each: (1) period of seed preparation, from 1st
of September to December 31; (2) period of plant development, from 1st of January to April 30th ;
(3) period of harvest, from 1st of May to August 31th. Cereals are sown in the fall season, around
the month of November, complete their vegetative growth and development in the winter season,
and the reproductive growth in spring season. The rainy season starts in October, but cereal
sowings are rarely done in this month. They usually start in November for practical reasons mainly:
Ease of tilling soils for seed bed preparation and weed control; the first rains of the month
of October loosen soils, and induce emergence of weeds ;
Optimization of crop growth and development, as November sowings seem to be optimal
for cereal yields expression.
Temperatures during October are still relatively high, which tend to accelerate development on the
expense of biomass accumulation, reducing the crop cycle and therefore yields. In dry areas like
the Abda region, farmers plant their cereal crops before the first rains, to take advantage of early
rains.
Physiological maturity of cereals, which marks the end of seed growth and the beginning of
desiccation, is reached at the end of the rainy season. Cereal harvests are usually done at grain
moistures ranging from 12 to 14%. The range of final harvest maturity date extends over more than
three months depending on region, ranging from May in the southern regions (e.g. Haouz,
Rehamna and Abda) to August in Mountain regions. Farmers are not sufficiently equipped to
36
Readers are referred to the publication of De Martonne et al. (1924), which describes the morphology and the aridity
of Rehamna.
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
52
harvest their crops earlier in the season at grain moistures above 14%. In majority, they await for
the arrival of combine harvesters coming up from the south, where harvests begin earlier.
Figure 22 : Geographic areas where cereal growing seasons extend from September until March
(areas in blue, green and brown colors), or until April (areas in green and brown colors) or until
May (areas in brown color), according to Ombrothermic index of Bagnouls and Gaussen (1953).
The map was generated from data of www.worldclim.org (Hijmans et al., 2005).
6. LENGT H O F THE GROWI NG PERIOD
The ratio of evapotranspiration to rainfall was used by FAO (1978) to identify lengths and time
limits (start and end) of major phases within the growing cycle, mainly:
Period of soil tillage for seed bed preparation ;
Period of crop growth (growing cycle) ;
Humid period which is part of the period of crop growth.
Agrometeorological Cereal Yield Forecasting in Morocco
Analyses of Moroccan climate
Balaghi R., Jlibene M., Tychon B., Eerens H.
53
Based on identified limits of each period, dates of sowing and harvests can be identified.
This method was used in Morocco by Jlibene and Chafai (2002) to determine optimal cultural
phases for wheat crop in the north-western part of the country (Figure 23).
Figure 23 : Lengths and time limits of different periods within the cereal crop cycle: (1) soil
preparation (W in purple), (2) period of growth (green) which includes sowing period (S), (3)
humid period (light green), flowering (F) and grain filling (G); and (4) grain maturity and ripening
(M in pink). The humid period includes times of emergence (L), tillering (T), and heading (H) of
cereal crops in the north-western Morocco (Jlibene and Chafai, 2002). Weather data are a series
of 70 years from Arbaoua Meteorological station (Data source: ORMVAL).
It turns out, according to this study, that for the Loukkos region in particular and North-West
Morocco in general (Figure 23):
The period of land preparation starts September 23th and is limited by the start of the
growing period, October 18th ;
The growing period starts October 18 and ends April 19th ;
The humid period included in the growing period starts November 5th and ends Mach 22th
;
Evaporation of 100 mm of soil residual moisture would take 60 days to deplete.
Applied to wheat cultivation in this region, the weather conditions allow:
A long growing period of 183 days, 155 days for vegetative growth, and 28 days for
reproductive growth ;
Sowings have 18 days to be completed, before the fields become wet and difficult to
access (November 5th), preferably during the first week of November. Major vegetative
stages take place during the humid period: emergence, tillering, elongation, and heading ;
At the end of the humid period (March 22th), flowering should take place, and seeds have
28 days to complete their development ;
Remobilization of carbohydrates from leaves and stems to seeds can be done outside the
growing period using reserve water available in the soil.
Agrometeorological Cereal Yield Forecasting in Morocco
Analysis of cereal production
Balaghi R., Jlibene M., Tychon B., Eerens H.
54
IV. ANALYSYS OF CEREAL PRODUCTION
The three main cereal crops in Morocco include soft wheat, durum wheat and barley. They are
grown for grain, and straw is used as animal feed. Grain of soft wheat is mostly used to baking
bread and biscuit; grain of durum is used to make pasta and semolina (couscous) and the grain of
barley is used as feed and a small proportion of it is used as food. They are grown in all provinces
and all agro-ecological zones. However, the geographic distribution of each crop follows market
demand for yield potential and grain quality, in addition to their inherent geographic adaptive
ability. In high yield environments, soft wheat is more grown because of its high productivity. In
intermediate environments where grain quality is better expressed, durum wheat is more frequent.
In dry environments where durability of agriculture system requires diversification, barley is mostly
grown as feed in association with sheep production.
Technological trend is a result of cereal yield improvement, which is positive for durum wheat and
soft wheat, and negative for barley. The positive trend observed for both wheat crops is mostly due
to genetic improvement. The negative technological trend for barley yields does not necessarily
result from negative yield improvement, since significant genetic improvement was achieved, but
mainly from the shift of barley areas to drier environments for the benefit of soft wheat.
Technological trend, quantified at the country or the agro-ecological zone levels, is taken into
consideration in agrometeorological cereal yield forecasting in this document.
1. END USES OF CEREAL S
Harvested grains, of the three major cereal crops (soft wheat, durum wheat and barley), are
designed for various specific end uses depending on the crop. The grains of barley are used
primarily (97%) to feed animals, with a small share (7%) being used for human consumption in the
form of flat bread or semolina (Table 5). Wheat grains are used exclusively in human diet, basically
in the form of bread and pasta, and also biscuit and semolina. At the industrial level, soft wheat is
used for the production of bread and biscuit, while durum wheat is used for the production of
couscous and pasta. At the household level or artisanal small businesses, wheat is used to produce
a multitude of products made from flour and wheat grain.
Soft wheat has become the largest contributor to food security of the country's due to the large
consumption of leavened bread by Moroccans (133 kg/person x year), accounting for two-thirds of
the total cereal human consumption, followed by durum wheat which contributes to 25%. As a
consequence of the high demand on soft wheat for bread, cropped area increased on the expense
of durum wheat, barley and fallow. Areas of durum wheat and fallow were reduced, while that of
barley shifted to drier areas. The market demand for soft wheat continues to increase due to
population growth, while the agricultural area is almost constant, which limits the alternatives for
cereal grain security in Morocco to the sole option of improving land productivity.
Agrometeorological Cereal Yield Forecasting in Morocco
Analysis of cereal production
Balaghi R., Jlibene M., Tychon B., Eerens H.
55
Table 5 : Main uses of cereal grains in Morocco (Source: ONICL, 2012).
Cereal crop
End use
Consumption
share
Main use
Secondary use
Soft Wheat
Human
67%
Bread and biscuit
Noodles, other local products
Durum wheat
Human
25%
Pasta and semolina
Bread, other local products
Barley
Animal
7%
Animal feed
Semolina, flat bread
2. GEOGR APH ICAL DISTRIBUT ION OF CEREALS
Cereals have large capacity to adapt to geographical, climatic, and cultural environmental
conditions. They are grown all over the country, from the Sahara south to the Mediterranean
border north, and from the Atlantic Ocean west to borders of Algeria east. Cereals occupy more
than two-thirds of agricultural land (Figure 24). Although they are grown in 36 provinces,
production is mainly concentrated in the coastal areas, followed by central and mountainous
regions. Cereals are part of most crop rotations, in addition to continuous cereal cropping, and are
grown on a wide range of soils, production systems and environments. The distribution of cereal
areas by agro-ecological zones is as follows:
Favorable zone: 39.3% of total cereal production on 31.1% area ;
Intermediate zone: 18.3% of total cereal production on 16.8% area ;
Unfavorable South: 18% of total cereal production on 25.5% area ;
Unfavorable East: 8.5% of total cereal production on 9.8% area ;
Mountainous zone: 12.6% of total cereal production on 10% area ;
Saharan zone: 3.3% of total cereal production on 4.1% area.
Soft wheat production is mainly concentrated in the Favorable zone, i.e. humid and sub-humid
agro-systems, which account for 53% of total production (Table 6). Intermediate, Unfavorable South
and Mountains zones account for 40% of wheat production while 8% of the production comes from
Unfavorable East and Saharan zones. For durum wheat, 41% of the production comes from the
Favorable zone fallowed by the Intermediate zone (25%). The Mountain and Unfavorable South
zones produce 27%, and Unfavorable East and Saharan zones produce about 8%. Barley production
is spread across four major zones: Unfavorable South with 28% of the total production, followed by
Favorable (22%), Intermediate (21%) and Unfavorable East (16%). Areas follow the same
distribution as the production for the three cereals. In contrast, productivity is driven by the rainfall
gradient and the rate of irrigation.
Agrometeorological Cereal Yield Forecasting in Morocco
Analysis of cereal production
Balaghi R., Jlibene M., Tychon B., Eerens H.
56
Table 6: Partitioning of cereal production, area and yield, by agro-ecological zones (Average of
1990 to 2011).
Production (%)
Area (%)
Yield (Q/ha)
Agro-ecological
Zone
Soft
wheat
Durum
wheat
Barley
Soft
wheat
Durum
wheat
Barley
Soft
wheat
Durum
wheat
Barley
Favorable
53
41
22
46
39
17
15.9
13.7
11.7
Intermediate
13
25
21
13
23
18
12.7
13.7
10.4
Unfavorable East
5
4
16
7
5
15
10.0
8.7
9.3
Unfavorable South
11
15
28
18
19
36
8.4
9.6
6.9
Mountain
15
12
9
13
12
8
16.8
13.2
10.8
Saharan
3
3
5
3
2
6
12.9
20.2
6.7
Cereal productivity of soft wheat, durum wheat and barley, is strongly influenced by rainfall,
varying in space and time: in space, the northern regions are more productive than those in the
south, ranging in a magnitude of 1 to 3.5, and in time, the yield varies from about 1 to 5, 3.6 Q/ha
recorded during the dry crop season of 1999-2000 to 18.5 Q/ha recorded during the wet season of
2008 - 2009 (official agricultural Statistics, 1979-2011).
Production of soft wheat is concentrated in the Atlantic plains of the country, from arid zones south
to wet zones north. The provinces that contribute most to wheat production are: Beni Mellal
(11.4%), Sidi Kacem (10.0%), Kenitra (8.9%), Khemisset (6.7%), El Kalaa Sraghna (6.0%) El Jadida
(5.7%), Settat (5.5%), Taounate (5.3%), Meknes (4.9%) and Fez (4.4%) (data series from 1990 to
2010). These ten provinces contribute more than two thirds (69%) of the country’s wheat
production. Among these 10 provinces, eight provide 61% of the national production of durum
wheat. Each one of the remaining provinces contributes less than 4% of the total soft wheat
production.
The provinces that contribute most to the country’s production of durum wheat are: Settat (12.4%),
El Jadida (10.9%), Taounate (9.3%), Taza (7.1%), El Kalaa Sraghna (6.1%), Safi (6.1%), Beni Mellal
(5.1%) and Sidi Kacem (4.4%) (data series from 1990 to 2010). These eight provinces provide 61% of
the national production of durum wheat. Each of the remaining provinces contributes less than 4%
of the total durum wheat production.
Barley production is concentrated in arid and semi-arid zones and in mountainous areas. The
provinces that contribute the most to the national production of barley include: Settat (8.4%), El
Jadida (7.2%), Safi (6.8%), Nador (6.6%), El Kalaa des Sraghna (6.1%), Essaouira (5.3% ), Oujda
(4.6%), Khouribga (4.5%), Taza (4.4%) and Marrakech (4.2%) (data series from 1990 to 2010). These
ten provinces provide 58% of the country’s production of barley. Each of the remaining provinces
contributes less than 4% of the total barley production.
Agrometeorological Cereal Yield Forecasting in Morocco
Analysis of cereal production
Balaghi R., Jlibene M., Tychon B., Eerens H.
57
Figure 24 : Map of average production (in Mega tons) of autumn cereals (soft wheat, durum
wheat and barley confounded) by province (data series from 1990 to 2010).
The final use of the grain produced by each of the three cereals influences their geographical
distribution. Soft wheat is concentrated in favorable (wet) areas where high productivity is sought
to meet the strong demand for this commodity. Durum wheat is concentrated in less humid areas,
which are suitable for semolina pasta-making quality. Excessive moisture in the grain formation of
durum wheat, causes "yellow berry" a phenomenon that transforms the grain starch from a state of
hard glassy grain (suitable for the production of semolina) to a soft floury grain. Barley is
concentrated in areas where sheep farming is dominant, mainly in Intermediate, Unfavorable
South, and Unfavorable East agro-ecological zones.
Average yield at administrative province level can explain the overall yield average at the country
MALI
MAURITANIA
ALGERIA
MOROCCO
SPAIN
PORTUGAL
OUAD EDDAHAB
TATA
ESSEMARA
FIGUIG
BOUJDOUR
ERRACHIDIA
OUARZAZATE
OUJDA
TAZA
ASSA-ZAG
LAAYOUNE
SAFI
FES
AZILAL
TIZNIT
SETTAT BOULMANE
TAROUDANTE
TAN-TAN
GUELMIM
NADOR
KHENIFRA
KHEMISSET
AGADIR
EL JADIDA
MARRAKECH
CHICHAOUA
KENITRA
ESSAOUIRA
BENI MELLAL
IFRANE
TAOUNATE
EL KELAA SRAGHNA
KHOURIBGA
CHEFCHAOUEN
SIDI KACEM AL HOCEIMA
LARACHE
RABAT EL HAJEB
TETOUAN
MEKNES
BEN SLIMANE
TANGER
CASABLANCA
< 30
30 < <100
100 < <250
250 < <500
Agrometeorological Cereal Yield Forecasting in Morocco
Analysis of cereal production
Balaghi R., Jlibene M., Tychon B., Eerens H.
58
level. Because of high correlations between province yields and country yield, some provinces can
be indicators of the country’s yield. Among the provinces most representative
37
of the country
average cereal yield, Settat Province comes first with an R2 of 95%, followed by El Jadida (R2=92%),
Casablanca (R2=92%), and Khouribga (R2=87%). For soft wheat, Settat comes first (R²=93%),
followed by Casablanca (R²=89%), Benslimane (R²=89%) and Taounate (R²=84%). For durum wheat,
Settat province comes first (R²=92%), followed by Casablanca (R²=89%), Benslimane (R²=86%) and
Rabat (R²=85%). For barley, Khouribga province comes first (R²=92%), followed by Azilal (R²=90%),
Safi (R²=89%) and Settat (R²=88%). Settat province which may represent the country yield for soft
wheat and durum wheat, belongs to Intermediate zone, and Khouribga province, which may
represent the country for barley yield, belongs to Unfavorable South zone. Preliminary cereal yield
forecasts at the country level can thus be done based on the yield of one province, namely Settat
for soft and durum wheat and all cereals confounded, and Khouribga for barley. Similar to yields,
the country’s production can also be forecasted based on these provinces productions.
3. TE CHN OLO GICAL TREND
Technological trend, reflected in yield improvement, is undeniable for cereals in Morocco. During
season of 1914-1915 which experienced heavy rains, Augustin (1921) noted that the yields of three
main cereals, reached: 10-12 Q/ha for durum wheat and barley, and 8-10 Q/ha for soft wheat. In
normal seasons, yields were around 7 Q/ha. At that time, the low yields of soft wheat compared to
barley and durum wheat was due to inadequacy of imported cultivars from Europe to local
conditions (Jlibene, 2009). The wheat was traditionally grown in the Saharan oases of the country.
Since 1921, the level of productivity of cereals raised significantly. For example, during the
extremely wet season of 2008-2009, country yields were 21.0 Q/ha for durum wheat, 21.6 Q/ha for
soft wheat and 17.3 Q/ha for barley. A quick comparison between the two rainy seasons of 1914-
1915 and 2008-2009, with regard to cereal yields, indicates that yields have doubled since early
1920’s, at an annual average rate of 3%.
If we consider only the post climate change period (since 1980), agricultural statistics from 1979 to
2006 (Source: Direction of Strategy and Statistics) show that there has been an average annual yield
increase at the country level of nearly 0.2 quintal per hectare (Q/ha.year) for wheat (0.19 and 0.18
for durum wheat and soft wheat, respectively), while for barley, technological trend was negative
(Table 7). This improvement was greater for wheat than for barley, despite similar investment in
agricultural research in Morocco. For barley, yields at the country level have not increased mainly
due to the low utilization rate of improved cultivars and certified seed (less than 1%), and shift of
part of the traditional barley areas to unfavorable areas for the benefit of soft wheat, following the
large scale wheat promoting operation launched in 1985.
37
Representativeness is measured by the correlation (R2) between a Province yield and country yield, using a recent
data base: series of 1999-2000 to 2010-2011.
Agrometeorological Cereal Yield Forecasting in Morocco
Analysis of cereal production
Balaghi R., Jlibene M., Tychon B., Eerens H.
59
The technological trend is not the same in all agro-ecological zones of Morocco. It is particularly
high for Favorable and Intermediate agro-ecological zones. This trend is the result of research
efforts deployed by INRA, especially in the fields of production system improvement in arid
environments, and cereal breeding cultivars with high productivity and resistance to drought and
weather associated diseases and pests.
Table 7: Yield trend over years (Quintal/ha.year), due to technology improvement of the three
main cereals, by agro-ecological zone, and coefficients of determination (R2) of the regression
lines. (Data series of 1979 till 2006; Source: Balaghi and Jlibene, 2009).
Agro-ecological Zone
Soft wheat
Durum wheat
Barley
Trend
R2
Trend
R2
Trend
R2
Favorable
0.24
0.79
0.19
0.71
-0.15
0.12
Intermediate
0.22
0.73
0.26
0.85
0.01
0.00
Unfavorable South
0.11
0.48
0.15
0.70
-0.01
0.00
Unfavorable East
0.06
0.20
0.08
0.37
-0.09
0.08
Mountain
0.17
0.60
0.13
0.70
-0.32
0.02
Saharan
0.05
0.08
0.13
0.20
-0.04
0.03
National
0.19
0.87
0.18
0.86
-0.03
0.03
Because of the high variability of cereal yields, which is due to the uncertainty of the climate
especially since 1980, the technological trend cannot be distinctly observed in the curve
representing yield evolution over years. When fitting a curve through the records of country’s
wheat yields observed during the crop seasons of 1961-1962, 1967-1968, 1985-1986, 1987-1988,
1990-1991 and 2005-2006, significant technological improvement made in favorable environments
can be observed (Table 7). We also note that wheat yields can fall down to a very low yield level in
seasons of extreme agricultural drought, around an average of 5 Q/ha at the national level. This
lower limit of country yields, which has not been raised since 1961, continues to weigh on the
country's food security. However, the fact that yields were kept at this lowest limit of 5 Q/ha
despite the decline in rainfall during the 1990s and 2000s can be seen an indication of a relative
technological improvement.
Agrometeorological Cereal Yield Forecasting in Morocco
Analysis of cereal production
Balaghi R., Jlibene M., Tychon B., Eerens H.
60
Figure 25 : Country soft wheat yield variation and trend in Morocco (in green) (Source of data:
FAOSTAT and DSS).
The blue curve on top shows trend of record yields, for the last 50 years, obtained during 1961-62,
1967-68, 1985-86, 1987-88, 1990-91 and 2005-06 seasons. The red line below shows the lower level
of yield obtained during seasons of extreme drought, which were: 1960-61, 1980-81, 1994-95, 1999-
2000 and 2006-07.
The gap between yields obtained in experimental stations (Potential situation) and those obtained
at the farmers' fields was analyzed through a modelling study done in 1994 at INRA’s Regional
Centre for provinces of Meknes, Annoceur and Douyet (Boughlala et al., 1994). This study showed
that the yield gap increases with improved production environments, from 20 Q/ha in adverse
conditions to 50 Q/ha in extremely favorable environmental conditions in average (Figure 26). One
might expect wet seasons to recover for dry seasons losses. This is not the case, since potential
production during wet seasons is not unlocked, due to premeditated farmer’s attitude in reducing
inputs for minimizing climatic risk or due to the lack of farmer’s knowledge concerning improved
technologies (improved cultivars, certified seeds, fertilization, weed control, supplemental
irrigation, etc.).
Agrometeorological Cereal Yield Forecasting in Morocco
Analysis of cereal production
Balaghi R., Jlibene M., Tychon B., Eerens H.
61
Figure 26 : Curves of cumulative probability of soft wheat yields at farm level (real situation) and
at potential situations at Meknes (Source: Boughlala et al., 1994).
In experimental stations of INRA, the annual rate of yield increase since 1980 reached 0.5
Quintal/ha.year due to genetic improvement for wheat (Jlibene, 2011), against 0.19
Quintal/ha.year at farmer level (Table 7). This difference in productivity shows how much efforts
remain to be done in terms of technology transfer. In the fields of some elite farmers that used
improved wheat varieties, record yields greater than 80 Q/ha were recorded during wet seasons.
