Content uploaded by Mohammad Ali Sharifi
Author content
All content in this area was uploaded by Mohammad Ali Sharifi on Nov 22, 2015
Content may be subject to copyright.
Content uploaded by W.G.M. Bastiaanssen
Author content
All content in this area was uploaded by W.G.M. Bastiaanssen on Apr 06, 2015
Content may be subject to copyright.
DEVELOPMENT OF A DECISION SUPPORT SYSTEM FOR NATURAL DAMAGE
ASSESSMENT BASED ON REMOTE SENSING AND BIO-PHYSICAL MODELS
M.A. Sharifi
a*
, W.G.M. Bastiaanssen
b
, S.J. Zwart
b
a
ITC, P.O. Box 6, 7500 AA, Enschede, The Netherlands - alisharifi@itc.nl
b
WaterWatch, Generaal Foulkesweg 28, 6703 BS Wageningen, The Netherlands - (w.bastiaanssen, s.zwart)@waterwatch.nl
KEY WORDS: Hazard damage assessment, Data fusion, SEBAL, Yield assessment, Risk management
ABSTRACT:
Recent developments in aerospace survey technology, modelling of crop production processes, and geographic information systems
have created promising opportunities for agricultural resource management. One of the potential fields of application is in the
agricultural risk management, in particular in assessing the impacts of natural hazards on agricultural productions. This seems to be
the bottleneck and one of the critical issues in agricultural risk management everywhere in the world, especially in Iran. This paper
presents the findings of a joint research program between a number of local and international organizations to consider the state of
the art development in remote sensing technology, biophysical science, the local infrastructure and to develop an appropriate model
for natural hazard impact assessment on the agricultural production.
*
Corresponding author
INTRODUCTION
Agricultural crop insurance has been used as one of the major
supporting policies for agricultural development in many
countries, as it tends to reduce farmer’s risk and protect them
against production failure due to natural hazards. Risk is
defined as the possibility of meeting danger or suffering loss as
results of an undesired event in the future, and insurance is the
business of transferring risk that is based on uncertainty and the
effect of an adverse or harmful event (SSYS Consulting, 2000).
Organizations and individuals learn to mitigate risk and provide
for risk in different ways. One way to mitigate the financial
implications of a future event is to transfer the liability of
potential losses arising from the event to another party, through
insurance.
Agricultural insurance therefore, is often characterized by moral
hazard and high administrative costs, due, in part, to the risk
classification and monitoring systems those insurers must put in
place to forestall asymmetric information problems. Other costs
include acquiring the data needed to establish accurate premium
rates and conducting claims adjustments. As a percentage of the
premium, the smaller the policy, typically, the larger the
administrative costs (World Bank 2005).
In this context the Iranian Ministry of Agriculture has lunched
an insurance policy to cover farmers against natural disasters
and promote agricultural development. The policy is covering
the production losses of the major agricultural productions due
to flood, frost, hails, intensive rainfall and droughts. The Iranian
Insurance Agency who is responsible for the implementation of
the policy is facing serious challenges regarding the overall
costs, low financial performance (under 20), establishment of
proper rates, assessments of damages and handling the related
claims. To improve the situation a joint research program
between a number of international and local organizations was
formulated to develop an appropriate model for natural hazard
impact assessment on the agricultural production. This research
considers the state of the art development in remote sensing
technology, biophysical science, with the existing local
infrastructure. The main international organizations have been
ITC, Wageningen University, WaterWatch consulting firm in
the Netherlands, the International Rice Research Institute in
Philippines and from Iran the main contributing organizations
were the Agricultural Insurance Funding Agency, Iranian Rice
Research Institutes and the Iranian meteorological organization.
In this context, a prototype model for natural hazard impact
assessment on rice production in Sumea-sara Township in Gilan
province (north of Iran) has been designed, developed, and
evaluated for the growing period of 2006. The study area has
typically local (strong flavour; low yield; good price) and high
yielding rice varieties (no flavour; high yield; moderate price).
A wide range of crop yields can thus be expected. This paper
briefly presents the finding of the research.
METHODOLOGY
To assess the impacts of natural hazards on rice yield
production, a yield model which simulates the biomass
production of the rice in the course of growing period was
developed and applied. The model is making use of high and
low resolution temporal satellite imagery which is collected in
the course of the growing period, the SEBAL model
(Bastiaanssen and Ali 2003; Zwart and Bastiaanssen, 2007) and
daily meteorological data. An important input in the SEBAL
models are standard meteorological data obtained from several
stations in the area. During the growing season these data were
collected, analyzed and interpreted. Thirteen cloud free
MODIS/Aqua images were selected (May 8 to September 1,
2006) and together with the instantaneous, average daily and
10-daily averages metrological data from Rasht weather station
were processed using SEBAL model. This process resulted in
maps of potential, actual and deficit evapotranspiration,
biomass production (total dry matter related to stem, leaves,
roots and grains), soil moisture and yield reduction caused by
extreme temperature, drought, vapour pressure deficit maps for
each decade.