The productivity of rainwater (i.e. ratio of yield to rainfall) at national level was also improved by
67% between the 1960s and 1970s to the 1980s and 1990s, to the order of 2.5 kg per mm (Jlibene,
2011). The average technological trend at the country level was estimated more precisely at about
0.15 Quintal/ha.year (Figure 25) instead of 0.19 Quintal/ha.year, as previously reported in this
document (Table 6). This average hides differences in improvement between contrasting seasons.
The various seasons in Morocco can be classified into three contrasted classes of productivity
corresponding to three classes of rainfall: wet (>400 mm), intermediate (300 to 400 mm) and dry
(<300 mm). When analyzing separately each of these three classes, technological trend is found to
be of more than 0.3 Quintal/ha during wet seasons, for both soft wheat (Figure 27) and durum
wheat (Figure 28). During intermediate seasons, slow trend was observed for soft wheat and
negative for durum wheat, while during dry seasons the trend was negative for all the two species.
This seeming low productivity in intermediate and dry seasons can be explained by the severity of
droughts experienced during the period of 1980 to 2008.
Agrometeorological Cereal Yield Forecasting in Morocco
Analysis of cereal production
Balaghi R., Jlibene M., Tychon B., Eerens H.
62
Figure 27 : Technology trend or improvement achieved in 28 years for soft wheat in Morocco.
Cropping seasons (series of 1979 to 2006) have been subdivided into three classes of yields, corresponding to
three classes of rainfall (<300, 300 to 400 and >400 mm/season) : dry seasons (lozenge), Intermediate
(triangle) and wet (cross). Technological trend is +0.15 Quintal/ha.year on average (green line), -0.3
Quintal/ha.year in dry seasons (brown line), nil in intermediate environments (dark green line) and +0.3
Quintal/ha.year in wet seasons (blue line). (Source: Balaghi and Jlibene, 2009).
Figure 28 : Technological trend or improvement achieved in 28 years for durum wheat in
Morocco.
Cropping seasons (series of 1979 to 2006) have been subdivided into three classes of yields, corresponding to
three classes of rainfall (<300, 300 to 400 and >400 mm/season) : dry seasons (lozenge), intermediate
(triangle) and wet (cross). Technological trend is +0.15 Quintal/ha.year on average (green line), -0.46
Quintal/ha.year in dry seasons (brown line), -0.26 Quintal/ha.year in intermediate seasons (dark green line)
and +0.35 Quintal/ha.year in wet seasons (blue line). (Source: Balaghi and Jlibene, 2009).
Agrometeorological Cereal Yield Forecasting in Morocco
Analysis of cereal production
Balaghi R., Jlibene M., Tychon B., Eerens H.
63
4. CEREAL C ROP DEVELO PMEN T
All the three cereals are sown in the fall, within a time frame of two months, starting from last
week of October to second week of December. In many parts of the cereal growing areas, sowing
starts after the first significant rains which allow germination of weeds and soften soil for seed bed
preparation.
The development of the aerial part of cereal undergoes several stages, of which the most important
ones taken as benchmarks are:
Germination takes place just after sowing, by seed imbibition, followed by swelling and
issuance of the radicle, and then of coleoptile which protects the first leaf, and breaks
through the ground. This stage requires moisture in the surface soil horizon and lasts
about a week.
Emergence is recognized by the appearance of the coleoptile, which wraps the first leaf
for protection and pushes through soil up to the surface. It takes place ten days after
sowing, but later if rains come late or if temperatures are lower than usual. The first leaf
will be followed by a second and then a third one, which are supported by a stem, called
main shoot. This stage lasts about a month.
Tillering begins when a secondary stem (or tiller) appears at the axils of the primary stem,
at the base of the eldest of the three formed leaves. The main shoot continues to issue
leaves and tillers. Tillers in turn emit secondary tillers. Tillers continue to form
(theoretically unlimited) at the basis of the oldest of the three leaves formed, but limited
by temperature increase and water deficit. All tillers are grouped near ground level,
constituting the basal tillering. Each leaf on the main shoot or a tiller is initiated from a
node. At this point the nodes are grouped. Tillering ends when the nodes begin to
separate and become discernible by touch at the base of tillers. The number of tillers
determines the potential number of spikes. The issuance of tillers lasts about one month
after emergence (three dekads).
Stem elongation is the process of elongation of internodes, which begins with the
judgment of the sheet formation and early spike initiation. During internodes elongation,
parts of the spike are formed, in particular number of spikelets, number of flowers per
spikelet and floral organs. Drafts of the spike are visible to the naked eye, since few inches
of elongation. The spike is protected by rolled leaves, eight in number generally. Other
intermediate stages can be identified as swelling that indicates the rise of the spike, or
heading that indicates the appearance of the spike outside the sheath. This phase, which
lasts more than one month, corresponds to maximum biomass accumulation. It is
sensitive to water stress. The potential number of grains is determined at this stage.
Flowering is the stage between the vegetative phase of the reproductive phase of the
crop. It takes place about a week after heading, visually recognized when stamens come
Agrometeorological Cereal Yield Forecasting in Morocco
Analysis of cereal production
Balaghi R., Jlibene M., Tychon B., Eerens H.
64
out. Stamens, three in number per flower, are expelled outside the spike, once the pollen
they contain is released to fertilize the stigma that will lead the genetic material of pollen
towards the ovary, in order to the fertilize it. The exit of stamens indicates that
fertilization took place, and also eventually fertilization.
Grain filling begins immediately after fertilization, the egg develops into the seed whose
development goes through benchmark stages: the milk stage and dough stage. During
dough stage, stalk of the spike turns yellow, indicating physiological maturity. This phase
determines the weight of the grain, and lasts one month and a half, approximately.
Final grain yield is the product of: (1) number of spikes per square meter, determined during
tillering, (2) number of grains per spike, determined during elongation, and (3) grain weight,
determined during grain filling.
Agrometeorological Cereal Yield Forecasting in Morocco
Agrometeorological analyses
Balaghi R., Jlibene M., Tychon B., Eerens H.
65
V. AGROMETEOROLOGICAL ANALYZES
Climate and cereal production analyzes revealed a strong dependency of cereal yields on weather. In
this chapter, the relationship between weather and cereal yields is studied in order to explore the
agrometeorological indices that can be used to forecast yields. Evapotranspiration, which is a
function of rainfall and temperature, is one of the variables that best explain cereal yields on fine
scales, as it is directly related to the cereal production of biomass. On wider scales, rainfall used
alone explains a substantial part of the inter-annual variability of cereal yields, when cumulated
over the cropping season from September till May. At national level, this dependency is valid for
cumulated rainfall below 378 mm, a threshold which is slightly above the long term rainfall average.
Rainfall above this amount is not necessarily associated with higher yields, leading to introduction
of the concept of water productivity (i.e. the ratio of yield per unit of rainfall). Improvement of water
productivity over the seasons was correlated with improvement of technological trend achieved on
wheat. Water productivity is significant in irrigated Saharan zones, followed by Intermediate and
Unfavorable zones. In high rainfall areas or in rainy seasons, water productivity is relatively low, due
to inefficiency of farmers in taking advantage of heavy rainfall. Drought events during the cropping
season also impact cereal yields. NDVI taken as an environmental predictor index is positively
correlated with both rainfall and cereal yields. It is positively and linearly correlated with rainfall
below 550 mm/cropping season. NDVI is a powerful cereal yield predictor, when averaged from
February till March, or from February till April.
1. RELAT IONSHIP BETWEEN TRANSPIRATION AND GRO W TH
In the mid-1950s, De Wit (van Keulen and van Laar, 1986) was among the first authors to recognize
the direct relationship between transpiration
38
and plant productivity. This relationship was later
used by FAO to develop models of crop response to evapotranspiration (Doorenbos and Kassam,
1979). Actual evapotranspiration
39
(ETa) of plants is one of the variables that best explain
agricultural productivity, as it is directly related to the production of biomass. The relationship is
linear, at various spatial scales (leaf, plant and region), and reconfirmed by many authors, in
particular in Settat province of Morocco (Bazza and Tayaa, 1998).
Transpiration may be low, due to a shortage of water within the root zone (drought), or due to low
amount of solar energy received (cold). One of the basic principles of agrometeorology is that plant
growth (biomass accumulation) is governed by the incident energy which has to be released by
plants through transpiration.
In Morocco, the relationship between evaporation and temperature has been demonstrated a long
time ago (Loup, 1957). Temperature is the factor that most determines evaporation, either average
38
Water evaporated from plant tissues. It depends on the energy received, the vapor pressure gradient and wind speed
(Allen et al. 1998).
39
Sum of plant transpiration and water evaporated from soils, in limiting water supply.
Agrometeorological Cereal Yield Forecasting in Morocco
Agrometeorological analyses
Balaghi R., Jlibene M., Tychon B., Eerens H.
66
or maximum temperature. Temperature determines evapotranspiration as well, which combines
both air evaporation and plant transpiration of water from the soil (Figure 29).
Figure 29: Evolution of daily actual evapotranspiration of soft wheat, simulated on the basis of
temperature alone (Eta_T°) and of temperature in association with relative humidity (RH), wind
speed and global radiation, at Meknes (Balaghi, 2000).
The approach was then updated by FAO, developing "AQUACROP" tool
(http://www.fao.org/nr/water/aquacrop.html), which can simulate crop productivity by
disaggregating field crop evapotranspiration (ETa), into soil evaporation and crop canopy
transpiration (Hsiao et al., 2009; Raes et al., 2009; Steduto et al., 2009).
The relationship between transpiration and biomass is shown schematically in Figure 30 (Jlibene
and Balaghi, 2009). The transpired water, which is essential for the metabolism, is transported from
the soil to the atmosphere passing through plants. The amount of water transpired is proportional
to the amount of water available in the soil. The latter results from the existing residual moisture
within the soil, in addition to rainfall and irrigation, minus losses, due to runoff, drainage, soil
evaporation and weeds transpiration. The soil acts as a buffer, keeping the humidity between two
water supplies.
Agrometeorological Cereal Yield Forecasting in Morocco
Agrometeorological analyses
Balaghi R., Jlibene M., Tychon B., Eerens H.
67
Figure 30: Schematic figure illustrating the relationship between transpiration and biomass
production (Jlibene and Balaghi, 2009).
The direct action of these factors on biomass (Figure 30) has been experimentally demonstrated in
a variety of agronomic studies. It has become a fact that rainfall, irrigation, weed control, mulching
and soil moisture increase crop yields.
1.1. RELATIONSHIP BETWE EN RAINFALL AND CEREAL YIELDS
Rainfall is the factor that induces the major proportion
40
of cereal yields' variation in Morocco. The
relationship between rainfall and cereal yields has, from long time and on several occasions, been
demonstrated in Morocco (Papy, 1979; Douguedroit and Messaoudi, 1998; Douguedroit et al.,
1998). Cereals yields (and thus production) vary with rainfall, depending on the cropping season. In
contrast, the area of the three cereals (soft wheat, durum wheat and barley) is relatively constant
(5.3 million hectares), with low inter-annual variation (CV = 6%, between seasons of 1999-2000 and
2010-2011).
In general, cereal yields and cereal production evolve in the same direction as the rainfall. An
40
The relative importance of other factors that could explain cereal yields' variation depends on the spatial scale
considered. For instance, agro-technical factors and soil types are relevant to consider only at fine spatial scales, for
explaining cereal yields. Unfortunately, in Morocco as in most developing countries, these data are often unavailable.
Agrometeorological Cereal Yield Forecasting in Morocco
Agrometeorological analyses
Balaghi R., Jlibene M., Tychon B., Eerens H.
68
increase in rainfall, relative to previous season, is associated with an increase in yield and in
production. Vice-versa, a decrease of rainfall is associated with a decrease in yields and in
production (Figure 31). The coefficient of variation of rainfall is equal to 32%, while it is 37% for soft
wheat, 41% for durum wheat and 47% for barley, over the period 1988-2011. There is a strong
positive correlation (R2> 65%) between cumulated rainfall over the cropping season and cereal
yields. However, there are some exceptions to this rule. During both 1989-1990 and 1999-2000
cropping seasons, rainfall increased while yields decreased. In contrast, during the cropping season
of 1997-1998, yields had increased despite a drop in rainfall compared to average. These
exceptions demonstrate the importance of rainfall distribution within the cropping season.
Figure 31: Deviation (%) from mean values of cumulated rainfall over September to May period,
and main cereal yield (soft wheat, durum wheat and barley), at the country level (data from 1988
to 2011); indicating the co-variation of rainfall and cereal yields.
1.2. RELATIONSHIP BETWE EN RAINFALL AND CEREAL AREA
The total area of cereals has steadily increased since 1980, at an average rate of 39,600
hectares/season. This increase was mainly the result of increased soft wheat area, boosted by the
large scale state promotion of this crop (in French, Opération de promotion du blé tendre),
launched in 1985-1986 (Jlibene, 2011). This operation was designed to double the cultivated area of
soft wheat, from 0.5 million hectares to 1.0 million hectares, using as incentives: the distribution of
Agrometeorological Cereal Yield Forecasting in Morocco
Agrometeorological analyses
Balaghi R., Jlibene M., Tychon B., Eerens H.
69
new improved varieties and fixed prices for harvested grain for producers, and fixed marketing
margins for industrials (Ait El Mekki, 2006). The area of soft wheat which had stagnated at around
half a million hectares for over thirty years, until the early 1980s, has been doubled in 1986, and
multiplied by four, ten years later. The area stabilized then at around two million hectares,
occupying more than a fifth (1.9 million hectares, from 1998 to 2011) of the agricultural area of the
country (8.7 million hectares). Over the period 1998-2011, the area of soft wheat has increased at
an average rate of 34,900 hectares/season, accounting for 88% of total cereal area increase. The
area of durum wheat and barley declined at a rate of 11,600 and 13,800 hectares/season,
respectively (Figure 32).
Figure 32: Deviation (%) from mean value of cumulated rainfall over September to May period,
and of main cereal area (soft wheat, durum wheat and barley), at the country level (data from
1988 to 2011).
In Morocco, cereal area is also influenced by rainfall, especially rainfall of start of the season, from
the first significant rains
41
till around November - early December (Figure 33). These first significant
rains were estimated at a minimum amount of 25 mm, cumulated over a period of 10 days, in semi-
arid regions of Morocco (Watts and El Mourid, 1988). The arrival date of these rains, which is highly
inconsistent from one year to another, determines the sowing date for cereals in arid and semi-arid
41
The first significant rains are defined as the amount of rainfall that allows germination and emergence.
Agrometeorological Cereal Yield Forecasting in Morocco
Agrometeorological analyses
Balaghi R., Jlibene M., Tychon B., Eerens H.
70
environments (Benaouda and Bouaziz, 1992). The cropped area is larger when the rains come early
and heavy in the season, as most farmers sow after these first rains. When rainfall is abundant at
the beginning of the season, even arid areas of southern Morocco are sown, as for example in the
region of Marrakech, Essaouira and Kelâa Sraghna.
Rainfall of the months of November till December best explains the cropped cereal area, especially
in the case of barley, then for soft wheat and durum wheat. However, during the last cropping
seasons, sowing tend to start from the first week of November, even in the absence of rain.
Rainfall of the months of November till December best explains the cropped area of cereal,
especially in the case of barley, then for soft wheat and durum wheat. However, during the last
cropping seasons, sowing tend to start from the first week of November, even in rain deficiency.
Figure 33: Coefficient of determination (R2) of the relationship between main winter cereal area
(soft wheat, durum wheat and barley) and cumulated rainfall during the cropping season
(starting in October in average), at the country level (Data series from 1988 to 2011). Average
dekadal rainfall is plotted in second axis (blue bars), for illustration.
Agrometeorological Cereal Yield Forecasting in Morocco
Agrometeorological analyses
Balaghi R., Jlibene M., Tychon B., Eerens H.
71
1.3. RELATIONSHIP BETWE EN RAINFALL AND RAINWATER PRODUCTIVITY
1.3.1. AT NATIONAL SCALE
When cereal yield is regressed against cumulated rainfall during the cropping season, the slope of
the regression line represents Rain Water Productivity
42
(RWP). At national level and for the three
winter cereal combined, the average RWP translated in grams/liter rainwater (g/l), is in average
0.332 g/l, for the period 1988-2011. RWP varies depending on the cropping season, from a
maximum of 0.506 g/l reached in 1993-1994 to a minimum of 0.149 g/l in 1999-2000. RWP is on
average higher for soft wheat (0.404 g/l) than for durum wheat (0.370 g/l) or for barley (0.261 g/l).
Grain yield of three winter cereal (soft wheat, durum wheat and barley), increases with the
cumulated rainfall from September to April, up to the limit of 378 mm slightly above long term
average (Figure 34, Figure 35 and Figure 36). Beyond this limit, yields do not increase, indicating an
undervaluation of wet seasons, which can be explained by the adaptation of cropping practices to
slightly above average rainfall conditions in Morocco.
Figure 34: Relationship between soft wheat country yield (Kg/ha) and cumulated rainfall during
the cropping season (mm) (data from 1988 to 2011).
Slope of the regression line (in blue) represents Rain Water Productivity of soft wheat in Morocco, which is
0.309 g/l. The slope of the regression line above (in red) denotes maximum Rain Water Productivity
manageable in optimum conditions (0.652 g/l).
42
Rainwater Productivity (g/l) is defined as the ratio of yield (kg/ha) on the cumulated rainfall (mm) during the cropping
season (September to April).
Agrometeorological Cereal Yield Forecasting in Morocco
Agrometeorological analyses
Balaghi R., Jlibene M., Tychon B., Eerens H.
72
Figure 35 : Relationship between durum wheat country yield (Kg/ha) and rainfall during the
cropping season (mm) (data from 1988 to 2011).
Slope of the regression line (in blue) represents Rain Water Productivity of soft wheat in Morocco, which is
0.347 g/l. The slope of the regression line above (in red) denotes maximum Rainwater Productivity
manageable in optimum conditions (0.653 g/l).
Figure 36: Relationship between barley country yield (Kg/ha) and rainfall during the cropping
season (mm) (data from 1988 to 2011).
Slope of the regression line (in blue) represents Rain Water Productivity of soft wheat in Morocco, which is
0.277 g/l. The slope of the regression line above (in red) denotes maximum Rainwater Productivity
manageable in optimum conditions (0.498 g/l).
Agrometeorological Cereal Yield Forecasting in Morocco
Agrometeorological analyses
Balaghi R., Jlibene M., Tychon B., Eerens H.
73
The valuation of wet seasons requires more farmers' input investments, mainly concerning selected
seeds, nitrogen and pesticides. Agriculture is also "the art of managing uncertainty" (Faurès et al.,
2010), as it aims at minimizing production risks related to the variability and unpredictability of
weather or of market. In unreliable environments, strategies for risk management are hardly
compatible with the objective of maximizing crop yields, which involves larger investments and
therefore higher risk-taking. The “policy of dams" launched in the early 1960s in Morocco, aimed at
reducing climatic risk and in the same time ensuring water supply to cities (hydraulic potential of 21
billion m3). Currently, irrigated lands account for 17% (1.46 million hectares) of agricultural lands, of
which 76% are still under surface irrigation (MAPM, 2011).
On average, durum wheat (Figure 35) seems to better use rainwater with a RWP of 0.346 g/l,
followed by soft wheat (0.309 g/l) (Figure 34) and barley (0.277 g/l) (Figure 36). The apparent
advantage of durum wheat in RWP is probably related to the fact that durum wheat is mainly
grown in semi-arid areas, unlike soft wheat which is relatively more cultivated in favorable (less
valued) areas (Table 6).
Potential RWP, expressed by the slope of the regression line between the potential cereal yield and
rainfall indicates that soft and durum wheat better value rainwater (0.652 g/l), unlike barley (0.498
g/l) (Table 8). Potential RWP is reached in optimal production conditions. It is almost double
average RWP, highlighting shortfall of agriculture concerning the rainwater valuation, in particular,
and management of natural hazards in general.
Theoretically, minimum rainfall required for grain production is 70 mm for soft wheat, 92 mm for
durum wheat and 100 mm for barley (Figure 36). For the three species, this rainfall level is very low,
indicating drought management improvement in Morocco. The difference between this minimum
rainfall amount, for these three species is in favor of soft wheat, which is relatively a rustic species
in Morocco, compared to durum wheat and barley.
Table 8 : Average and Potential Rain Water Productivity (RWP) of cereals, at national level. RWP
is equal to the slope of the relationship between yield and rainfall.
Average RWP is computed using all yield data, while Potential RWP is computed using only the
maximum yields observed for each level of rainfall.
Species
Rainwater Productivity (gram/liter)
Average
Potential
Soft wheat
0,309 ; R² = 49,6%
0,652 ; R² = 93,3%
Durum wheat
0,346 ; R² = 59,7%
0,653 ; R² = 94,6%
Barley
0,277 ; R² = 58,1%
0,498 ; R² = 84,6%
Agrometeorological Cereal Yield Forecasting in Morocco
Agrometeorological analyses
Balaghi R., Jlibene M., Tychon B., Eerens H.
74
1.3.2. AT AGRO-ECOLOGICAL ZONE SCALE
For the three cereals together, RWP increases from Favorable agro-ecological zone (0.281 g/l),
Unfavorable South (0.321 g/l), Mountain (0.333 g/l), Unfavorable East (0.334 g/l), Intermediate
(0.340 g/l) to the Saharan zone (0.548 g/l) (Table 9). In Saharan zone, high RWP is due systematic
irrigation. From these three species, soft wheat has the highest RWP in Favorable, Mountain and
Unfavorable East zones. Durum wheat has the highest RWP in Unfavorable South, Intermediate,
and more significantly, in Saharan zone. Barley has the lowest RWP in the six agro-ecological zones.
Table 9: Rain Water Productivity (gram/liter), for the three winter cereals, at the national and
agro-ecological zones levels (average from 1988 to 2011).
Rainwater Productivity (gram/liter)
Rainfall (mm)
Soft wheat
Durum wheat
Barley
Three cereals
Favorable
524.9
0.316
0.271
0.232
0.281
Unfavorable South
243.4
0.349
0.408
0.292
0.321
Mountain
436.4
0.409
0.314
0.253
0.333
Unfavorable East
273.3
0.349
0.297
0.333
0.334
Intermediate
350.3
0.360
0.393
0.293
0.340
Saharan
190.1
0.763
1.196
0.393
0.548
National
346.6
0.404
0.370
0.261
0.332
2. AG RIC ULT URA L DROUGHT
Cumulated rainfall during the cropping season can be an indicator of drought. Already in 1921 in
Morocco, Augustin (Augustin, 1921) stated: "It is the amount and distribution of rainfall that
indicate which areas are suitable for cropping and for what crops".
The number of dry seasons, with rainfall less than 400 mm, increased since climate change
observed in the early 1980s in Morocco. Frequency of dry seasons has increased fivefold, from one
dry season every 15 normal seasons, during the 1930s, 1940s, 1950s, 1960s and 1970s, to one dry
season every three normal seasons, in the last three decades (Jlibene, 2011).
The accumulation of rainfall during the cropping season can be interrupted, thereby informing on
drought. Interruptions can occur on one or many time periods, in the beginning, middle or end of
the season. Examples of these interruptions are given in Figure 37.
During season of 1994-1995, there has been no rain for almost three months (November,
December and January), affecting grain and straw cereal production ;
Season of 2004-2005 was almost identical to that of 1994-1995 (driest historical season),
in terms of total rainfall, but not in terms of its distribution. During season of 2004-2005,
dry spells during January and three consecutive months in March, April and May, were
recorded ;
Agrometeorological Cereal Yield Forecasting in Morocco
Agrometeorological analyses
Balaghi R., Jlibene M., Tychon B., Eerens H.
75
During season of 1999-2000, 180 mm were recorded till January, then a drought settled
for three consecutive months ;
Season of 1996-1997 was rainy till mid-January, followed-on by two consecutive dry
months ;
During season of 2000-2001, 300 mm were recorded till late February, and then rains
stopped during the following two consecutive months of March and April.
Figure 37 : Dekadal cumulated rainfall between September and May, of dry cropping seasons.
Periods of drought are recognized as flat segments of the curves (in bold). Drought can occur at
diverse periods of the crop cycle.
Figure 38 illustrates the similarity of rainfall distribution during the cropping season, covering many
weather stations distributed all over the country. In this figure, the flat zones of cumulated rainfall
curves indicate drought spells. Drought is hence generalized all over the country, whenever it
occurs. For example, during the dry season of 1996-1997, there was almost no rainfall, in 23
provinces of the country taken as wide examples, during the first ten days of January and the last
dekad of March (Jlibene, 2011). This situation systematically occurs whenever there is drought.
Drought is thus due to low rainfall, inadequate intra-annual distribution of rainfall or the two
reasons at once.
Drought can be explained by a deviation of rainfall compared to long term average, in terms of total
amount or distribution. The idea behind this is that crops have adapted over the centuries to a
usual rainfall distribution, and are thus affected when this normal distribution is disturbed.
Agrometeorological Cereal Yield Forecasting in Morocco
Agrometeorological analyses
Balaghi R., Jlibene M., Tychon B., Eerens H.
76
Agricultural drought occurs when the amount of rainfall and soil water reserve are not sufficient to
ensure crop water needs. In conclusion, drought is a water deficiency, sufficiently prolonged to
cause a negative effect on crop. In this definition, there are two important concepts: sufficiently
prolonged water deficit and the negative effect on crop. The effect on the crop depends on its
sensitivity to water deficit, also known as vulnerability or tolerance.
Figure 38: Spatial distribution of cumulated dekadal rainfall (from September till May) for the dry
season of 1996-1997. Locations used cover 23 provinces of Morocco, scattered all over the
country (Source: Jlibene, 2011).
3. RELATION SHIP BETWEEN GROWTH A ND RAINFALL
The normal growth cycle of wheat in Morocco, in relation with the temperature and precipitation,
is shown in Figure 39. Crop growth has adapted, over long years, to intra-annual distribution of
rainfall. Rainfall peak occurs in autumn and winter, which allow water reserves in the soil and thus
installation of cereal crops. Sowing is generally carried out between the months of September and
December, depending on how early the first significant rains of autumn come. The harvests begin in
the month of May in southern areas of Morocco, where temperatures rise first, and continue until
June in the northern areas.
Agrometeorological Cereal Yield Forecasting in Morocco
Agrometeorological analyses
Balaghi R., Jlibene M., Tychon B., Eerens H.
77
Figure 39: Typical weather conditions, during wheat growing season. Long term average of
dekadal rainfall (series from 1988 to 2011) and dekadal temperature (series from 1998 to 2009)
were used. Major development phases of wheat crop are also displayed (green arrow).
4. RELATION SHIP BETWEEN NDVI AND RA INF ALL
In Morocco, rainfall varies to a great extent, from south to north and from east to west. Cropping
season's rainfall (September till April) has been related to NDVI (derived from NOAA-AVHRR) of
agricultural lands (using a crop mask), for a total of 23 stations / provinces of the country, using
time series data from 1990 till 2004 (Figure 40). Average dekadal NDVI, over the 9 dekads from
February till April, is highly correlated to cumulated rainfall over the cropping season (September
till April) in Morocco, as country's climate is mainly semi-arid, and as most of agricultural lands are
covered by cereals (5.3 out of 8.7 million hectares).
In Figure 40, is illustrated the strong relationship between these two variables, which is linear for
rainfall values less than 550 mm (blue line). Beyond this limit, the relationship is no longer
consistent. Similarly, the linear relationship was indeed found, in the range 200-600 mm, in African
semi-arid regions (Martiny et al., 2006). It can be observed that the relationship gradually slacks off
as rainfall increases. Variation of NDVI, for a given level of season's rainfall, is most likely due to
intra-annual distribution of rainfall.
Agrometeorological Cereal Yield Forecasting in Morocco
Agrometeorological analyses
Balaghi R., Jlibene M., Tychon B., Eerens H.
78
NDVI can thus be used as indicator for forecasting cereal yields, using a statistical linear regression
methodology, for the majority of Moroccan agricultural lands, and for a wide range of cropping
seasons. Also, from Figure 40, one can thus explain why, in most European countries where
generally rainfall exceeds 600 mm, NDVI is useless for crop forecasting.
Figure 40: Relationship between NDVI (NOAA-AVHRR), averaged over the period from February
till April and over agricultural zones, and cumulated rainfall over the cropping season from
September till April, in Morocco. The 345 dots in this figure are data from 23 weather stations,
over the period from 1990 to 2004.
NDVI can be used to monitor vegetation status, during the cropping season, at dekadal time step. In
Figure 41 are presented NDVI maps, of the 3rd dekad of March, for two contrasted seasons, 1999-
2000 (dry: 264 mm) and 2005-2006 (wet: 378 mm). Differences between these two seasons NDVI
maps are obvious. In 2006, high NDVI values (shown in dark green in the figure) cover almost all
croplands. NDVI is relatively high, even in the Saharan and eastern provinces of the country. In 2000
instead, high NDVI values appear only in limited areas, around few coastal and mountainous areas
and in the northwest of the country. One can also notice the foehn
43
effect, encountered on the
Atlas Mountains, which creates aridity in the eastern areas of the country.
43
The foehn (or Föhn), from Swiss dialect, is a wind that results from the warming of air mass that has lost its moisture
on Mountain slopes, creating thus an arid region on the leeward side of the Mountain.
Agrometeorological Cereal Yield Forecasting in Morocco
Agrometeorological analyses
Balaghi R., Jlibene M., Tychon B., Eerens H.
79
NDVI of the 3rd dekad of March 2006
NDVI of the 3rd dekad of March 2000
Figure 41: Comparison of the vegetation status in Morocco, between a wet season (2005-2006)
and a dry season (1999-2000), using NDVI (SPOT-VEGETATION) of the 3rd dekad of March. Dark
green colored pixels indicate high NDVI values.
5. NDVI PRO FILE
5.1. INTRA AND INTER-ANNUAL VARIATION OF NDVI
NDVI profile during the cropping season (average on all agricultural lands of Morocco) shows a
gradually increase of NDVI, driven by early rains of November, a peak around the first dekad of
March and then a decline (Figure 42). NDVI profile can reflect the evolution of cereal growth, since
cereals are grown on most of croplands in Morocco (5.3 out of 8.7 million hectares). The vegetative
cover of cereals is in fact low in November (planting time), with a peak in March (flowering period)
and a decline thereafter (maturation period).
Starting from December, cropping seasons are easily recognizable by the NDVI profile, over a given
geographical area which may vary from one plot to the whole country. Through NDVI profile, the
specific characteristics of each cropping season can be described, particularly those which suffered
from drought spells and those which were wet, such as (Figure 42):
Cropping season of 1999-2000, which experienced a severe drought in the middle of the
cycle, during which NDVI fell very early in the season, in February ;
Cropping season of 2001-2002, during which the rainy season was delayed, displaying a late
NDVI growth, in January ;
Cropping season of 2006-2007, which experienced drought throughout almost all the
season, displaying very low levels of NDVI ;
Cropping season of 2008-2009, during which rains were abundant and well distributed,
Agrometeorological Cereal Yield Forecasting in Morocco
Agrometeorological analyses
Balaghi R., Jlibene M., Tychon B., Eerens H.
80
displaying a fast growing NDVI, with record values in spring, and began to decrease only late
March.
Figure 42: NDVI profile, during the cropping season for all croplands of the country (SPOT-
VEGETATION) and for 12 cropping seasons (1999-2000 to 2010-2011), arranged in descending
order of observed cereal yields.
5.2. REGIONAL VARIATION OF NDVI
NDVI profile (average on all croplands), during the cropping season, is shown in Figure 43, for each
of the six agro-ecological zones of Morocco. On average, from 1999-2000 to 2010-2011, NDVI
varied in a ratio of 1: 2.5: 2, between the months of September, March and April. NDVI differences
are more pronounced between agro-ecological zones, around vegetation peak, i.e. around the
month of March. In average, all over the season, NDVI values are higher in Favorable, Intermediate
and Mountain zones than in Unfavorable South, Unfavorable East and Saharan zones, consistent
with their respective usual spatial rainfall distribution. During the months of September and
October, NDVI is relatively low and stable throughout the country, due to little rainfall, or because
vegetation is still at its starting phase. During the start of the season, NDVI is however higher in
Favorable, Mountain and Saharan zones, average in Intermediate zone and low in Unfavorable
South and Unfavorable East zones. In the particular case of the Saharan zone, NDVI is relatively high
early in the season, due to the systematic use of irrigation.
Agrometeorological Cereal Yield Forecasting in Morocco
Agrometeorological analyses
Balaghi R., Jlibene M., Tychon B., Eerens H.
81
Figure 43: Average NDVI profile (SPOT-VEGETATION), during the cropping season and over all
agricultural lands, for the six agro-ecological zones of Morocco (Data from 1999-2000 to 2010-
2011).
NDVI increases gradually from November and reaches a maximum, between the 3rd dekad of
February and the 3rd dekad of March, depending on the zone, and then decreases. Maximum NDVI
is reached earlier in the season, during the 3rd dekad of February, in the Intermediate and
Unfavorable South zones, which are semi-arid, then during the 2nd dekad of March, in the Favorable
zone (sub-humid) and Mountain zone (wet), and finally during the 3rd dekad of March, in
Unfavorable East and Saharan zones. These differences are due to differences in the length of the
growing cycle, due to water availability in irrigated areas, to higher rainfall in the north than in the
south, and to lower temperatures in mountains than in plains. The increase rate in NDVI is relatively
fast in Favorable and Intermediate zones and slow in the Saharan zone.
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
82
VI. CEREAL YIELD FORECASTING APPROACHES
Cereal yields can be forecasted based on combined approaches, which relies on a concurrent use of
four approaches and various types of predictors (rainfall, temperature, NDVI)
The non-parametric approach is used for qualitative description of the effect of weather on cereals.
Cereal yields can be forecasted, according to whether drought spells happen in the beginning,
middle or end of the cycle, in relation with critical growth and development phases of the crops.
The similarity analysis approach consists in identifying, among the long term historical weather
data, the cropping seasons that are similar (or analog) to the current one, in terms of rainfall,
temperature and NDVI, selected as predictors. This type of analysis assumes that similar
environmental conditions would result in similar yields, considering that predictors used, as proxy of
the environment, account for most of yield variation. Yields of similar cropping seasons are then
averaged and adjusted for technological trend. A forecasting error can be estimated and associated
with the forecast, when many similar seasons are considered.
The regression models approach relies on simple or multiple linear regression analyses, between
yields and some climatic or bio-climatic indicators, for forecasting cereal yields. At both national and
agro-ecological zones levels, the correlations are high when considering average dekadal rainfall,
cumulated over the cropping season, from October till late March. Forecasts can further be
improved, by splitting season's rainfall into several non-overlapping time periods: Two periods, from
September till November and from January till March; or three periods, from October till November,
from December till January, and from February till March. Cereal forecasts are obviously more
accurate, when considering three time periods' rainfall, rather than two or only one. In addition to
rainfall, bio-climatic indicators, from FAO's AgroMetShell water balance model, can be used as
predictors. However, correlations are higher when using average dekadal NDVI, from February till
March, peaking at an R2 of nearly 90%. Although forecasts can begin as early as February,
forecasting error is progressively improving as approaching the end of the season. The regression
model approach leads to accurate cereal yield forecasts, even at provincial level. Several indicators
can be combined to improve the forecasts, such as NDVI, rainfall and temperature. Among these
indicators, NDVI best explains yields variation, followed by rainfall and temperature. Temperature,
which always has a negative impact on yields, since it increases evapotranspiration, is relevant only
in high rainfall areas, like in provinces located in the far north-west of the country, such as Tangier
and Tetouan.
Simulation models approach, which simulates the behavior of crop depending on weather variables
and crop and soil parameters.
The combined approach relies on the concomitant use of non-parametric, similarity, regression
models and simulation approaches. It allows minimizing the forecasting error of the forecasts.
Operational cereal forecasts were performed using this approach, for the cropping season of 2008-
2009, 2009-2010, 2010-2011 and 2011-2012, by INRA in partnership with national and international
institutions.
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
83
1. NON-PARAMETRIC APPROACH
The non-parametric statistical approach (Gommes, 2006) is of descriptive interest, allowing for a
qualitative assessment of the effect of weather on crops. It is particularly useful for the evaluation
of qualitative and indirect effects of weather on crops. One of the reasons why the non-parametric
approach can be very powerful is that climatic variables do not vary independently but constitute a
"complex", like the case of drought. This approach requires prior expertise on the environment and
the main factors that affect agricultural production, and their inter-correlations. Wheat yield at the
national level can be forecasted, using non-parametric statistical approach, i.e. using indicators of
season's environment. The season can be divided in three phases of the cropping cycle, which are
known to be critical for growth and development of cereals: The beginning, middle and end of
season, which correspond approximately to the months October-November-December, January-
February-March, and April-May.
Table 10 : Non-parametric relationship between wheat grain yield (Q/ha) and drought indicators,
during critical season’s phases (beginning, middle and end of cycle). (0: No drought; 1:Drought).
Season
Start
Middle
End
Rainfall (mm)
Yield (Q/ha)
1987-1988
0
0
0
387.7
18.60
1988-1989
0
0
0
350.9
14.80
1989-1990
0
1
0
384.9
13.59
1990-1991
0
0
1
392.4
19.50
1991-1992
1
1
0
279.4
7.72
1992-1993
0
0
0
233.9
8.00
1993-1994
0
0
0
343.3
20.06
1994-1995
1
1
0
196.0
5.68
1995-1996
0
0
0
587.2
18.57
1996-1997
0
1
0
465.4
9.43
1997-1998
0
1
0
377.4
14.46
1998-1999
0
0
1
226.9
8.39
1999-2000
0
1
0
263.6
5.23
2000-2001
0
0
1
320.6
13.21
2001-2002
0
1
0
332.2
13.33
2002-2003
0
0
0
411.3
17.83
2003-2004
0
1
0
419.2
18.00
2004-2005
0
1
1
228.4
11.03
2005-2006
0
0
0
377.5
20.76
2006-2007
0
1
0
237.1
6.18
2007-2008
0
1
0
268.1
13.11
2008-2009
0
0
0
547.9
21.60
2009-2010
0
0
0
604.8
16.62
2010-2011
0
0
0
492.9
18.60
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
84
Drought in Morocco usually occurs at the end of the cycle, but it can occur during the other phases
as well. Experience shows that drought has a dissimilar impact on wheat yields according to
whether it occurs at the beginning, middle or end of the cycle (Table 10). This approach obviously
doesn't take into account the intensity of drought, as parametric approach does.
From Table 10, a multiple regression analysis can be performed using four predictors: three non-
parametric variables (0/1 drought in beginning, middle and end of the cycle) and the cumulated
rainfall from September till May. After removal of predictors that do not significantly explain yield
variance, through "Stepwise" regression analysis procedure, it results the following equation:
Yield (Q/ha) =6,38 – 3,62 Middle (-) + 0,025 Rainfall (mm) R2=61% (Pr. <0.001)
This equation explains significantly a high proportion (R2 = 61%) of wheat yield variance. Aside from
the fact that yield depends on the rainfall (partial R2 = 50%), this equation shows that it is mainly
affected by mid-cycle drought (partial R2 = 11%) comparatively to the other two drought periods
occurring during wheat cycle, which is in accordance with observations from field experiments.
Through non-parametric statistical approach, wheat yields can therefore be forecasted in Morocco
with an average error of 3.27 Q/ha, nationwide.
2. SIMILARI TY APPROAC H
2.1. SIMILARITY ANALYSIS USING RAINFALL
2.1.1. CUMULATED RAINFALL OVER THE SEASON
The easiest and fastest way for forecasting cereal yields consists in performing a similarity analysis,
using cumulated rainfall over the cropping season. For example, a simple graphical comparison of
historical cropping seasons (from 1987-1988 till 2011-2012), based on cumulated rainfall from
October till April, shows that cropping season of 2011-2012 is similar to those of 1999-2000 and
2004-2005, which also both experienced severe drought (Figure 44). This approach is obviously very
approximate as it ignores the intra-annual distribution of rainfall. However, cropping seasons can
be qualitatively assessed, using this approach. From official statistics, cereal yields of the cropping
season of 2011-2012 were lower than average, with a national yield of 10.1 Q/ha for the three
cereals combined (against 11.9 Q/ha, on average for period 2000-2011), due to mid-cycle drought
(Balaghi et al., 2012).
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
85
Figure 44 : Cropping seasons, from 1987-1988 to 2011-2012, arranged in ascending order of
cumulated rainfall between 1st of September and 10th of April of the following calendar year
(Source: Balaghi et al., 2012). Numbers in x-axis correspond to years of harvest. Green line
indicates average cumulated rainfall over the 25 years. The red bar points to cumulated rainfall of
the latest season i.e. 2011-2012.
2.1.2. INTRA-ANNUAL RAINFALL DISTRIBUTION
In Figure 45, cropping seasons of 1960 till 2000 were classified based on cluster analysis, using
monthly rainfall of the eight months from September till April (Balaghi, 2000). Cluster analysis
technique consists in identifying homogeneous groups (clusters) which are relatively more similar
to each other than to those in other groups
44
, in terms of monthly rainfall for the eight months
considered (September till April). In this case, no specific weights have been assigned to the
different months, i.e. rainfall of each of the eight months are supposed to have same impact on
crops. Cluster analysis suggested a strong similarity between seasons of 1999-2000 and 1994-1995,
which in fact were both very dry. The cluster analysis also shows two distinct groups, which can
typically be representative, on one hand of good seasons, and on the other hand of dry seasons.