The MODIS results have a resolution of 250 meter which is too
coarse to distinguish individual fields. Therefore the MODIS
products were upscaled to field level with SEBAL results of
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. 34, Part XXX
Rice yield
MODIS biomass
(250x
250
m
)
Final Biomass
(30*30)
Landsat biomass
(30
x
30
m
)
Field assessment
ET-deficit
……….
MODIS
Anomaly
maps
Meteorology Field measurement
Landsatt
Damage map Yield maps
Rice Map
Damage
Assessment
Biomass
assessment
Figure 1. modules (each module in one colour) of the Damage Assessment System.
high resolution Landsat images (30 meter), which is suitable for
detecting rice at field scale. The upscaled model results were
calibrated using field measurements of dry matter production
and yield in the pilot area in more than 200 locations.
To assess the impacts of natural hazard on yield two different
methods, one based on biophysical and the other based on
empirical relations were designed and applied. The biophysical
approach tries to detect the impact of natural hazards on rice
yields, by estimating yield losses due to meteorological
conditions that temper photosynthesis. Yield losses or yield gap
is defined as the difference between a hypothetical potential
yield and the actual yield calculated by SEBAL model. The
impact of low and high temperature extremes, vapour pressure
deficit and water shortage was quantified during growing
period. The remaining part of the yield loss that could not be
explained was caused by factors that cannot be directly
measured with remote sensing and crop growth modelling.
These include farm management, hail and rain storms, soil
organic matter, amongst others. The impact of these events was
detected using a time series of NDVI maps that were derived
from MODIS satellite images acquired throughout the growing
season. A fitting model was established on a pixel by pixel basis
to establish the rice growing period. Anomalies from this model
were mapped and related to the events that could have had
impact on the yield production (rain events, hail, frost, flooding,
etc.).
The empirical approach compares the derived yield or biomass
at each point in time with its corresponding yield or biomass
average of the same period (point in time) at village, sub-district
and township level. In this approach, it is assumed that a natural
hazard will not only affect one field but will affect a larger area;
therefore the yield per point is compared with the averages of its
neighbourhood to assess the impacts of the natural hazards. In
practice the combination of the two is more fruitful. Finally a
Damage Assessment System (DAS) was developed to organize
input maps and images, visualization, and facilitating the
analysis (Sharifi 2007).
SYSTEM COMPONENTS AND IMPLEMENTATION
The rice damage assessment system as designed and developed
is composed of several modules as follows (Figure 1):
• Rice land cover map, based on temporal Landsat data
• Biomass assessment, based on 10-daily MODIS biomass
production maps, and monthly Landsat biomass production
maps, measuring growing period per pixel based on time
series of MODIS NDVI time series, and assessing biomass
on 30 meter resolution , The SEBAL model was applied to
obtain biomass production maps..
• Rice yield assessment, converting the biomass map into
the actual rice yield. The conversion is carried out through
harvest index which is derived from field measurements
• Natural damage assessment, combing the derived layers
and provide potential damage extent and intensities
The main objective of the DAS was to measure the extent, and
intensity of natural hazards and their impact on the obtained
rice yields. In fact the system should help verifying the
existence of any natural hazard, and measuring its extent and
impact on the rice yield per farm, so that it could be used by
insurance company to process the damage claims that is coming
from farmers. This means that the result has immediate legal,
social and economical impacts and therefore the result has to be
reliable and accurate. To derive this product, the DAS will
basically monitor the rice production process at each point
(pixel) from cultivation to harvesting date. In this process large
amounts of information (over 700 map layers) is generated that
have their own specific applications. The generated information
includes:
− Accumulated values of biomass production per decade.
− Accumulated values of actual and potential evapotranspira-
tion and evapotranspiration deficit (defined as potential mi-
nus actual evapotranspiration as an indicator of drought) per
decade.
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. 34, Part XXX
stations, the daily and monthly precipitation products from
the Tropical Rainfall Measurement Mission (TRMM) were
analyzed and tested for accuracy and suitability for the
Damage Assessment System “DAS”.
− In the following section some of the main products are
briefly presented.
− The average biomass production (Figure 2) for the entire
period was 13.4 tons per hectare (between 10-17). This was
based on the assumptions that all fields were cultivated first
May and harvested on august 31 which of course are not
true. To correct for this an automated procedure was
developed to determine the start, end and length of the rice
growing period for each pixel using a time series of 18
NDVI maps between May and September WaterWatch
(2006). The results were then used for calculation of the
seasonal biomass production. In the same process maps of
the sudden changes of NDVI “anomaly” in the growing
season were also generated. At the next stage the MODIS
250 meter biomass was up scaled into a high resolution 30
meter paddy rice yield map. This was carried out in three
stages:
− Compute accumulated MODIS biomass production for three
periods by incorporating the growing season start and end.
− Upscale the accumulated biomass production map of step 1
with three Landsat high resolution relative biomass produc-
tion maps from May 30, July 17 and August 2.
− Calibrate the high resolution seasonal biomass production
map with field measurements of paddy yield.