The first group contains in particular seasons of 1970-1971 and 1961-1962, and the second group
seasons of 1999-2000 and 1974-1975. Season of 1984-1985, which resulted in average country
44
The distance between two groups is given either by the distance between the most distant objects in the groups
(method of "complete linkage") or by minimizing the intra-group variance (Ward's method).
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
86
yield, was falsely assigned by cluster analysis to dry seasons, like 1980-1981, because low
temperatures observed during season of 1984-1985, were not taken into account.
So, we can see that this approach requires a good knowledge of the climatic conditions that
occurred during the different cropping seasons. Cluster analysis can thus only be used as
supplementary technique to other methodologies for forecasting cereal yields.
Figure 45: Dendrogram of cluster analysis, discriminating cropping seasons from 1960-1961 to
1999-2000, with regard to monthly rainfall of 8 months, from September to April, for Meknes
region (Source: Balaghi, 2000).
Similarity analysis, based on rainfall distribution over the cropping season, can be used to
accurately forecast cereal yields. This methodology was used for the first time in January 1995 at
the request of the Provincial Direction of Agriculture (Direction Provinciale de l'Agriculture) of
Meknes, following persistent drought which prevailed for more than two consecutive months
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
87
(November and December). This Direction asked INRA to provide with plausible scenarios of the
evolution of the cropping season of 1994-1995. This season (in red in the graph in Figure 46) was
not similar to none of the 35 previous seasons, since 1961, in terms of rainfall distribution over the
first four months of the season (September till December). From the historical time series, only the
cropping season of 1974-1975 has experienced more or less a similar drought at start of cycle. The
scenarios for the forecasts have thus been developed on the basis of this season. However, it
turned out later on, after the month of January, that cropping season of 1994-1995 was
exceptionally dry and far from all experienced seasons, in term of amount and distribution of
rainfall. It was thus concluded that trying to forecast yields too early in January, based on similarity
analysis, is a risky task.
Figure 46: Cumulated rainfall over the cropping season (September till May) for the 1994-1995
cropping season, at Meknes, compared to historical data from 1960-1961 to 1993-1994. Seasons
are arranged in ascending order of cumulated rainfall.
In the case of the cropping season of 1999-2000, drought lasted for three consecutive months
(January till March), which had never been experienced in living memory in Morocco. During the
month of March, a cluster analysis was thus used to look for similar seasons, based on time series
of monthly rainfall data (1960-1999), from September till April.
Similarity approach was used to forecast cereal yields, using all dekadal cumulated rainfall between
September till the end of February (18 dekads), for the cropping season of 2007-2008, taken as an
example. Cropping seasons of 1988-1989 (230 mm) and 2001-2002 (168 mm) were identified as
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
88
closest to season of 2007-2008 (209 mm) (Figure 47). Similarly, the result has not changed till end
of March: similar seasons to 2007-2008 (223 mm) were 1988-1989 (276 mm), 2001-2002 (244 mm)
and 2004-2005 (225 mm). However, in late April, similar seasons to 2007-2008 (251 mm) were
1999-2000 (240 mm), 2001-2002 (316 mm) and 2004-2005 (276 mm).
Figure 47: Similarity analysis of 2007-2008 cropping season, using cumulated dekadal rainfall, at
the country level.
Most similar (analogous) cropping seasons to 2007-2008 are: 1999-2000, 2001-2002, and 2004-
2005. Shaded area in blue embodies range of variation of cumulated dekadal rainfall from historical
data (1988 to 2008).
Concerning the season of 2007-2008, at the 2nd dekad of April, forecasted yields for soft wheat,
durum wheat and barley, were: 13.0, 12.3 and 9.0 Q/ha, respectively, which were relatively close to
observed yields (from official statistics: 13.1, 13.4 and 6.2 Q/ha).
The use of this methodology, should take into consideration that seasons are defined as similar in a
subjective way, because no statistical threshold rule is available for splitting the times series
between similar and non-similar seasons, i.e. they are only relatively similar or dissimilar to each
other.
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
89
2.1.3. TECHNOLOGICAL TREND
Similarity approach must take account of yield trend resulting from technology trend (0.15
Q/ha.year for wheat, Figure 27 and Figure 28), given that identified similar seasons could be far
enough so that technological trend has increased yields. Before computing weighted average yield
of similar past seasons, the respective yields of these seasons must be incremented by the
technological yield trend. However, for the case of barley, no technological trend was statistically
detected from historical time series (Table 7).
2.1.4. FORECASTING ERROR
Accuracy of similarity analysis method can be evaluated, from computation of the average and
confidence interval of some arbitrarily selected similar cropping seasons. The forecasted yield is
taken as equal to average historical yields of the similar seasons, weighted by their relative
Euclidian distances to the current season. Four arbitrarily closest seasons were selected, but fewer
seasons can also be considered if some of them are too distant, which can be visually judged based
on figure like Figure 47. Then, a weighted confidence interval of the yield forecast can be
computed.
Figure 48: Forecasted country yields of soft wheat, for cropping seasons of 2007-2008, 2008-2009,
2009-2010 and 2010-2011, based on similarity analysis of cumulated rainfall since October.
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
90
The forecasting error of the yield forecast can be statistically calculated, considering that yields of
similar seasons are a random sample of a population of forecasted yields. The forecasting error is
thus estimated on the basis of the variation (confidence interval 90%, for example) of similar
seasons' yields. From there, results of similarity analysis show that yields can be forecasted with,
reasonable accuracy, using cumulated rainfall during the season. The forecasts were more accurate
for season of 2007-2008 and 2010-2011 than those of 2008-2009 and 2009-2010 (Figure 48),
because these two latter seasons were exceptionally wet and very dissimilar to all historical
seasons. The differences between forecasted and observed yields, is however significant for
exceptionally rainy seasons like 2008-2009 (547.9 mm) and 2009-2010 (604.8 mm).
The forecasting error
45
for soft wheat, based on similarity analysis method, is relatively low at the
3rd dekad of March, for seasons of 2007-2008 (12.6 ± 0.9 Q/ha against 13.1 Q/ha observed
46
), 2008-
2009 (17.3 ± 0.8 Q/ha against 21.6 Q/ha observed), 2009-2010 (18.2 ± 0.9 Q/ha against 16.6 Q/ha
observed) and 2010-2011 (19.3 ± 0.6 Q/ha against 19.6 Q/ha observed) (Figure 48). Note that the
yield forecast is not very far from observations, even earlier in the season, in February, for seasons
of 2009-2010 and 2010-2011.
2.2. SIMILARITY ANALYSIS USING NDVI
In the framework of a research and development agreement with JRC, INRA uses "MARSOP3" Web
viewer (http://www.marsop.info/marsop3/) developed by Alterra, which is connected to agro-
climatic database of the European Union. MARSOP3 allows real-time analysis of climatic conditions
(temperature, rainfall, solar radiation, potential evapotranspiration, number of cold and hot days,
degrees x days, etc.), and vegetation conditions (NDVI) across the country, which are interpolated
and displayed in grids of 25x25 km spatial resolution. Among its features, this application allows
similarity analysis of cropping seasons, using these weather and vegetation indices (Figure 49).
Similarity analysis should take into account not only rainfall, but also temperature in order to reflect
real weather conditions surrounding crops, since temperature affects the evaporative demand of
the air, and therefore water balance. An example is given for season of 2011-2012, which was
exceptionally dry and cold. Similarity analysis was performed, using dekadal NDVI from early
February till the first dekad of April. Results show that this season was similar to season of 2000-
2001, then relatively more distant to seasons of 2007-2008 and 2004-2005 (Balaghi et al., 2012)
(Figure 50). In contrast, similarity analysis, using cumulated rainfall identified similarity with seasons
of 1999-2000 and 2004-2005. The difference between these two results is due to very cold
temperatures experienced during season of 2011-2012, starting from the 2nd half of January (Figure
45
Average yield and its related 90% confidence limits are presented.
46
Confidence limits of observed yields are not available in historical crop statistical databases, provided by DSS at
province level.
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
91
50). In fact, low temperatures contributed at reducing evapotranspiration and thus at mitigating
drought, which therefore contributed at avoiding low NDVI values.
Figure 49: "MARSOP3" Web viewer, for agro-climatic analysis of the cropping season. Deviation
of NDVI (SPOT-VEGETATION), at the first dekad of April for agricultural lands, from long term
average NDVI is provided by this application. Comparison with other seasons can also be
performed. Spatial resolution of grids is 25x25 km (Source: Balaghi et al., 2012).
From all these examples, one can appreciate the interest and limitations of the similarity analysis.
These examples also show that these statistical methodologies could be used only as
complementary tools to the necessary field expertise of the agro-meteorologist.
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
92
Figure 50: Similar seasons to the 2011-2012 cropping season, based on dekadal NDVI (SPOT-
VEGETATION) over agricultural lands, at the country level. The 2011-2012 cropping season (in
orange) is compared to historical seasons. The 2000-2001 cropping season is the most similar to
2011-2012 with regard to NDVI profile from 1st of February to 1st of April, followed by 2007-2008
and 2004-2005 (Source: Balaghi et al., 2012).
2.2.1. FORECASTS BASED ON SIMILARITY ANALYSIS AND USING NDVI
Soft wheat yield forecasts for seasons of 2007-2008 to 2010-2011, based on similarity analysis and
using dekadal NDVI since start of February (datasets of NOAA-AVHRR data from 1988 to 2011), is
shown in Figure 51. For each of these four seasons, the forecasted yield is equal to the arithmetic
average of the similar seasons weighted by their relative distances to the season to be forecasted.
Arbitrarily, the four most similar seasons were selected.
2.2.2. TECHNOLOGICAL TREND
Technological trend was not considered when using NDVI, as it is already detected through NDVI
values. In fact, this index reflects not only the state of humidity and temperature prevailing during
the season, but also the efficiency use of factors that affect yields such as cultivars and cropping
techniques. In other words, NDVI will be high when weather is favorable (range of rainfall less than
550 mm) or when varieties and production factors are improved.
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
93
Figure 51: Forecasted soft wheat grain yields at the country level, for cropping seasons of 2007-
2008, 2008-2009, 2009-2010 and 2010-2011, based on similarity analysis and using dekadal NDVI
(NOAA-AVHRR) since 1st of February.
2.2.3. FORECASTING ERROR
In general, the forecasting error decreases as the cropping season is ending, except for season of
2007-2008. For example, at the end of February 2011, the forecasted yield was 17.5 ± 1.09 Q/ha. At
late March 2011, it was 19.6 ± 0.51 Q/ha, and at late April 2011 it was 20.0 ± 0.44 Q/ha, against
19.6 Q/ha observed. The forecasting error based on similarity analysis was relatively low at the 3rd
dekad of March, for seasons of 2007-2008 (13.5 ± 1.1 Q/ha, against 13.1 Q/ha observed), 2008-
2009 (17.9 ± 0.5 Q/ha, against 21.6 Q/ha observed) and 2009-2010 (16.1 ± 0.7 Q/ha, against 16.6
Q/ha observed). Note that the forecast was not far from observed, even early in the season, in
February, i.e. when it is not yet possible to use other parametric methodologies, like regression
analysis, to forecast yields. However, for the very particular cropping season of 2008-2009, the
forecast was again very far from observed yield. Nonetheless, the forecasting error is smaller when
using NDVI instead of rainfall.
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
94
2.3. SIMILARITY ANALYSIS USING NDVI AND RAINFALL CO MBINED
The CGMS Statistical Toolbox (CST, http://e-agri.wikispaces.com/CGMSStatTool), is a software
dedicated to statistical crop yield forecasting analysis (Figure 52). Other similar types of software
like CST exist worldwide, such as STATCAT (Curnel et al., 2004). CST has been adapted for Morocco
context in the framework of E-AGRI project. It allows forecasting cereal yields based on two types
of statistical analysis: A scenario analysis relying on Principal Component Analysis (PCA), and a
multiple regression analysis. Both types of analyzes could be performed on time series of cereal
yields and environmental predictors. These predictors are of three types: (1) outputs of WOFOST
simulation model (http://www.wofost.wur.nl/UK/), (2) cumulated rainfall during the cropping
season, and (3) vegetation indices derived from remote sensing (NDVI, DMP). Yields can be
forecasted at multiple spatial scales (provinces, agro-ecological zones, country), at dekadal time
step, starting from late February until the end of April.
Figure 52: The "CGMS Statistical Toolbox" software, for cereal yield forecasting, adapted to
Morocco.
The result of the PCA analysis is presented in Figure 53, for season of 2010-2011, combining both
average NDVI (NOAA-AVHRR) since February till late March and cumulated rainfall since September
till March (times series from 1985 to 2010). Again, season of 1996-1997 was removed from the time
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
95
series due to the peculiarity of its intra-annual rainfall distribution. For example, at the 3rd dekad of
March 2011, result of PCA analysis (Figure 53) shows seasons which were similar to 2010-2011.
They are, in order of similarity, those of 1997-1998, 2002-2003, 2005-2006 and 2003-2004. The
forecasted yield for 2010-2011 is equal to the average yield of these similar seasons, weighted by
their relative distances from 2010-2011, and to which technological trend was added (0.15
Q/ha.year). PCA analysis considers only unique values of NDVI (the average over several dekads)
and rainfall (the total over several dekads), unlike similarity analysis which relies on their
distribution, at dekadal time step, along the cropping season.
Figure 53: Similarity analysis of 2010-2011 cropping season, based on Principal Component
Analysis, using in "CGMS Statistical Toolbox" software CST (http://e-
agri.wikispaces.com/CGMSStatTool).
Similarity analysis was performed using country average NDVI (NOAA-AVHRR) and cumulated
rainfall over the cropping season (data from 1988 to 2011). Similar seasons to 2010-2011 are, in
order of similarity: 1997-1998, 2002-2003, 2005-2006 and 2003-2004.
The exercise was repeated for seasons of 2008-2009, 2009-2010 and 2010-2011, for forecasting
durum wheat yields at the national level (Figure 54). PCA identified diverse similar seasons,
depending on the dekad at which forecasts has been performed. For example, at the 2nd dekad of
February, no match was found between the season's rainfall of 2008-2009 and all historical data
sets since 1985, because of the extreme rainfall experienced during this season (468 mm). At the 3rd
dekad of March, cumulated season's rainfall of 2008-2009 (530.5 mm) was similar to that of 1995-
1996 (523.6 mm) and more distantly, to that of 1986-1987 (413.6 mm). However, at the 3rd dekad
of April 2008-2009 (cumulated rainfall 542.7 mm), a similarity was detected with, in order: 1995-
1996 (546.9 mm), 1986-1987 (438.8 mm), 1990-1991 (371.5 mm) and 2002-2003 (404.2 mm). PCA
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
96
is therefore able to provide a yield forecast, even indicative, early in the season, starting from
March.
The forecast is all the more accurate if it exists in the available time series seasons which are
relatively similar to the current season, in terms of NDVI and rainfall. If the season is singular, PCA is
unlikely to lead to a reliable forecast. This was the case for seasons of 2008-2009 and 2009-2010,
which were remarkably wet (> 550 mm). In the case of season of 2010-2011, the difference
between forecasted and observed yields of durum wheat, at the national level, was 2.1 Q/ha at the
3rd dekad of February, 1.4 Q/ha at the 3rd dekad of March, and only 0.7 Q/ha at the 3rd dekad of
April (Figure 54).
Figure 54: Observed and forecasted durum wheat grain yield at the country level, based on
Principal Component Analysis, for the 2008-2009, 2009-2010 and 2010-2011 cropping seasons,
using NDVI (NOAA-AVHRR) and cumulated rainfall (Data from 1988 to 2011), at dekadal time
step.
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
97
3. REGRE SSI ON MODELS A PPR OACH
The regression models approach consists in identifying climatic variables (precipitation,
temperature, etc.) or agro-climatic indices (NDVI, etc.) which are significantly correlated to crop
yields. Forecasting crop yields using regression models strongly relies on agronomic expertise, as
well as statistical analysis, Geographic Information Systems and satellite image processing skills.
The main precaution before using this approach is to ensure that non-climatic factors do not
interfere during regression analysis, such as technological trend (change of varieties, irrigation, new
agronomic practices, etc.), or socio-economic factors (prices, subsidies, etc.). This approach
requires to always considering "the law of the limiting factors" also known as the “law of the
minimum”. This law is one of the most important principles of practical agronomy, which was set in
1828 by Carl Sprengel, and then adapted by Liebig in 1850. The principle of this law states that crop
yield is limited by factors that first are insufficiently available.
A study covering 12 European countries revealed that simple climate, soil and economic variables
could explain most of the variation of wheat yields and their trend in Europe (Bakker et al., 2005).
Among these, climate variables (precipitation, temperature, evapotranspiration and global
radiation) account for the largest percentage of yield variability, more than the contribution of
variables related to soils and economy. In some countries, where climate is not a limiting factor,
either because rainfall is abundant or because irrigation is available, crop yields are mostly
dependent on biotic stresses (diseases, pests) or agronomic inputs (fertilization, pest treatment). In
general, in these countries, yields increase continuously due to technology improvement, and vary
little from one year to another. For example, in Belgium, the sophisticated B-CGMS crop monitoring
and forecasting system (http://b-cgms.cra.wallonie.be/en/), which is an adaptation of the European
CGMS, explains less wheat yield variation than only a technological trend, computed based on
statistical linear regression (Decrem et al., 2002).
In the case of Morocco, inter- and intra-annual rainfall distribution is obviously the factor that
explains much of crop yields' variability, in general, and cereal variability in particular, due to the
aridity of climate and limited irrigation capabilities (only 1.46 million hectares irrigated,
representing 17% of agricultural lands).
The regression models approach can be used at multiple spatial scales, depending on data
availability. With the development of weather interpolation methods and satellite imagery, the
main limitation to the use of this approach for finer spatial scales, comes from the availability of
agricultural statistics. Cereal statistics were provided by DSS at provincial level, based on sub-
provincial sampling technique, called "area frame sampling" (see chapter II_1.6).
The regression models approach start with the constitution of a database containing all relevant
agro-climatic factors or indices and crop yields statistics, at the largest (finest) spatial scale (the
provinces, in the case of cereals for Morocco). Once the database is established, the
agrometeorologist can start data analysis. Crop yields are then linked to single or multiple
predictors (factors or indices), using usual spreadsheets or using more sophisticated statistical
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
98
software packages. The "golden rules" to consider, when using linear regression models for crop
forecasting, are described by Gommes et al. (2010).
In crop forecasting, simple or multiple linear regression models are often used as an adjunct to crop
growth simulation models. This is the methodology used by the JRC for crop monitoring and
forecasting over all countries of the European Union and neighboring countries, including the
Maghreb. This is a two-step methodology, which involves first computation of indicators informing
on the state of the growing season (water balance, vegetative phases, etc.) through the WOFOST
model. Indices derived from this model are then used as predictors in multiple linear regressions for
crop yield forecasting, calibrated against official agricultural statistics. Forecasted yields are
therefore not provided directly by WOFOST model, but from linear regression between official
yields and outputs coming this.
Selection of the predictors of the linear regression models involves two steps, called "calibration"
and "validation". The calibration step consists in finding, among all potential predictors (1 to several
hundred depending on the case), those that are most correlated to crop yields, in order to
maximize the coefficient of determination (R2) of the regression model. Statistical procedures for
automatic selection of predictors (Forward, Stepwise, etc.) are available in most statistical software
products, so as to avoid collinearity
47
problems. However, regression models with high R2 have not
necessarily good predictive power. For this reason, the validation step is performed. Validation is
performed based on a "cross-validation" technique, which is recommended in crop forecasting
(Balaghi et al., 2008). This validation technique is used to check the reproducibility of results, and
thus to verify the predictive power of the regression model, for a "new" year which has not been
used during the calibration step. A coefficient of determination of the cross validation
48
(Rp2) as well
as the forecasting error of the model are thus computed. Preference should be given to regression
models with low “shrinkage” (difference between R2 and Rp2).
3.1. FORECAST ING USING RAINFALL AS PREDICTOR
Operational crop forecasting attempts, using rainfall as predictor, are numerous in Morocco
(Douguedroit and Messaoudi, 1998; Douguedroit et al., 1998; Skees et al., 2001; Stoppa and Hess,
2003; Balaghi et al., 2008). These various forecasting models were all based on empirical statistical
approaches. These statistical approaches are based on single or multiple linear regressions, at
different spatial scales (country, agro-ecological zones or provinces), between cereals yields and
one-time rainfall values (daily, ten-day, monthly or yearly), or cumulative rainfall over several
periods within the season. These linear regression models generally provide reliable forecasts,
47
The presence of high correlations, between explanatory variables of a multiple linear regression model.
48
The coefficient of determination in cross-validation is defined as the coefficient of determination between observed
and predicted values (see Balaghi et al., 2008). Rp
2 gives an indication on the replicability of results and checks the
prediction performance of a model.
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
99
varying in accuracy according to available time series, to the spatial scale of the study and to the
species considered (soft wheat, durum wheat and barley).
The relationship between cereal yields and cumulated rainfall during the cropping season is
improving continuously, proportionally to the accumulation of rainfall along the season, from mid-
October to late March (Figure 55). High coefficients of determination (R2) reached at the end of
March as well as the continuous improvement of R2 during the course of the cropping season,
demonstrate the reliability of using cumulated rainfall as predictor in forecasting models based on
statistical linear regression. At late March, R2 computed over the period 1988-2011, reaches values
of 61, 68 and 70%, for soft wheat, durum wheat and barley, respectively. Also, the R2 curves for the
three species suggest using cumulated rainfall between October and March for cereal yield
forecasting.
Figure 55: Coefficient of determination (R2), by step of one dekad, of the regression line between
grain yields of the three main cereals (soft wheat, durum wheat and barley) in Morocco and
cumulated rainfall since September till May, at the country level. Average (1988-2011) cumulated
rainfall is displayed in blue bars, for illustrating the steady increase of R2 with rainfall during the
season, from the 2nd dekad of October to the 2nd dekad of March.
3.1.1. FORECASTING USING RAINFALL CUMULATED OVER ALL THE CROPPING SEASON
Cereal yield forecasting based on linear regression models, can be achieved in a simple way using
only the cumulated rainfall from October till March. Forecasts can further be improved using a
logarithmic regression model, which fits more closely the scatter plots (El Aydam, Balaghi and
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
100
Baruth, 2010). Also, the logarithmic form suggests that rainfall should be first linearized
49
before its
use in combination with other predictors (NDVI, ET0, etc.), in case of using multiple regression
models. At national level, the correlation between yields of the three cereals and rainfall are high,
based on a logarithmic regression (Figure 56). However, some cropping seasons, like those of 1996-
1997 and 1999-2000, do not fit into this type of relationship, due to their particular rainfall
distribution. For example, in season of 1996-1997, although the rainfall was above average (466
mm), drought which lasted from February to early April (Figure 38) affected cereal yields (9.4, 9.1
and 6.6 Q/ha for soft wheat, durum wheat and barley, respectively). Disregarding season of 1996-
1997, correlations between yields and rainfall become strong, with R2 values of 84, 82 and 77% for
soft wheat, durum wheat and barley, respectively (Figure 56).
Soft wheat
49
A regression model is called "linear" when it is linear in parameters (coefficients of regression). A model is called
"intrinsically linear" when it is not linear in the parameters, but it could be linearizable using variable transformation.
Precautions should be taken when linearizing an intrinsically linear model. Finally, a model is called "intrinsically non-
linear" when its linearization is not possible.
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
101
Durum wheat
Barley
Figure 56 : Relationship between country yields of soft wheat, durum wheat and
barley, and cumulated rainfall over the cropping season (from September to
March), for data series of 1988 to 2008.
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
102
The forecasting methodology based on regression models and using cumulated rainfall as predictor,
was used for the forecasting of cereal harvests, since season of 2008-2009 by INRA in collaboration
with the JRC, in conjunction with other methodologies based on crop simulation models (WOFOST),
similarity analysis, and the use of remote sensing data as predictor (NDVI) (Narciso and Balaghi,
2009; El Aydam, Balaghi and Baruth, 2010; El Aydam and Balaghi, 2011; Balaghi et al., 2012). For
example, in 2010 the expected country yields during late March, based on cumulated rainfall during
the season, were 2.2, 2.0 and 1.6 tons/ha for soft wheat, durum wheat and barley, respectively,
against 1.7, 1.8 and 1.4 tons/ha, recorded in official statistics. As seen in this example, although the
forecasts were not far from observations, there is still room for improvement which could be
sought in a combination of other methodologies and yield indicators.
Similarly, cereal yields can be forecasted at the level of agro-ecological zones, with reasonable
accuracy, except for the Saharan zone where production depends heavily on irrigation. Crop yield
forecasting for the agro-ecological zones is feasible based on simple linear regression models, using
cumulated rainfall from October till March (Table 11). The form of the relationship can be
approximated by a logarithmic function in 5 of the 6 areas. R2 of the regression models are above
62%, in the Mountain and Saharan zones, which contribute very little to national production. R2 is
highest in the Intermediate zone, where average season's rainfall is equal to 366 mm.
Table 11 : Coefficient of determination (R2) of the linear regression models between cereals yields
(soft wheat, durum wheat and barley) and cumulated rainfall over the cropping season (from
October till March), at the level of the agro-ecological zones (Data from 1988 to 2008).
Agro-ecological zone
Shape of the relationship
R2
Soft wheat
Durum wheat
Barley
Favorable
Logarithmic
0.70
0.66
0.71
Intermediate
Logarithmic
0.80
0.73
0.70
Unfavorable East
Logarithmic
0.62
0.75
0.59
Unfavorable South
Logarithmic
0.71
0.65
0.79
Mountain
Logarithmic
0.53
0.61
0.63
Saharan
Linear
0.44
0.03
0.66
3.1.2. FORECASTING USING RAINFALL CUMULATED BY PHASES DURING THE CROPPING
SEASON
Cereal yield forecasting based on single or multiple regression models, using cumulated rainfall
over non-overlapping periods during the cropping season (Balaghi et al., 2008). Forecasts can be
performed using rainfall, cumulated over several periods of 2 to 8 consecutive months during the
season (Figure 57). In all cases, the periods to be taken into account orbit the high rainfall period of
the cropping season, which runs in average from November till February (Figure 39). For periods of
4 months and over, simple regressions models are satisfactory. For these periods, the forecast is
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
103
better (R2=61%), when considering a 6 months period, between October and March. For periods of
3 months and less, multiple regressions are more appropriate to achieve significant R2. For periods
of 3 months, best forecasts are achieved (total R2=70%) when using rainfall cumulated over January
till March (partial R2=38%) and September till November (partial R2=32%). Maximum accuracy of
the forecasts are obtained (R2=78%), when using 3 periods of 2 consecutive months, from February
till March (partial R2=34%), from October till November (partial R2=35%), and from December till
January (partial R2=9%). The consecutive periods were selected so as to not overlap and avoid thus
multicollinearity between predictors. The optimal number of periods (predictors) to consider
depends on the length of the data series available to ensure to avoid overfitting
50
problems.
.
Figure 57: Coefficient of determination (R2) of linear regression models between grain yields of
the three main cereals (soft wheat, durum wheat and barley) in Morocco and rainfall cumulated
over significant periods during the cropping season, at the country level (data from 1988 to 2011).
R2 is higher when using cumulated rainfall over 6 months, and in partitioning the cropping
season’s rainfall into two groups of three months and three groups of 2 months each (in these
two latter cases, lengths of colored bars correspond to partial R2 of multiple regressions).
50
Overfitting occurs when a statistical model describes random error or noise instead of the underlying relationship.
Overfitting usually occurs when a model is overly complex, with too many predictors (multicollinearity between
predictors) compared to the number of observations.
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
104
3.2. FORECAST ING USING AGRO-CLIMATIC INDICES
Some agro-climatic indices computed from water balance models, combining climatic factors with
crop and soil characteristics, are highly correlated with cereal yields. Agro-climatic indices are
expected to provide an added value, in terms of vegetation monitoring or crop forecasting,
compared to climatic factors used separately. Examples of such agro-climatic indices are given by
AgroMetShell software, which is part of a set of tools developed by FAO for decision making in the
context of food security.
Three indices directly computed from this software were used for illustration: Water Surplus-Deficit
Water (WSD), Water Requirement Satisfaction Index (WRSI), and Soil Water Storage (SWS). As well,
a fourth new index was added, computed as the cumulated WSD over consecutive dekads (ΣWSD)
starting from the start of the season in November. Meknes province was selected as sub-humid test
site, assuming 150mm soil water holding capacity (Balaghi, 2006). These four indices were
statistically correlated with soft wheat yields at the provincial level (Balaghi, 2006; Figure 58).
Among these indices, ΣWSD (maximum R2=80% at the 3rd dekad of March) and SWS (maximum
R2=77% at the 3rd dekad of February) were best correlated to soft wheat yields.
Figure 58: Coefficient of determination (R2) of regression models between soft wheat grain yields
and agro-meteorological indices derived from AgroMetShell program of the FAO, in Meknes
province (Source: Balaghi, 2006).
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
105
3.3. FORECAST ING USING REMOTE SENSING
Compared to weather data, indices derived from remote sensing, such as NDVI, have two main
advantages: first, they express the effects of the main environmental factors affecting yields
(rainfall, temperature, soils, diseases, cropping practices, etc.); second, they are discontinuously
available at the extent of country and at high spatial and temporal resolution (as "raster", also
known as pixel). Indeed, in Morocco, synoptic weather stations belonging to DMN (44 in total) are
sparse, and most of them are located in the Atlantic plains and very little in mountainous areas or in
rangelands (Figure 12).
3.3.1. RELATIONSHIP BETWEEN YIELDS AND NDVI
NDVI is a powerful indicator of the resulting effect of the environmental factors and their
interactions with crops. It expresses the combined effect of weather, varieties, diseases, soils,
terrain, cropping techniques, etc. Indeed, there is a close and linear relationship between NDVI
(SPOT-VEGETATION) and yields for the three winter cereals (soft wheat, durum wheat and barley),
at the national level (Figure 59).
Figure 59: Coefficient of determination (R²) of regression models between grain yields of the
three main cereals (soft wheat, durum wheat and barley) in Morocco, and dekadal NDVI (SPOT-
VEGETATION), at the country level (data from 1999 to 2011), by step of one dekad, since 1st
dekad of January.
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
106
The relationship becomes stronger in February and peaks at the first dekad of March (R2=89%), for
soft and durum wheat, and at the first dekad of April (R2=81%) for barley. Peaks of correlation
coincide with long term NDVI peak during within the season, given that peak of NDVI varies from
one year to another, depending on season's rainfall and temperature and on sowing dates.
However, seasons of drought slacken the relationship between cereal yields and NDVI. When
removing dry seasons (1999-2000, 2004-2005 and 2006-2007) from time series, yield forecasts are
improved and can be performed as early as January, with a coefficient of determination (R2) of 70%
for wheat at national level.
The relationship between cereal yields and dekadal NDVI (Figure 59) at national level improves over
time, and reaches a peak at the first dekad of March for wheat and at the 2nd dekad of March for
barley. The relationship persists until the end of April for barley, whereas it relaxes for wheat.
In order to improve accuracy of the forecasts based on linear regression models, a transformation
of variable is performed on NDVI (Genovese et al., 2001). It consists in performing regression
analysis on average dekadal NDVI values over several dekads, usually taken around period when
NDVI is usually at maximum during the cropping season (Figure 60). The use of average NDVI over
dekads is a way of taking into account differences in NDVI profile, from one season to another.
Relationship is gradually improving during the cropping season starting from February till April, with
maximum R2 reached in late April. In Morocco, best correlations with cereal yields were obtained,
when averaging NDVI from February till April, followed by February till March. This time period has
been identified by testing all possible combinations average NDVI over several dekads, between
November and April, at both national and provincial levels (Balaghi et al., 2008).
Wheat yields can be forecasted, based on linear regression models, with coefficients of
determination (R2) greater than 80%, using average dekadal NDVI from 1st February till end March
(Figure 60). Coefficient of determination exceeds 90%, when using average dekadal NDVI from 1st
February till end April. For barley, the forecast is more accurate till the 2nd dekad of April. The
reasons why optimal period, for averaging NDVI, varies from one species to another remain to be
elucidated. Besides, the relationship between cereal yields and NDVI is likely to improve gradually
as time series become longer.
NDVI time series are delivered by NOAA-AVHRR sensor, since 1982 at 1x1 km spatial resolution and
at dekadal time step (10 days). However, a number of scenes of this sensor suffer from poor quality
due to poor geometric correction and cloud detection. NDVI values delivered by this sensor can be
used for forecasting cereal yields in Morocco (Balaghi et al., 2008), at both national and provinces
levels. NDVI delivered by MODIS sensor, which has better spatial resolution (250 meters), can also
be used in crop forecasting. Research findings (unpublished) indicate that NDVI values delivered by
SPOT-VEGETATION (onboard SPOT-5) are better correlated to cereal yields, than those of NOAA-
AVHRR and are equivalent to those of MODIS. However, SPOT-VEGETATION mission will end in
2013-14. It will be replaced by Proba-V
51
sensor (Project for On-Board Autonomy - Vegetation) to
51
For more information, see VITO website: http://www.vgt.vito.be/.
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
107
be launched in 2013, shipping a lighter version of the VEGETATION instrument. A restatement of
SPOT-VEGETATION images is expected to ensure interoperability with Proba-V, in order to take
advantage from SPOT-VEGETATION time series available since 1998.
Figure 60: Coefficient of determination (R2) of regression models between grain yields of the
three main cereals (soft wheat, durum wheat and barley) in Morocco, and average NDVI (SPOT-
VEGETATION), at the country level (data from 1999 to 2011), by step of one dekad, since the 1st
dekad of February.
3.3.2. RELATIONSHIP AT NATIONAL LEVEL
Cereal yields can be forecasted based on linear regression with NDVI at national level. The
relationship between cereal yields (soft wheat, durum wheat and barley) and NDVI (SPOT-
VEGETATION) is linear and very strong (Figure 61), unlike the relationship with rainfall (Figure 56)
which is logarithmic.
Coefficient of determination (R2) of the relationship between yields and average dekadal NDVI from
February till March reaches 89% for soft wheat, 88% for durum wheat and 69% for barley. The
relationship is even stronger when average dekadal NDVI from February till April is used instead,
with values of R2 equal to: 93% for soft wheat, 94% for durum wheat and 84% for barley. This
relationship was used by INRA, in combination with other methods, for performing cereal yield
forecasts, which were published in 2009, 2010, 2011 and 2012 (Balaghi et al., 2012; El Aydam et al.,
2010, 2011; Narciso and Balaghi, 2009).
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
108
Soft wheat
Durum wheat
Barley
Figure 61: Linear regression between country grain yields of the three main cereals (soft wheat,
durum wheat and barley) and average dekadal NDVI (SPOT-VEGETATION) between 1st of
February and end of March, and between 1st of February and end of April (data of 1999 to 2011).
3.3.3. RELATIONSHIP AT THE LEVEL OF THE AGRO-ECOLOGICAL ZONES
Cereal yield forecasts can be performed at the level of the agro-ecological zones of Morocco with
good accuracy, starting from the end of March. The relationship between yields of winter cereals
(soft wheat, durum wheat and barley) and NDVI (SPOT-VEGETATION) is very strong at this spatial
level, except for the Saharan zone (Table 12). Forecasts can be achieved with minimum error, in
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
109
late April, for the three species. The relationship is less consistent for agro-ecological zones than for
national level. The relationship is stronger for the Favorable, Intermediate and Unfavorable South
zones than for the Mountain, Unfavorable East and Saharan zones.
Table 12: Coefficient de determination (R2) of the regression models between grain yields of the
three winter cereals (soft wheat, durum wheat and barley) and average NDVI (SPOT-
VEGETATION) from February till March and from February till April, at the level of the agro-
ecological zones of Morocco (Data of 1999 to 2011).
Agro-ecological zone
February till March
February till April
Soft wheat
Durum wheat
Barley
Soft wheat
Durum wheat
Barley
Favorable
0.73
0.74
0.66
0.85
0.83
0.83
Intermediate
0.79
0.77
0.72
0.87
0.83
0.84
Unfavorable South
0.70
0.69
0.69
0.74
0.77
0.79
Unfavorable East
0.24
0.46
0.38
0.41
0.69
0.67
Mountain
0.51
0.58
0.45
0.52
0.65
0.57
Saharan
0.13
0.03
0.30
0.16
0.03
0.35
3.3.4. RELATIONSHIP AT ADMINISTRATIVE PROVINCE LEVEL
Cereal yields can be forecasted at province level, based on linear regression models, and using
average dekadal NDVI (SPOT-VEGETATION) as predictor from February till March, or with higher
accuracy from February till April. Four classes of coefficient of determination (R2) are shown in
Figure 62, for soft wheat, durum wheat and barley. R2 higher than 62% are statistically very highly
significant (probability level < 0.001). Those between 45 and 62% and between 30 and 45% are
highly significant (probability level < 0.01) and significant (probability level < 0.05), respectively. R2,
lower than 30%, is considered not statistically significant.
Classes of high R2 (> 62%) are located in the south-west to north-east axis (Essaouira to Al Hoceima)
of the country, where most of cereal areas are located. The relationship between cereal yields and
NDVI are relatively lower in the Northern provinces (Tangier, Tetouan, Larache), eastern provinces
(Nador and Oujda), mountainous provinces (Taza, Khenifra, Beni Mellal, Azilal) and in southern
provinces (Marrakech, Chichaoua, Taroudant, Agadir, Tiznit). In late March, R2 are above 62% in
most provinces, for soft wheat. Instead, at that period of the cycle, high R2 (> 62%) are encountered
in fewer provinces, in the Settat - Taounate axis for durum wheat, and further south in the
Essaouira - Khemisset axis for barley.
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
110
Soft wheat
February to March
February to April
Durum wheat
February to March
February to April
Barley
February to March
February to April
Figure 62 : Mapping of four classes of coefficient of determination (R2) of the linear regression
models between yields of soft wheat, durum wheat and barley, and average dekadal NDVI (SPOT-
VEGETATION) from February till March and from February till April.
Provinces in blue, green and orange, indicate statistical significance of R², at probability levels of
0.001, 0.01, and 0.05, respectively. Provinces in yellow indicate non statistical significance of R2. Non
colored provinces correspond to non-agricultural areas, as identified by GICropV2 crop mask (Data
of 1999 to 2011).
R2
TATA
FIGUIG
ERRACHIDIA
OUARZAZATE
OUJDATAZA
SAFI
FES
SETTAT
AZILAL
TIZNIT
BOULMANE
TAROUDANTE
KHENIFRA
GUELMIM
NADOR
ASSA-ZAG
KHEMISSET
EL JADIDA
AGADIR
MARRAKECH
KENITRA
CHICHAOUA
IFRANE
EL KELAA SRAGHNA
BENI MELLAL
TAOUNATE
KHOURIBGA
LARACHE
ESSAOUIRA
CHEFCHAOUEN
SIDI KACEM
TAN-TAN
RABAT
AL HOCEIMA
EL HAJEB
TETOUAN
MEKNES
BEN SLIMANE
TANGER
CASABLANCA
> 0,62 ***
0,45 < < 0,62 **
0,30 < < 0,45 *
< 0,30 ns
TATA
FIGUIG
ERRACHIDIA
OUARZAZATE
OUJDATAZA
SAFI
FES
SETTAT
AZILAL
TIZNIT
BOULMANE
TAROUDANTE
KHENIFRA
GUELMIM
NADOR
ASSA-ZAG
KHEMISSET
EL JADIDA
AGADIR
MARRAKECH
KENITRA
CHICHAOUA
IFRANE
EL KELAA SRAGHNA
BENI MELLAL
TAOUNATE
KHOURIBGA
LARACHE
ESSAOUIRA
CHEFCHAOUEN
SIDI KACEM
TAN-TAN
RABAT
AL HOCEIMA
EL HAJEB
TETOUAN
MEKNES
BEN SLIMANE
TANGER
CASABLANCA
TATA
FIGUIG
ERRACHIDIA
OUARZAZATE
OUJDATAZA
SAFI
FES
SETTAT
AZILAL
TIZNIT
BOULMANE
TAROUDANTE
KHENIFRA
GUELMIM
NADOR
ASSA-ZAG
KHEMISSET
EL JADIDA
AGADIR
MARRAKECH
KENITRA
CHICHAOUA
IFRANE
EL KELAA SRAGHNA
BENI MELLAL
TAOUNATE
KHOURIBGA
LARACHE
ESSAOUIRA
CHEFCHAOUEN
SIDI KACEM
TAN-TAN
RABAT
AL HOCEIMA
EL HAJEB
TETOUAN
MEKNES
BEN SLIMANE
TANGER
CASABLANCA
TATA
FIGUIG
ERRACHIDIA
OUARZAZATE
OUJDATAZA
SAFI
FES
SETTAT
AZILAL
TIZNIT
BOULMANE
TAROUDANTE
KHENIFRA
GUELMIM
NADOR
ASSA-ZAG
KHEMISSET
EL JADIDA
AGADIR
MARRAKECH
KENITRA
CHICHAOUA
IFRANE
EL KELAA SRAGHNA
BENI MELLAL
TAOUNATE
KHOURIBGA
LARACHE
ESSAOUIRA
CHEFCHAOUEN
SIDI KACEM
TAN-TAN
RABAT
AL HOCEIMA
EL HAJEB
TETOUAN
MEKNES
BEN SLIMANE
TANGER
CASABLANCA
TATA
FIGUIG
ERRACHIDIA
OUARZAZATE
OUJDATAZA
SAFI
FES
SETTAT
AZILAL
TIZNIT
BOULMANE
TAROUDANTE
KHENIFRA
GUELMIM
NADOR
ASSA-ZAG
KHEMISSET
EL JADIDA
AGADIR
MARRAKECH
KENITRA
CHICHAOUA
IFRANE
EL KELAA SRAGHNA
BENI MELLAL
TAOUNATE
KHOURIBGA
LARACHE
ESSAOUIRA
CHEFCHAOUEN
SIDI KACEM
TAN-TAN
RABAT
AL HOCEIMA
EL HAJEB
TETOUAN
MEKNES
BEN SLIMANE
TANGER
CASABLANCA
TATA
FIGUIG
ERRACHIDIA
OUARZAZATE
OUJDATAZA
SAFI
FES
SETTAT
AZILAL
TIZNIT
BOULMANE
TAROUDANTE
KHENIFRA
GUELMIM
NADOR
ASSA-ZAG
KHEMISSET
EL JADIDA
AGADIR
MARRAKECH
KENITRA
CHICHAOUA
IFRANE
EL KELAA SRAGHNA
BENI MELLAL
TAOUNATE
KHOURIBGA
LARACHE
ESSAOUIRA
CHEFCHAOUEN
SIDI KACEM
TAN-TAN
RABAT
AL HOCEIMA
EL HAJEB
TETOUAN
MEKNES
BEN SLIMANE
TANGER
CASABLANCA
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
111
3.4. FORECAST ING USING WEATHER DATA AND NDVI
The cereal yield forecasting approach, based on multiple regression models
52
and using a
combination of predictors (rainfall, NDVI or temperature), improves accuracy at fine spatial scales.
This approach was recently tested in Morocco using NDVI alone or in combination
53
with rainfall or
temperature, at province level (Balaghi et al., 2008). This was made possible through to the
collaboration with University of Liege (ULg, Belgium), JRC and VITO. NDVI images provided by JRC
cover the entire agricultural areas, from northern Morocco to south to 20° latitude north.
Compared to data coming from ground weather stations, indicators derived from remote sensing,
like NDVI, have the advantage of being available throughout the extent of the country, without
discontinuity at high spatial and temporal resolutions (in "raster" form, known as pixel), which is
well suited for crop yield forecasting.
Forecasts can be performed operationally at the levels of the provinces, agro-ecological zones and
country. Regression models can be developed separately, for each province or agro-ecological zone.
Forecasted yield for each province can even be aggregated to national level, proportionally to their
respective cropped areas, for providing yield forecast for the entire country, with more
computational workload but with better accuracy. The coefficients of determination (R2) of these
models were very high, between 72 and 98%, except in the arid southern provinces of the country
which contribute little to national production (Balaghi et al., 2008). These regression models were
relatively stable, since their coefficients of determination in cross-validation (Rp2) were still between
59 and 94%. Also, forecasts were relatively accurate and early, from two months before harvests.
Among the variables used, NDVI explains the largest share (partial R2>40%) of yield variation for
wheat in 14 out of 23 provinces studied, i.e. in provinces which account for more than 69.4 % of the
total national production. Rainfall explains most of yield variation in the provinces of Larache,
Agadir, Oujda, Beni Mellal, Tangier, Marrakech and Ouarzazate. Temperature explained most of
yield variation only in Tangier and Tetouan, which are located in wet environment. In these two
provinces, temperature affects negatively wheat yields, because it increases evapotranspiration
and reduces period of growth.
A recrudescent interest of these statistical approaches has been prompted by the growing
availability of climate data, over long time series, and by software development, allowing storage
and process of large amounts of datasets, as well as by the availability of software products
dedicated to statistical analysis and remote sensing processing. Generally, statistical forecasting
approaches are criticized because they do not apply outside the range of variation of the data used
for their calibration. This criticism can be largely avoided in the case of Morocco, due to the high
variability of climate and geography. In fact the probability of encountering climatic seasons outside
52
This approach has been successfully applied also in Canada (Mkhabela et al., 2011), in Romania (Lazar et al., 2009),
Senegal (Kouadio, 2007) and in Burkina Faso (Ramde, 2007).
53
In these models, the « Stepwise » procedure for automatic selection of predictors has been applied in order to avoid
multicollinearity between explanatory variables.
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
112
the range of historical variation is always possible, but nevertheless reduced due to the extreme
cropping seasons experienced, both very dry (1994-1995, 1996-1997, 1999-2000, 2000-2001 and
2004-2005) and very wet (1995-1996, 2008-2009, 2009-2010 and 2010-2011).
Meknes
Settat
Figure 63 : Relationship between observed and forecasted yields of soft wheat at Meknes and
Settat, based on multiple linear regression models, involving NDVI, rainfall and temperature
(Source: Balaghi et al., 2008).
3.4.1. ESTIMATION OF THE FORECASTING ERROR
The forecasting error of the linear regression models using average dekadal NDVI (SPOT-
VEGETATION) is improving gradually, as the cropping season is ending, from February till end of
April, regardless of the spatial scale. In Figure 64 are shown examples of soft wheat forecasts, at
national level, for cropping seasons of 2008-2009, 2009-2010 and 2010-2011. These cropping
seasons were of course not used during the calibration process. Coefficients of determination (R2)
and consequently forecasting errors, have gradually improved starting from February and peaked in
late April (R2> 88%).
During the extreme rainy season of 2008-2009 (548 mm), the forecast was close to the official yield
(difference of 1.65 Q/ha). This cropping season was particular, outside the range of historical
rainfall variation (between 227 and 419 mm, from 1999 to 2008). Also, the forecasts were very
close to official yields for cropping seasons of 2009-2010 and 2010-2011. Whatever are the season
and the cereal, confidence intervals of the forecasts have improved gradually as the cropping
seasons reach completion. Also, confidence interval at 70% always comprises official yields.
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
113
2008-2009
2008-2009
2008-2009
Figure 64 : Increase of precision and accuracy of the country soft wheat yield forecast, based
linear regression models, using NDVI (SPOT-VEGETATION), as the cropping seasons reach
completion. Examples of the 2008-2009, 2009-2010 and 2010-2011 cropping seasons are shown.
4. COMBINED APP ROACH
Cereal yields can be forecasted based on a combined approach (Table 13), which relies on a
combination of four approaches and use of various types of predictors (rainfall, temperature,
NDVI):
1. Non-parametric approach: multiple regression models, using qualitative predictors
(rainfall) for describing the status of the cropping season;
2. Similarity approach: similarity analysis based on Euclidean distance and Principal
Component Analysis, using rainfall and NDVI ;
3. Regression models approach: Single or multiple linear regression models, using rainfall,
temperature and NDVI ;
4. Simulation approach: Crop growth simulation models (WOFOST), using raw climatic, crop
and soil data.
This approach has the advantage of providing yield forecasts by various independent ways
(Gommes et al., 2010), allowing an estimation of the uncertainty of the forecasts. This combined
approach has been formalized in the CST software for Morocco, in the framework of E-AGRI
project.
Agrometeorological Cereal Yield Forecasting in Morocco
Cereal yield forecasting
Balaghi R., Jlibene M., Tychon B., Eerens H.
114
Table 13: Schematic table of the combined approach and predictors developed for cereal yield
forecasting in Morocco. Shaded areas indicate the predictors used for each method.
Combined Approach
Method
Predictor
Rainfall
Temperature
NDVI
morpho physiological
Non-parametric
Similarity
Euclidian
distance
PCA
Parametric
Simple
regression
Multiple
regression
Simulation WOFOST
This approach was used to forecast cereal yields with high accuracy, for cropping seasons of 2008-
2009 (Balaghi and Narciso, 2009), 2009-2010 (El Aydam, Balaghi and Baruth, 2010), 2010-2011 (El
Aydam and Balaghi, 2011) and 2011-2012 (Balaghi et al., 2012). Forecasts concerned soft wheat,
durum and barley, while the data on the area of each species were provided by DSS from field
surveys.
The accuracy of the forecasts is continuously refined, since it depends on three key conditions that
are still improving:
1. Time series are increasing, allowing to take account of more contrasted cropping
environments, improving hence the robustness of regression models and similarity analysis ;
2. The number of weather stations is increasing in Morocco, which will give floor to better
spatial characterization of climatic conditions at fine spatial scale ;
3. The crop mask, which separates agricultural areas from other lands (forest, rangeland,
wasteland, lakes, cities, etc.) can be replaced by a cereal mask which will closely filter NDVI
images ;
4. The NDVI images are continually affordable and also improved, in terms of quality and
spatial resolution.
Agrometeorological Cereal Yield Forecasting in Morocco
Institutionalization &
operationalization
Balaghi R., Jlibene M., Tychon B., Eerens H.
115
VII. INSTITUTIONALIZATION AND OPERATIONALIZATION OF THE
FORECASTING SYSTEM
The combined approach for cereal yield forecasting has been automated in an operational
forecasting system, in order to forecast cereal yields in Morocco, called "CGMS-MAROC". This
system is an adaptation and improvement of the European CGMS system. The native CGMS consists
of three levels: (1) the acquisition of meteorological data and their interpolation on a square grid of
25x25 km spatial resolution, (2) the simulation of crop growth using the European WOFOST model,
and (3) cereal yield forecasting based on outputs derived from WOFOST model. CGMS-MAROC has
the same architecture than the European CGMS but has been improved, since it includes statistical
models for forecasting crop yields developed at INRA. Also, CGMS-MAROC runs with spatialized
weather data on finer climate grids of 10x10 km resolution. CGMS-MAROC can currently forecast
cereal yields at the level of the country, agro-ecological zone and province levels. It is operated by
INRA and managed in collaboration with DMN and DSS, in the framework of an official partnership.
It is intended to support DSS, as this administration is officially mandated to provide crop statistics
and forecasts. CGMS-MAROC is installed on a central server at DMN, which is responsible for
updating and delivering the weather grid. A Web viewer installed at INRA, was specifically
developed for checking data and making preliminary analysis concerning the season. CGMS-MAROC
is the first operational agrometeorological crop forecasting system available in Morocco,
institutionalized by a strategic partnership which allows its development and sustainability.
To effectively deal with climate change and climate risks in general, initiatives to improve
governance systems
54
and policies for better adaptation and proactivity, should be promoted. The
success of the adaptation of agriculture to climate risks depends on national institutions which are
responsible for food security, directly or indirectly. Participants to the 3rd World Climate Conference
organized by the World Meteorological Organization (http://www.wmo.int/wcc3/) highlighted the
need for capacity building and linkages between meteorological, agricultural and scientific research
institutions (Balaghi et al., 2010).
The combined approach for cereal yield forecasting has been automated in an operational
forecasting system, in order to forecast cereal yields in Morocco, called "CGMS-MAROC"
(www.cgms-maroc.ma), which is an adaptation and improvement of the native European CGMS
system. The European CGMS consists of three levels: (1) the acquisition of meteorological data and
their interpolation on a square grid of 25x25 km spatial resolution, (2) the simulation of crop
growth using the European WOFOST model, and (3) cereal yield forecasting based on outputs
derived from WOFOST model.
54
Such initiatives include the useful work undertaken by FAO for dissemination of data and tools, and knowledge
transfer for the processing of agrometeorological information, through the "Climpag" network
http://www.fao.org/nr/climpag/index_en.asp
Agrometeorological Cereal Yield Forecasting in Morocco
Institutionalization &
operationalization
Balaghi R., Jlibene M., Tychon B., Eerens H.
116
The forecasting system of Morocco, called "CGMS-MAROC", is developed in the framework of E-
AGRI project. The European institutions partners of E-AGRI project are: JRC, Alterra and VITO.
CGMS-MAROC is coordinated by INRA and jointly managed by DSS and DMN, in the framework of a
tripartite official agreement. It is intended to support DSS, as this administration is officially
mandated to provide crop statistics and forecasts. CGMS-MAROC is installed on a central server at
DMN, which is responsible for maintaining and delivering the climatic weather grid. CGMS-MAROC
is the first operational system for cereal yield forecasting in Morocco, institutionalized through a
strategic partnership in order to promote its development and sustainability.
The role of the national institutions in charge of CGMS-MAROC:
The National Institute for Agronomic Research (INRA) is responsible for:
o Acquisition and provision of agronomic data, for the calibration of the system at
Level 2 of CGMS-MAROC ;
o Contribution with DMN and DSS to the statistical analysis for cereal yield forecasting
at Level 3 of CGMS-MAROC ;
o Data analysis of remote sensing, to forecast crop yields at Level 3 of CGMS-MAROC.
The Direction of Strategy and Statistics (DSS) is responsible for:
o Acquisition and provision of data on official crop statistic ;
o Area estimates based on remote sensing and field surveys.
The National Direction of Meteorology (DMN) is responsible for:
o Acquisition and provision of weather data and hosting of CGMS-MAROC ;
o Interpolation of weather data, collected from synoptic weather stations.
CGMS-MAROC system has been improved compared to the native European CGMS system. CGMS-
MAROC has the same architecture than the European system but has been improved, since it
includes in addition to WOFOST simulation model, statistical models for forecasting crop yields
developed at INRA. Also, CGMS-MAROC runs with spatialized weather data on finer climate grids of
10x10 km resolution. CGMS-MAROC can currently forecast cereal yields at the level of the country,
agro-ecological zones and provinces. CGMS-MAROC is installed on a central server at DMN, which is
responsible for maintaining the system and updating and delivering the weather grid.
CGMS-MAROC is accompanied by a web interface (Figure 65) for visualization and preliminary
analysis of the cropping season, which was developed in Morocco in the framework of E-AGRI
project. The Web interface is hosted on a server located at INRA and supplied by daily climatic data
(rainfall, maximum and minimum temperature, reference evapotranspiration) coming from all
synoptic stations of DMN, as well as by satellite imagery products (NDVI, DMP) delivered every 10
Agrometeorological Cereal Yield Forecasting in Morocco
Institutionalization &
operationalization
Balaghi R., Jlibene M., Tychon B., Eerens H.
117
days by VITO. This Web interface allows thus a bio-climatic monitoring of the season at fine grid
cells of 10x10 km over all the agricultural areas of Morocco. Similarity analysis capabilities have
been implemented in this Web interface, allowing preliminary analysis of the cropping season and
rapid cereal forecasts.
Figure 65: The Web viewer of CGMS-MAROC (www.cgms-maroc.ma) for bio-climatic monitoring
of cereals, and preliminary statistical analysis of the cropping season. On the left side window is
displayed the cumulated rainfall over the season on agricultural lands (10x10 km grid). On the
right window is shown, the similarity analysis using cumulated rainfall, in the district of Ben
Ahmed (province of Settat).
Agrometeorological Cereal Yield Forecasting in Morocco
Conclusion & Outlook
Balaghi R., Jlibene M., Tychon B., Eerens H.
118
VIII. CONCLUSIONS AND OUTLOOK
The combined agrometeorological approach, for cereal yield forecasting developed at INRA in
collaboration with national and international research and development institutions, is described in
this document. It is a pragmatic and powerful approach, which is adapted to the different types of
available observed data: meteorological, agricultural, and satellite. It has a major advantage over
other approaches of crop yields forecast, which use models that require simulation data often not
readily available in Morocco. It also adapts to the spatial scale of the data (national, agro-ecological
zone, province).
The development of this combined approach began with the analysis of the Moroccan climate
which revealed the relevance of using amount and distribution of rainfall for cereal yield
forecasting. Analysis of cereal crops and their interactions with weather helped developing
agrometeorological indices that can be used to forecast yields. Such indices were derived from
climatic factors most limiting cereal production (rainfall, temperature, water balance), indices
derived from satellite imagery (NDVI), and from WOFOST simulation model. Indices derived from
rainfall and NDVI appeared to be more relevant and more practical for forecasting cereal yields in
Morocco. Indices derived from temperature are useful only in wet areas of the northern parts of
the country where rainfall is not a limiting factor.
This combined approach is currently implemented in a computerized system for cereal yield
forecasting, named "CGMS-MAROC." This system is designed in a modular way, to adapt to the
available data, to the spatial scale, and to desired accuracy and time of forecast delivery. Accuracy,
reliability, objectivity, traceability and forecast delivery time are important qualities that have been
targeted during the development of this system. The system is based on the combined approach
developed at INRA, in scientific and technical collaboration with the Luxembourg University
Foundation (now became Arlon Campus Environment of the University of Liège, ULg - Belgium)
since 2000 and JRC and VITO, since 2007. The forecasts are provided based on regression models
between official agricultural statistics and rainfall and NDVI during the growing season as well as
outputs of WOFOST simulation model. Bulletins on cereal yields forecasts have been published on
the websites of JRC (http://ec.europa.eu/dgs/jrc/index.cfm) and INRA (http://www.cgms-
maroc.ma/ newsbulletin.htm).
CGMS-MAROC system for cereal yield forecasting is currently the only institutionalized system in
Morocco providing published season's monitoring and operational cereal yields forecasts. To
sustain the system and allow further development, a formal consortium was established, composed
of INRA, DSS and DMN. CGMS-MAROC is installed on a central server at DMN. A Web viewer
(http://www.cgms-maroc.ma/) installed at INRA was specifically developed for checking data and
making preliminary analysis of the growing season. It is managed by a panel composed of members,
of recognized scientific and technical authority. INRA provides scientific support to the
Agrometeorological Cereal Yield Forecasting in Morocco
Conclusion & Outlook
Balaghi R., Jlibene M., Tychon B., Eerens H.
119
development of the system. DMN is responsible for updating and delivering spatial data of climatic
parameters on the weather grid provides covering the entire territory, and also manages the
modelling (WOFOST) aspects. DSS provides official data areas and grain yields. The first joint
bulletin for forecasting cereal yields was published in April 2011 on the website of the INRA website
(www.inra.org.ma).
Cereal yields are forecasted for different spatial resolutions: the country, the agro-ecological zones
and provinces. The country average cereal yields have been accurately forecasted for the growing
seasons of 2008-2009, 2009-2010, 2010-2011, and 2011-2012.
CGMS-MAROC system is currently being improved to overcome some accuracy limitations
encountered in some specific cases, for example:
Relatively high rainfall growing seasons but with severe drought in the middle of the crop
cycle like in 1996-1997 ;
High rainfall growing seasons, like those of 2008-2009 and 2010-2011 ;
Low temperature growing seasons like the 2011-2012.
These limitations of the system could be overcome by taking into account phenological stages of
the crop during periods of forecasting and during periods of drought, excess water or cold.
The prospects for this system are promising. Methods for forecasting yields can be developed for
other annual crops similar to cereals, since one of the modules of the WOFOST model provides
parameters for many other crops, particularly corn grain, rice, sugar beet, potato, bean, soybean,
rapeseed and sunflower.
CGMS-MAROC allows not only to instantly forecast grain yields two to three months before harvest,
but can also be adapted to other uses, through additional improvements, like drought insurance,
agricultural warning, mapping potential of land and land use, seasonal forecasting of crop yields
and impact of climate change on agricultural productivity.
CGMS-MAROC can be improved and used in agricultural early warning, by incorporating modules
for forecasting short-term weather events (rain, drought, heat waves and cold) and biotic hazards
(diseases, insects). These products can be generated directly from local weather stations, national
networks of weather stations or numerical models of spatial interpolation of climate data. This
implies a consorted effort in local data collection, transmission, processing and dissemination of
agrometeorological information to farmers and agricultural advisors.
The scope of agricultural warning is large, which may include in addition to the plant protection,
fertilization management, supplemental irrigation, and the choice of sowing date. This is a field of
research and development to develop in Morocco, which can have significant positive impact on
Agrometeorological Cereal Yield Forecasting in Morocco
Conclusion & Outlook
Balaghi R., Jlibene M., Tychon B., Eerens H.
120
agriculture and the environment. It is in this context that the National Health Security Office of
Food Products (in French Office National de Sécurité Sanitaire des Aliments, ONSSA) and DMN
included in their priorities the development of a national program on agricultural warning.
CGMS-MAROC can also be adapted for forecasting cereal yields in countries with similar climate to
that of Morocco, such as the Mediterranean countries. In these countries, behavior of cereals crops
to weather and crop management are similar.
Satellite imagery data, NDVI in particular, has been proven to be relevant in forecasting cereal
yields for Morocco. Based on this result, the possibility of estimating cereal areas using satellite
imagery data can be hypothesized.
Agrometeorological Cereal Yield Forecasting in Morocco
References
Balaghi R., Jlibene M., Tychon B., Eerens H.
121
IX. REFERENCES
Aït El Mekki, A., 2006. Les politiques céréalières au Maroc. Les notes d’analyse du CIHEAM N°7 –
mars 2006. Centre International de Hautes Etudes Agronomiques Méditerranéennes. Paris.
http://portail2.reseau-concept.net/Upload/ciheam/fichiers/NAN07.pdf
Allard, D.W., van Diepen, K., Boogaard, H., 2009. Regional crop yield forecasting: a probabilistic
approach. Geophysical Research Abstracts, Vol. 11, EGU2009-4820-1.
http://meetingorganizer.copernicus.org/EGU2009/EGU2009-4820-1.pdf
Allen, R., G., Periera, L., S., Raes, D., and Smith, M., 1998. Crop evapotranspiration - guidelines for
computing crop water requirements. FAO Irrigation and drainage 56, FAO - Food and
Agriculture Organization of the United Nations, Rome.
http://www.fao.org/docrep/X0490E/X0490E00.htm
Augustin, B., 1921. Le régime des pluies au Maroc. Contribution préliminaire à l'étude de la
pluviométrie du Maroc (avec deux planches et une carte en couleurs). "Mémoires de la
société des sciences naturelles du Maroc" ; "Tome I. Nº 1", 1921. Rabat : au siège de l'Institut
Scientifique Chérifien ; Paris: 11, rue Victor-Cousin (Ve) chez Émile Larose, Éditeur.
http://bibdigital.rjb.csic.es/ing/Libro.php?Libro=3403
Augustin, B., 1911. Le dry-farming et ses applications dans l'Afrique du Nord. In: Annales de
Géographie. t. 20, n°114. pp. 411-430.
http://www.persee.fr/web/revues/home/prescript/article/geo_0003-
4010_1911_num_20_114_3660
Bagnouls, F., Gaussen, H., 1957. Les climats biologiques et leur classification. In: Annales de
Géographie. t. 66, n°355. pp. 193-220.
http://www.persee.fr/web/revues/home/prescript/article/geo_0003-
4010_1957_num_66_355_18273
Bagnouls, F., Gaussen, H., 1953. Saison sèche et indice xérothermique. Bull. Soc. Rist. Nat.
Toulouse, 88 (3.4), 193·239.
http://www.persee.fr/web/revues/home/prescript/article/geoca_0035-
113X_1954_num_29_3_1980
Balaghi, R., Tahri, M., El Hareich, T., 2012. Bulletin de prévision de la récolte céréalière. Campagne
agricole 2011-2012. 3p. http://www.INRA.org.ma/docs/environ/bullprev1112fr.pdf
Balaghi, R., Badjeck, M.-C., Bakari, D., De Pauw, E., De Wit, A., Defourny, P., Donato, S., Gommes, R.,
Jlibene, M., Ravelo, A.C., Sivakumar, M.V.K., Telahigue, N., Tychon, B., 2010. Managing Climatic
Risks for Enhanced Food Security: Key Information Capabilities. World Climate Conference -
3. Procedia Environmental Sciences 1, 313–323
Balaghi, R., Tychon, B., Eerens, H., Jlibene, M., 2008. Empirical regression models using NDVI,
rainfall and temperature data for the early prediction of wheat grain yields in Morocco.
International Journal of Applied Earth Observation and Geo-information 10, 438 – 452.
doi:10.1016/j.jag.2006.12.001
Balaghi, R., 2006. Wheat grain yield forecasting models for food security in Morocco. University of
Liège, Belgium. Ph.D. Thesis, 103 p. http://aridoculture.googlepages.com/publications
Agrometeorological Cereal Yield Forecasting in Morocco
References
Balaghi R., Jlibene M., Tychon B., Eerens H.
122
Balaghi, R., Fadlaoui, A., Jlibene, M., 2005. Defining the spatial extent of the agricultural regions.
Development of an Integrated Natural Resources Management Framework for Durable
Agriculture in Central Morocco (ICARDA-INRM project). INRA, Rabat. 35p.
Balaghi, R., 2000. Suivi de l'évolution des réserves en eau sous culture de blé en région semi-aride
marocaine: calibration et utilisation du modèle "Soil". Fondation Universitaire
Luxembourgeoise - Belgique, 2000. Diplôme d’Études Approfondies en sciences de
l’environnement. http://riad.balaghi.googlepages.com/DEA.pdf
Balaghi R., Jlibene M., 1995. Diagnostic de la campagne agricole 1994-95. Centre Régional de la
Recherche Agronomique de Meknes.
Bennani, S., Ouattar, S., El Mourid, M., 1993. Simulation du rendement grain par le modèle SIMTAG.
Hommes, Terres et Eaux n°93. http://www.anafide.org/doc/HTE%2093/93-14.pdf
Benaouda, H. et Bouaziz, A., 1992. Stratégies de travail du sol pour une meilleure réussite du
peuplement pieds de depart et de rendement final : Cas de la région de Chaouia (Maroc).
Hommes, Terre et Eaux n°86. http://www.anafide.org/doc/HTE%2086/86-3.pdf
Bénichou P. et Le Breton O., 1987. Prise en compte de la topographie pour la cartographie des
champs pluviométriques statistiques. La Météorologie 7e série, 19, 23-34.
Boegh, E., M. Thorsen, M.B. Butts, S. Hansen, J.S. Christiansen, P. Abrahamsen, C.B. Hasager, N.O.
Jensen, P. van der Keur, J.C. Refsgaard, K. Schelde, H. Soegaard and A. Thomsen,
Incorporating remote sensing data in physically based distributed agro-hydrological
modelling. Journal of Hydrology. 287 (2004) 279.
Bontron, G., 2004. Prévision quantitative des précipitations : adaptation probabiliste par recherche
d'analogues - utilisation des réanalyses NCEP/NCAR et application aux précipitations du sud-
est de la France. http://www.lthe.fr/OHM-CV/Documents/theses/these_bontron.pdf
Boughlala, M., Balaghi, R., El Mourid, M., 1994. Validation et utilisation du modèle MULTISIM
simulateur des distributions de rendements. In : Anonyme, eds. Journée d'Information sur les
Recherches Agricoles. Outils d'aide à la décision et rentabilité de la recherche en agriculture
aléatoire. Rabat: INRA, 1994: 7-21.
Bouman, B.A.M. Crop modeling and remote sensing for yield prediction. Netherland’s Journal of
Agricultural Science. 43 (1995) 143.
Brohan, P., J.J. Kennedy, I. Harris, S.F.B Tett and P.D. Jones, 2006. Uncertainty estimates in regional
and global observed temperature changes: a new dataset from 1850. J. Geophys. Res., 111,
D12106, doi: 10.1029/ 2005JD006548.
Chaponniere, A. and V. Smakhtin, 2006. A review of climate change scenarios and preliminary
rainfall trend analysis in the Oum er Rbia Basin, Morocco. Working Paper 110 (Drought Series:
Paper 8) Colombo, Sri Lanka: International Water Management Institute (IWMI).
Confalonieri, R., Bregaglio, S., Cappelli, G., Francone, C., Carpani, M., Acutis, M., El Aydam, M.,
Niemeyer, S., Balaghi, R., Dong, Q., 2012. Wheat modelling in Morocco unexpectedly reveals
predominance of photosynthesis versus leaf area expansion plant traits. Agronomy for
Sustainable Development, in press.
Agrometeorological Cereal Yield Forecasting in Morocco
References
Balaghi R., Jlibene M., Tychon B., Eerens H.
123
Curnel, Y., Buffet, D., Oger, R., 2004. First functionalities of a statistical calibration toolbox for crop
yield forecasting. In: Development of an Agri-environmental Monitoring System for the
Chinese Province Heilongjiang. Final Report of a Sino-Belgian Cooperation Project. Brussels,
Belgian Science Policy.
De Martonne, E., Célérier, J., Charton A., 1924. Le massif Rehamna. In: Annales de Géographie. t.
33, n°183. pp. 244-256. http://www.persee.fr/web/revues/home/prescript/article/geo_0003-
4010_1924_num_33_183_9655
De Wit, A., Duveiller, G., Defourny, P., 2012. Estimating regional winter wheat yield with WOFOST
through the assimilation of green area index retrieved from MODIS observations. Agricultural
and Forest Meteorology 164, 39–52.
Debaeke, Ph., Aboudrare, A., 2004. Adaptation of crop management to water-limited
environments. Europ. J. of Agronomy 21. pp 433-446.
Decrem, M., Oger, R., Buffet, D., Gego, E., Ozer, P., Tychon, B., Weinzaepflen, E., Eerens, H., Piccard,
I. 2002. Questioning the importance of crop growth simulation for prediction of winter wheat
yields in Belgium. In: VII International Congress of the European Society for Agronomy,
Cordoba, Spain, 14-19/07/2002.
Doorenbos, J. and Kassam, A.H., 1979. Yield response to water. Irrigation and Drainage Paper n°. 33.
FAO, Rome, Italy, 193 p.
Doraiswamy, P.C. and P.W. Cook, 1995. Spring wheat yield assessment using NOAA AVHRR data.
Canadian Journal of Remote Sensing. 21 (1995) 43.
Douguédroit, A., Messaoudi, A., 1998. Essai de prédiction des rendements des céréales dans le
Maroc du Centre-Ouest. In: Méditerranée, Tome 88, 1-1998. Précipitations et cultures
céréalières dans le Centre-ouest du Maroc. pp. 45-49.
http://www.persee.fr/web/revues/home/prescript/article/medit_0025-
8296_1998_num_88_1_3032
Douguedroit, A., Durbiano, C., Messaoudi, A., Aït Hamza, M., 1998. Précipitations et rendements du
blé dur et de l'orge en culture «bour» dans le Maroc du Centre-ouest. In: Méditerranée, Tome
88, 1-1998. Précipitations et cultures céréalières dans le Centre-ouest du Maroc. pp. 39-44.
http://www.persee.fr/web/revues/home/prescript/article/ medit_0025-
8296_1998_num_88_1_3031
Driouech, F., 2010. Distribution des précipitations hivernales sur le Maroc dans le cadre d’un
changement climatique : descente d’échelle et incertitudes. Thèse de doctorat. Université de
Toulouse. France. 164p. http://ethesis.inp-toulouse.fr/archive/00001300/01/ driouech.pdf
Duchemin, B., Hadria R., Erraki, S., Boulet, G., Maisongrande, P., Chehbouni, A., Escadafal, R.,
Ezzahar, J., Hoedjes, J.C.B., Kharrou, M.H., Khabba, S., Mougenot, B., Olioso, A., Rodriguez,
J.C., Simonneaux, V., 2006. Monitoring wheat phenology and irrigation in Central Morocco: on
the use of relationships between evapotranspiration, crops coefficients, leaf area index and
remotely-sensed vegetation indices. Agricultural Water Management 79, 1-27.
El Aydam, M., Balaghi, R. MARS Agrometeorological Vol. 19 – Special Issue. Crop Monitoring in
Morocco. 2011. ftp://mars.jrc.ec.europa.eu/Bulletin/Europe/2011/MARSBULL
_2011_19_XX_special_Morocco.pdf
Agrometeorological Cereal Yield Forecasting in Morocco
References
Balaghi R., Jlibene M., Tychon B., Eerens H.
124
El Aydam, M., Baruth, B., Balaghi, R. MARS Agrometeorological Vol. 18 No. 7 - Crop Monitoring in
Morocco. 2010. http://mars.jrc.ec.europa.eu/mars/Bulletins-Publications/ MARS-
Agrometeorological-AGRI4CAST-INRA-Morocco-30-06-2010-Vol.18-No.-7
El Mourid, M., 1991. Jalons pour une agronomie des zones arides et semi-arides. Mémoire pour le
concours d'Ingénieur en Chef. INRA. Maroc.
Falkenmark, M., 1995. Coping with water scarcity under rapid population growth. Conference of
SADC Ministers, Pretoria 23-24 November 1995.
FAO, 2003. The FAO crop monitoring and forecasting approach (R. Gommes). In: Crop and
Rangeland Monitoring in Eastern Africa for Early Warning and Food Security (D. Rijks, F.
Rembold, T. Nègre, R. Gommes and M. Cherlet, eds). Proceedings of a JRC/FAO International
Workshop, Nairobi, 28–30 January 2003. Rome, FAO.
FAO, 2001. FAOCLIM 2.0: A World-Wide Agroclimatic Database. Food and Agriculture Organization
of the United Nations: Rome, Italy.
FAO, 1978. Report on the agro-ecological zones project. Vol. 1: Results for Africa. World Soil
Resources Report 48/1. FAO, Rome, 158pp.
Faurès, J.-M., Bernardi, M., Gommes, R., 2010. There Is No Such Thing as an Average: How Farmers
Manage Uncertainty Related to Climate and Other Factors, International. Journal of Water
Resources Development, 26: 4, 523-542. http://dx.doi.org/ 10.1080/07900627.2010.519515
Fischer, T., 1900. Regenkartc von Murroko, Zeitsch. d. Gesellsch. f. Erdkunde zu Berlin.
Gaussen, H., 1956. L'étude des climats par les courbes ombrothermiques : application à la
cartographie. In: L'information géographique. Volume 20 n°5. pp. 191-193.
http://www.persee.fr/web/revues/home/prescript/article/ingeo_0020-
0093_1956_num_20_5_1643
Genovese, G., Bettio, M. (eds)., 2004. Methodology of the MARS Crop Yield Forecasting System.
Volume 4. Statistical data collection, processing and analysis. In: European Communities (ISBN
92-894-8183-8)
Genovese, G., 2001. Introduction to the MARS Crop Yield Forecasting System (MCYFS). Meeting on
4 and 5 October 2001, Luxembourg. Space Applications Institute, Joint Research Centre of the
European Commission, Ispra, Italy, pp 15.
Genovese, G., Vignolles, C., Nègre, T., Passera, G., 2001. A methodology for a combined use of
normalised difference vegetation index and CORINE land cover data for crop yield monitoring
and forecasting. A case study on Spain. Agronomie 21: 91-111.
Gibbs, W.J. and Maher, J.V., 1967. Rainfall Deciles as Drought Indicators. Bulletin No. 48.
Melbourne, Bureau of Meteorology.
Glueck, M.F., Stockton, C.W., 2001. Reconstruction of the North Atlantic Oscillation, 1429-1983. Int.
J. Climatol., 21(12), 1453-1465.
Gobel, W., El Hairech, T., Moussadek, R., El Oumri, M., Douaik, A., Balaghi, R., Benaouda, H., El
Ouali, A., Mokssit, A., El Khatal, S., Sebbari, R., 2007. Atlas agro-climatique du Maroc.
Collection de 29 cartes. INRA et DMN, Rabat.
Agrometeorological Cereal Yield Forecasting in Morocco
References
Balaghi R., Jlibene M., Tychon B., Eerens H.
125
Gommes, R., Balaghi, R., Haripada, D., Mariani, L., Challinor, A., Tychon, B., Dawod, M.A.A. 2010.
Chapter 6: Agrometeorological forecasting. Guide to agricultural meteorological practices
n°134. Geneva, Switzerland, World Meteorological Organization. pp. 214-262.
http://www.wmo.int/pages/prog/wcp/agm/gamp/gamp_en.php
Gommes, R. 2006. Non-parametric crop yield forecasting, a didactic case study for Zimbabwe. In
Remote sensing support to crop yield forecast and area estimates. ISPRS Archives XXXVI-
8/W48 Workshop proceedings, page 79-84.
http://www.fao.org/nr/climpag/pub/non_parametric_yield_forecasting_FAO.pdf
Gommes, R. 2001. An Introduction to the art of agro-meteorological crop yield forecasting using
multiple regression. Report. Crop Monitoring and Forecasting Group Crop Yield Forecasting
and Agrometeorology Sub-Project, UTF/BGD/029, ASIRP/DAE, Dhaka.
Hadria, R., Duchemin, B., Lahrouni, A., Khabba, S., Er-Raki, S., Dedieu, G., Chehbouni, A., 2006.
Monitoring of irrigated wheat in a semi-arid climate using crop modeling and remote sensing
data: Impact of satellite revisit time frequency, IJRS, 27. 1093-1117.
Hanchane, M., 2009. Simulation de l’effet de la date de semis sur la satisfaction des besoins en eau
de l’orge par le modèle CERES en climat semi-aride marocain. Sécheresse 20, 357-359.
http://www.jle.com/e-docs/00/04/51/67/article.phtml
Hanchane, M., 1998. Estimation des risques climatiques en fonction de la date de semis de l'orge au
Maroc. In: Méditerranée. Tome 88, 1-1998. Précipitations et cultures céréalières dans le
Centre-ouest du Maroc. pp. 51-58.
http://www.persee.fr/web/revues/home/prescript/article/medit_0025-
8296_1998_num_88_1_3033
Herrera, R.G., Puyol, D.G., Martin, E.H., Presa, L.G., Rodriguez, P.R., 2001. Influence of the North
Atlantic Oscillation on the Canary Islands precipitation. J. Clim., 14(19), 3889-3903.
Hijmans, R.J., I.M. Guiking-Lens and C.A. van Diepen, 1994: WOFOST 6.0: User’s Guide for the
WOFOST 6.0 Crop Growth Simulation Model. Technical Document 12. Wageningen, DLO
Winand Staring Centre.
Hsiao, T.C., Heng, L.K., Steduto, P., Rojas-Lara, B., Raes, D., Fereres, E., 2009. AquaCrop—The FAO
Crop Model to Simulate Yield Response to Water: III. Parameterization and Testing for Maize.
Agron J. pp. 448–459
IIASA/FAO, 2012. Global Agro‐ecological Zones (GAEZ v3.0). IIASA, Laxenburg, Austria and FAO,
Rome, Italy. http://www.iiasa.ac.at/Research/LUC/GAEZv3.0/
IPCC, 2007. Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to
the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (S.
Solomon, D. Qin. M. Manning, Z. Chen, M. Marquis, K.B. Averyt, M. Tignor and H.L. Miller,
eds). Cambridge, Cambridge University Press.
Jlibene, M., 2011. Options génétiques d’adaptation du blé tendre au changement climatique :
variétés à résistance multiple (sécheresse, cécidomyie, septoriose, rouilles). Éditions INRA,
Maroc, 2011. 51 p.
Jlibene, M., 2009. Amélioration génétique du blé tendre au Maroc à l’aube du 21ème siècle. Edition
INRA, Maroc, 2009. 80 p.
Agrometeorological Cereal Yield Forecasting in Morocco
References
Balaghi R., Jlibene M., Tychon B., Eerens H.
126
Jlibene M., Balaghi R., 2009. Le risque sécheresse en agriculture pluviale. Bulletin mensuel de liaison
et d'information du Programme National de Transfert de Technologie n°181.
http://www.vulgarisation.net/bul181.htm
Jlibene, M. et Chafai, A. E., 2002. Variétés de blé adaptées au nord-ouest du Maroc. Bulletin de
Transfert de Technologie en Agriculture n°88. 4p. http://www.vulgarisation.net/bul88.htm
Jury, A., Dedebant, G., 1924. Études sur le Régime des Pluies au Maroc et Carte provisoire de la
Répartition des pluies. Mémoires de la Société des sciences naturelles du Maroc" ; "Nº IX.",
1924. Rabat : au siège de l'Institut Scientifique Chérifien ; Paris : 11, rue Victor-Cousin (Ve)
chez Émile Larose, Éditeur ; Londres : 44, great Russell Street chez Janson et Sons.
http://bibdigital.rjb.csic.es/spa/Libro.php?Libro=3423
Knippertz, P., Christoph, M., Speth, P., 2003. Long-term precipitation variability in Morocco and the
link to the large-scale circulation in recent and future climates. Meteorology and Atmospheric
Physics 83, 67-88.
Koch, E., 2011. Étude de faisabilité d'une assurance rendement basée sur indice climatique.
Université Paris-Dauphine, Mémoire d'actuariat. 126p. http://www.lynxial.fr/pda/FP/1226-
02.nsf/769998e0a65ea348c1257052003eb94f/48763f8ac4740188c12579d8006d8939/$FILE/
Memoire%20KOCH%20Erwan.pdf
Kogan, F.N., 2000. Contribution of Remote Sensing to Drought Early Warning. In: D.A.Wilhite,
M.V.K. Sivakumar, A.W. Deborah, (Editors). Early Warning Systems for Drought Preparedness
and Drought Management. World Meteorological Organization, Lisbon, pp. 86-100.
Kouadio A.L., 2007. Prévision de la production nationale d’arachide au Sénégal à partir du modèle
agrométéorologique AMS et du NDVI. DES Interuniversitaire En Gestion Des Risques Naturels.
Université de Liège. Belgique. http://www.hoefsloot.com/Downloads/TFE_Louis-Amani.pdf
Kutner, M.H., Nachtsheim, C.J., Neter, J., Li, W., 2005. Applied Linear Statistical Models: Fifth
edition New York: McGraw-Hill.
Lazar C., Baruth B., Micale F. 2009. Winter wheat yield estimation for Romania, based on
normalized difference vegetation index data available on MARSOP site. AN. I.N.C.D.A.
Fundulea, Vol. LXXVII, 2009. Fiziologia Plantelor. http://www.incda-
fundulea.ro/anale/77/77.25.pdf
Loup, J., 1957. Note sur l'évaporation au Maroc. In: Revue de géographie alpine. Tome 45 N°2. pp.
351-379. http://www.persee.fr/web/revues/home/prescript/article/rga_0035-
1121_1957_num_45_2_1901
Maas, S.J. Use of remotely-sensed information in agricultural crop growth models. Ecological
Modelling. 41 (1988) 247.
Mahyou H., Karrou M., Mimouni J., Mrabet R., El Mourid M., 2010. Drought risk assessment in
pasture arid Morocco through remote sensing. African Journal of Environmental Science &
Technology 4. 845-852.
MAPM, 2011. Situation de l'agriculture marocaine N°9. Ministère de l'Agriculture et de la Pêche
Maritime. Rabat, Maroc. Novembre 2011. 204 p.
http://www.marocagriculture.com/bibliotheque/bilan/Situation-de-l'Agriculture-
Marocaine.pdf
Agrometeorological Cereal Yield Forecasting in Morocco
References
Balaghi R., Jlibene M., Tychon B., Eerens H.
127
Martiny, N., Camberlin, P., Richard, Y., Philippon, N., 2006. Compared regimes of NDVI and rainfall
in semi-arid regions of Africa. International Journal of Remote Sensing, 27, pp. 5201–5223.
Maselli F., Romanelli S., Bottai L., Maracchi G., 2000. Processing of GAC NDVI data for yield
forecasting in the Sahelian region. International Journal of Remote Sensing, 21. 3509-3523.
Mayaux, P., Bartholomé, E., Fritz, S., Belward, A., 2004. A new land-cover map of Africa for the year
2000. Journal of Biogeography 31, 861-77.
Mereu, V., Iocola, I., Spano, D., Motroni, A., Balaghi, R., Duce, P., 2009. Analisi di Land Suitability per
la valutazione dell’impatto delle variazioni climatiche sulla coltivazione del frumento nella
provincia di Settat, Marocco. Italian Journal of Agrometeorology, anno 14 n. 2 giugno 2009,
pp. 44-45.
Mkhabela, MS., Bullock, P., Raj, S., Wang, S., Yang, Y., 2011. Crop yield forecasting on the Canadian
Prairies using MODIS NDVI data. Agricultural and Forest Meteorology, 151, 385-393.
Moricochi, L., R.R. Alfonsi, E.G. Oliveira and J.L.M. de Monteiro, 1995. Geadas e seca de 1994:
perspectivas do mercado cafeeiro. Informações Econômicas – Instituto de Economia Agrícola,
25(6):49–57.
Motroni, A., Canu, S., Iocola, I., Benaouda, H., Duce, P., 2009. Studi di impatto della variabilità
climatica sulla vocazionalità agricola nell’aerea mediterranea: il caso del Marocco. Italian
Journal of Agrometeorology, anno 14 n. 2 giugno 2009, pp. 12-13.
Narciso G., Balaghi R. MARS Bulletin Morocco - Morocco Joint Crop Bulletin - 09 06 2009 Vol. 1 No.
1. http://mars.jrc.ec.europa.eu/mars/Bulletins-Publications/MARS-Bulletin-Morocco-
Morocco-Joint-Crop-Bulletin-09-06-2009-Vol.-1-No.-1-English-Version
Neumann, K., M. Herold, A. Hartley, and C. Schmullius, 2007. Comparative assessment of
CORINE2000 and GLC2000: Spatial analysis of land cover data for Europe, Journal of Applied
Earth Observation and Geoinformation, 9, 425–437.
Nicholson S.E., Kim J., 1997. The relationship of the El Niño-Southern Oscillation to African rainfall.
Int. J. Climatol. 17:117–135.
Nicholson, S.E., 1986. The spatial coherence of African rainfall anomalies: Inter-hemispheric
teleconnections. J. Clim. Appl. Meteorol., 25, pp. 1365–1381.
ONICL, 2012. Importations des quatre céréales principales. Évolution de 1980 à 2012. Office
National interprofessionnel des céréales et des légumineuses. Maroc.
http://www.onicl.org.ma
Papy F., 1979. Analyse du comportement des cultures de blé dur et d'orge dans différentes régions
céréalières du Maroc à travers leur réaction aux variations interannuelles des régimes
pluviométriques. Acad. Agric. Fr., n° 3, pp.23 1-247.
Péguy, C.P., 1983. Informatique et climatologie. In: Annales de Géographie. t. 92, n°511. pp. 284-
304. http://www.persee.fr/web/revues/home/prescript/article/geo_0003-4010 1983_num
92_511_20189
Peterson, Thomas C. and Russell S. Vose, 1997. An Overview of the Global Historical Climatology
Network Temperature Data Base, Bulletin of the American Meteorological Society, 78, pp.
2837–2849.
Agrometeorological Cereal Yield Forecasting in Morocco
References
Balaghi R., Jlibene M., Tychon B., Eerens H.
128
Raes, D., Steduto, P., Hsiao, T.C., Fereres, E., 2009. AquaCrop—the FAO Crop Model to Simulate
Yield Response to Water: II. Main Algorithms and Software Description. Agron J. pp. 438–447.
Ramde, P. S. 2007. L’alerte précoce et la prévision des rendements agricoles au Burkina Faso: cas de
trois provinces Passore, Yatenga et Soum. Université de Liège.
http://www.memoireonline.com/03/11/4360/Lalerte-precoce-et-la-prevision-des-
rendements-agricoles-au-Burkina-Faso-cas-de-trois-province.html
Roux, G., Debrach, J., Gaussen, H., 1949. Précipitations Annuelles. Atlas du Maroc No. 4a. Comité de
Géographique du Maroc.
Roux, G., 1938. Troisième mémoire consacré à la Physique du Globe et à la Météorologie au Maroc.
Mémoires de la Société des sciences naturelles du Maroc" ; "Nº XLIV", 1938. Rabat : au siège
de l'Institut Scientifique Chérifien ; Paris : 11, rue Victor-Cousin (Ve) chez Émile Larose,
Éditeur ; Londres : 44, Great Russell Street chez Janson et Sons.
http://bibdigital.rjb.csic.es/spa/Libro.php?Libro=4344
Sebbar, A., Badri, W., Fougrach, H., Hsaine, M., Saloui, A., 2011. Étude de la variabilité du régime
pluviométrique au Maroc septentrional (1935-2004). Sécheresse 22. Pp. 139-48.
http://www.jle.com/e-docs/00/04/6C/61/vers_alt/VersionPDF.pdf
SEEE, 2010. Seconde Communication Nationale à la Convention Cadre des Nations Unies sur les
changements climatiques. Avril 2010. Secrétariat d’État chargé de l’Eau et de l’Environnement
(SEEE). Rabat, Maroc. 218p.
Seemann, J., Y.I. Chirkov, J. Lomas et B. Primault. 1979. Agro-meteorology. Springer-Verlag Berlin
Heidelberg, New York, Germany. 326p.
Sinclair, T.R., Seligman, N., 2000. Criteria for publishing papers on crop modelling. Field Crops
Research 68, pp. 165-172.
Skees, J., R., Gober, S., Varangis, P., Lester, R. and Kalavakonda, V., 2001. Developing Rainfall Based
Index Insurance in Morocco. World Bank Policy Research Working Paper 2577.
http://ideas.repec.org/p/wbk/wbrwps/2577.html
Snedecor, George W. and Cochran, William G., 1989. Statistical Methods, Eighth Edition, Iowa State
University Press.
Steduto, P., Hsiao, T.C., Raes, D., Fereres, E., 2009. AquaCrop—the FAO Crop Model to Simulate
Yield Response to Water: I. Concepts and Underlying Principles Agron. J 2009 (101): 426–437.
Stoppa, A., Hess, U., 2003. Design and Use of Weather Derivatives in Agricultural Policies: the Case
of Rainfall Index Insurance in Morocco. Contributed paper presented at the International
Conference Agricultural policy reform and the WTO: where are we heading? Capri (Italy), June
23-26, 2003. http://www.agriskmanagementforum.org/farmd/sites/
agriskmanagementforum.org/files/Documents/Design%20and%20Use%20of%20Weather%20
Derivatives%20Morocco.pdf
Supit, I., A.A. Hooijer and C.A. van Diepen (eds), 1994: System Description of the WOFOST 6.0 Crop
Simulation Model Implemented in CGMS. Vol. 1. Theory and Algorithms. Agriculture Series.
Luxembourg, Office for Official Publications of the European Communities.
Agrometeorological Cereal Yield Forecasting in Morocco
References
Balaghi R., Jlibene M., Tychon B., Eerens H.
129
Taky, A., 2008. Maitrise des excès d'eau hivernaux et de l'irrigation et leurs conséquences sur la
productivité de la betterave sucrière dans le périmètre irrigué du Gharb (Maroc). Analyse
expérimentale et modélisation. Thèse de doctorat. Institut des Sciences et Industries du
Vivant et de l’Environnement (Agro Paris Tech). France. www.eau-
sirma.net/media/files/these_de_taky
Tchuente, A.T.K., Roujean, J.L., De Jong, S.M., 2011. Comparison and relative quality assessment of
the GLC2000, GLOBCOVER, MODIS and ECOCLIMAP land cover data sets at the African
continental scale. International Journal Of Applied Earth Observation And Geoinformation,
13, 207-219. DOI: 10.1016/j.jag.2010.11.002
Van Diepen, C.A., J. Wolf, H. van Keulen and C. Rappoldt. 1989. WOFOST: a simulation model of
crop production. Soil Use and Management, 5(1):16–24.
Van Keulen, H. and H.H. van Laar, 1986. The relation between water use and crop production. In:
Modelling of Agricultural Production: Weather, Soils and Crops (H. van Keulen and J. Wolf,
eds). Simulation monographs. Wageningen, Pudoc.
Van Kraalingen, D.W.G., 1991: The FSE System for Crop Simulation. Simulation reports CABO-TT No.
23, CABO-DLO. Wageningen, Wageningen Agricultural University.
Van Oijen, M., 2002. On the use of specific publication criteria for papers on process-based
modelling in plant science. Field Crops Research 74, 197-205.
Varaldi-Conia A., 1953. Un exemple d'utilisation du milieu pour une étude sur l'érosion et ses
conséquences dans une zone aride. In: L'information géographique. Volume 17 n°4. pp. 148-
153. http://www.persee.fr/web/revues/home/prescript/article/ingeo_0020-
0093_1953_num_17_4_1299
Ward, M.N., Lamb, P.J., Portis, D.H., El Hamly, M., Sebbari, R., 1999. Climate Variability in Northern
Africa: Understanding Droughts in the Sahel and the Maghreb. In: Navarra A (ed), Beyond El
Niño: Decadal and Interdecadal Climate Variability. Springer Verlag, Berlin, pp. 119-14.
Watts, D., El Mourid, M., 1988. Rainfall patterns and probabilities in the semi-arid cereal production
region of Morocco. USAID Project No. 608-0136. Settat : Aridoculture Center INRA.
http://archives.cnd.hcp.ma/uploads/news/94-0207.pdf
Watson, D.J., 1947. Comparative physiological studies in growth of field crops. I. Variation in net
assimilation rate and leaf area between species and varieties, and within and between years.
Anatomy and Botany 11, 41–76.
WMO, 2010. Guide to Agricultural Meteorological Practices n°134. World Meteorological
Organization. Geneva, Switzerland. 799p.
http://www.wmo.int/pages/prog/wcp/agm/gamp/gamp_en.php
WMO, 2006. Commission for Agricultural Meteorology (CAgM). The first fifty years. WMO-No. 999,
Geneva, 44 pp. http://www.wamis.org/agm/pubs/brochures/wmo999e.pdf
Agrometeorological Cereal Yield Forecasting in Morocco
Authors
Balaghi R., Jlibene M., Tychon B., Eerens H.
130
Dr Riad BALAGHI
Dr Riad BALAGHI, graduated from the agronomy school « Institute of Agronomy and Veterinary
Hassan II » as an agricultural engineer in 1992. In 2006, he pursued his doctoral training, at the
University of Liège (Belgium), from which he was delivered a Ph.D. degree in Environmental
Sciences.
His joined the National Institute for Agronomic Research (INRA – Regional Centre of Meknes) as a
researcher, where he held responsibilities of coordinator of the Research Unit of Agronomy and
Crop Physiology, and later head of Environment and Natural Resources Department of INRA, Rabat.
His research carrier started as soon as he graduated in 1992 in the field of agro-climatology applied
to cereals at the “INRA-Meknes”. His technical and scientific national and international publications,
are mostly dealing with drought and means to attenuate climate impact on Moroccan agriculture.
Specific studies include: drought risk management, impact of drought on crop yields, and
forecasting crop productions, in addition to expected impact of climate change on future Moroccan
agriculture.
He has been a consultant for FAO in the areas of climate change and forecasting crops harvests.
Nationally, he has been helping to get the issue of climate change integrated within the projects of
the Moroccan Green Plan (Plan Maroc Vert). He coordinated national and international research
and development projects, and published the first specialized bulletins on cereal harvests
forecasting in Morocco. He contributed to the dissemination of scientific and technical information;
through his web site SAADA https://sites.google.com/site/aridoculture/.
Dr Riad BALAGHI was awarded for his work on climate change, the medal of merit by FAO in 2008.
He is associated researcher at the Royal Institute for Strategic Studies and member of the
International Society for Agricultural Meteorology and the Moroccan Association of Agro-economy.
Agrometeorological Cereal Yield Forecasting in Morocco
Authors
Balaghi R., Jlibene M., Tychon B., Eerens H.
131
Dr Mohammed JLIBENE
Dr Mohammed JLIBENE, holds a degree of Agricultural Engineer from the ‘Institut Agronomique et
Vétérinaire Hassan II’ in 1979, and a Ph.D. in Genetics and Agronomy from the University of
Missouri USA, in 1990.
He worked during his entire carrier for the National Institute for Agronomic Research (INRA),
starting 1979. He was head of the wheat breeding program at INRA, developed more than twenty
cultivars of soft wheat, which are largely adopted by Moroccan famers for their high productivity
and resistance to drought and diseases. « Marchouch », « Achtar », « Amal », « Rajae », « Mahdia »
et « Arrehane » are examples of highly successful cultivars which are currently largely grown in
Morocco, contributing to food security, combating drought and raising farmers revenues, in
addition to limiting cereal imports.
Dr. Jlibene, has also developed multidisciplinary research teams, and scientific laboratories, at
INRA-Rabat, -Settat and –Meknes, and managed them.
Dr Jlibene was awarded in 2009 the first Grand Prix Hassan II for invention and research in
agriculture, for his achievements in wheat breeding, specifically wheat cultivars resistant to drought
and high temperatures, and resistant to main parasites of wheat (septoriose, rusts, Hessian fly). His
interest in agrometeorology and breeding for drought resistance dates back to 1995.
He published more than a hundred publications (journal articles, technical documents, books and
book chapters), supervised degree training university students and non-degree trainees from
Morocco and abroad, advised farmers in cereal production. Refer to his web-site SAADA
https://sites.google.com/site/aridoculture/, for more information.
Agrometeorological Cereal Yield Forecasting in Morocco
Authors
Balaghi R., Jlibene M., Tychon B., Eerens H.
132
Dr Bernard TYCHON
Dr Bernard TYCHON, born in 1964, holds diploma of engineer in agronomy (Catholic University of
Louvain, 1987) and doctorate of science in Environment Sciences (Luxembourg University
Foundation, 1993). He has been assigned a teaching position in the Department of Sciences and
Environment Management at Liège University of Belgium.
He provides academic courses in agro-meteorology, data processing and modelling, remote sensing
and geographic information system, and diffuse pollution. He trained many students from
developing countries for DES, DEA, Master of Science, Complementary Master and Doctorate
degrees. He has published over a 100 scientific communications (books, research projects reports,
proceedings, national and international scientific revues.
For the last 15 years, he has been leading a team of research in agro-meteorology, studding the
relationship between agriculture and physical environment. More precisely, his main research
activities are concerned with identifying requirements and needs of crops under different latitudes
for water, solar radiation and temperature, in the purpose of forecasting agricultural production
levels. Outputs of theses research activities include development of operational tools to forecast
yields in many parts of the world (Belgium, Morocco, Ethiopia, Senegal, China,…), and conception
of an operational warning system on wheat diseases in Belgium and Luxembourg, and proposition
of an integrated management tool of water in agriculture at Burkina Faso. Dr Bernard TYCHON also
contributed training in these tools, in Niger, Pakistan, Bangladesh, Turkey, Afghanistan and Armenia
via programs of FAO.
He obtained the Outstanding Young Person Award of the Belgian Junior Economical Chamber in
2004, and he represents Belgium in European Society of Agronomy since 2004. He is also acting
President of College of Sciences Doctorate and Environment Management and co-responsible for
Complementary Master of Sciences and Environment Management in developing countries.
Agrometeorological Cereal Yield Forecasting in Morocco
Authors
Balaghi R., Jlibene M., Tychon B., Eerens H.
133
Mr Herman EERENS
Mr Herman EERENS holds a Master of Science in Agriculture and Applied Biology, from the
University of Leuven (Belgium) in 1980. He worked at the university for nearly 15 years, first in the
Department of Forestry and, from 1985, in the field of remote sensing.
His main activities are focused on research areas such as radiometry of crops and soils, modelling
the reflectance of crops, and atmospheric correction and classification of images of high spatial
resolution. In 2007, he joined the Flemish Institute for Technological Research (VITO, Belgium).
Since then, he has worked mainly on the satellite image processing of low spatial resolution images
provided by sensors from SPOT-VEGETATION, NOAA / METOP-AVHRR, MODIS and MSG and on
extracting useful information from these images to agricultural statistics.
Mr. Herman EERENS is currently working to provide the information needed for project Monitoring
Agricultural ResourceS (MARS) of the European Union as well as for projects in China and Africa. He
has developed software programs for processing satellite images such as Glimpse and Spirits in
these projects.
National Institute for Agronomic Research (INRA)
Division of Information and Communication
2013 edition
ISBN: 978 - 9954 - 0 - 6683 - 6
SUMMARY
The present document provides a summary of research work carried out, at National
Institute for Agronomic Research of Morocco (INRA), since early 1990s, in the area of
operational agrometeorology oriented toward forecasting crop harvests. Forecasting the
production of crops early before harvest allows decision makers to be prepared in advance for
eventual consequences of abnormal deviations of the climate, particularly for strategic
commodity crops to food security like cereals. To our knowledge, to date there is no official
method to forecast cereal production in Morocco on the basis of agrometeorological data.
However, cereal productions are estimated based on a sampling method some weeks before
harvest, every year by the Ministry of Agriculture and Marine Fishery (MAPM) through the
Direction of Strategy and Statistics (DSS). It is a direct method, precise, and applied directly
before harvest, but requires consequent human and financial resources. The need to
elaborate an indirect method to early forecast yields that is fast and economical, has been
understood at INRA as early as in 1995, triggered by the severe drought of that particular
season, described as the worst dry season of the 20th century in Morocco. Neither the
classical frequency analyses of the climate used to identify seasons of close similarity to 1994-
1995 season, nor the available mechanistic models for crop forecasting used in developed
countries, have been able to monitor crop development during that season and a fortiori
predict the catastrophic harvest of 1995. Therefore, it became necessary to come up with a
new approach for forecasting cereal yields using an innovative methodology which combines
empirical and statistical approaches with agronomic and meteorological expertise. First we
had to study the interaction between the climate and the cereal crops behaviors, particularly
climatic and crop cycles were analyzed together in a series of long term data, initially for
Meknes region where the first two authors were posted, extended later to other regions of
Morocco. Preliminary results indicated for the first time in Morocco that inter-annual
variation of cereals yields could be explained by variation in the amount of rainfall cumulated
during the crop cycle, with a relatively high accuracy. The relationship could be enhanced by
partitioning the season into three or more phases. In collaboration with the University of
Liège (ULg, Belgium) and later with the Joint Research Centre of the European Commission
(JRC), a new indicator was identified as highly correlated to cereal yields, which is the
Normalized Difference Vegetation Index (NDVI) derived from satellite images. Unlike many
European countries, this index was highly correlated to cereal yields in Morocco, mainly due
to the aridity of Moroccan climate and the predominating coverage of cereals of agricultural
areas. NDVI is correlated with cereal yields as long as cropping season rainfall did not exceed
550 mm, which explains the irrelevance of NDVI to forecast crop yields in Northern Europe.
The combination of both rainfall and NDVI allowed forecasting of cereal yields as early as
March, three months before harvest, and at a low cost, with a level of accuracy similar to the
one of the direct sampling method used at crop maturity by DSS. These astonishing results
have led INRA to publish for the first time in Morocco three crop forecasting bulletins
between 2009 and 2011, in collaboration with JRC. In these bulletins, an approach combining
four individual approaches was used: (1) similarity approach using rainfall and/or NDVI as
criteria of comparison, (2) regression models using rainfall and NDVI as predictors of cereal
yields, and (3) the JRC approach which is based on a simulation model of crop growth called
WOFOST. The deterministic model WOFOST is now being adapted to the Moroccan agro-
climatic context and incorporated in an operational forecasting system. To ensure durability
of the system, a strategic partnership between INRA, DSS and DMN was formalized in addition
to that bounding INRA and JRC. This new collaboration has allowed establishment of the first
national cereal yields forecasting system named “CGMS-MAROC”, based on the combined
approach developed in the present document. The system is carried out by the three national
institutions (INRA, DSS and DMN), leading to the edition of a fourth bulletin of cereal yields
forecasts issued for the 2012 season. The combined approach can be extended to forecast
yields for other crops in morocco as well as in countries of similar climatic pattern, provided
some adjustments. In parallel to yield forecasting, a new field of research can be explored,
dealing with estimating cropped areas, using low resolution and inexpensive satellite images.
Key words: Agrometeorology, yield forecasting, similarity analysis, NDVI, rainfall,
temperature, cereals, soft wheat, durum wheat, barley, drought, Morocco, INRA.