− Biomass was converted to yield through Harvest Index. Har-
vesting index was established based on measurements of
biomass and yields of 224 (1 meter) field samples. An aver-
age harvest index of 0.45 was measured with 14% moisture.
Since this index did not include under ground biomass, a
new harvest index was calculated considering the averages
field measurements at 224 points (3.7 ton) and the estimated
average SEBAL biomass (11.9 ton). The new harvest index
(0.31), was then applied to convert the SEBAL high resolu-
tion biomass map into fresh paddy yield at 14% moisture
(Figure 3). For validation, the model results in 59 fields were
compared with the actual field data which were collected by
the ministry of Agriculture. The results are presented in table
1 and Figure 4.
−
y = 0.648x + 1.2271
R
2
= 0.24
0.0
1.0
2.0
3.0
4.0
5.0
0.0 1.0 2.0 3.0 4.0 5.0
Modele d yield
Measured yield
− Figure 4. Modelled against measured yield in 59 sample
points. The one to one line is presented for comparison.
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
May-1
May-2
May-3
June-1
June-2
June-3
July-1
July-2
July-3
Augus-1
Augus-2
Augus-3
September-1
10-daily biomass production (ton ha
-1
)
0
2
4
6
8
10
12
14
16
cumulative biomass production (ton ha
-1
)
10-daily
cumulative
Poly. (10-daily)
Figure 2. Average ten-daily and cumulative (May 1, September 10
th
) biomass production for rice in the pilot area.
Figure 3. Rice yield map at 30 meter resolution.
Figure 4. Potentially affected areas.
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. 34, Part XXX
− Yield reduction caused by extreme temperature, drought, and
vapour pressure deficit per decade.
− Average soil moisture for each decade.
− Crop growing period (cultivation, harvesting dates per pixel)
and accumulated biomass production.
− Seasonal accumulated ET and water productivity maps.
− Trend analysis of all the above indicators.
− Land use/cover and anomaly maps.
− Meteorological grids of 1 km resolution for standard weather
parameters (air temperature, relative humidity, cloud cover
and wind speed) for each day in the growing season.
− Precipitation maps with daily intensities. As an alternative
for measured rainfall from meteorological stations, the daily
and monthly precipitation products from the Tropical
Rainfall Measurement Mission (TRMM) were analyzed and
tested for accuracy.
RESULTS
Thirteen cloud free MOIDS/Aqua images were selected (May
8
th
to September 1, 2006) and together with the instantaneous,
average daily and 10-daily averages metrological data from
Rasht weather station were processed using SEBAL model. The
process resulted in potential, actual and deficit ET, biomass
production (total dry matters related to stem, leaves, roots and
grains), soil moisture and yield reduction caused by extreme
temperature, draught, vapour pressure deficit maps for each
decades. The average biomass production (Figure 2), for the
entire period was 13.4 tons per hectare (between 10-17). This
was based on the assumptions that all fields were cultivated first
May and harvested on August 31 which does not count for all
fields. To correct for this an automated procedure was
developed to determine the start, end and length of the rice
growing period for each pixel using a time series of 18 NDVI
maps between May and September (WaterWatch, 2006). The
results were then used for calculation of the seasonal biomass
production. In the same process maps of the sudden changes of
NDVI “anomaly” in the growing season were also generated. At
the next step the MODIS 250 meter biomass was up scaled into
a high resolution 30 meter paddy rice yield map. This was
carried out in three stages:
− Compute accumulated MODIS biomass production for three
periods by incorporating the growing season start and end.
− Upscale the accumulated biomass production map of step 1
with three Landsat high resolution relative biomass produc-
tion maps from May 30, July 17 and August 2.
− Calibrate the high resolution seasonal biomass production
map with field measurements of paddy yield.
Biomass was converted to yield through the harvest index (HI),
which was established based on measurements of biomass and
yields of 224 (1 by 1 meter) field samples. An average harvest
index of 0.45 was measured with 14% moisture. Since this in-
dex did not include under ground biomass, a new harvest index
was calculated considering the averages field measurements at
224 points (3.7 ton) and the estimated average SEBAL biomass
(11.9 ton). The new harvest index (0.31), was then applied to
convert the SEBAL high resolution biomass map into fresh
paddy yield at 14% moisture (Figure 3). For validation, the
model results in 59 fields were compared with the actual field
data which were collected by the ministry of Agriculture. The
results are presented in table 1 and Figure 5.
As it can be seen from the table and figure the regional accuracy
of the model defined by the mean yield difference of the 59
fields is around 0.04 tone per ha (in the order of %1) and the
point accuracy defined by RMSE is 0.402 kg/ha with a CV of
around 11%, which are satisfactory results. The regression
coefficient (r
2
) between the filed measurements and the model
yield equals 0.49, which in comparison to the critical value of
0.330 for the 99% confidence interval, shows a significant
relation between the two values.
To support damage assessment, a potential damage map was
produced. This map is resulted from combination of yield, yield
gap (potential yield – the reductions caused by temperature, va-
pour pressure deficit and draught), and anomaly maps (Figure
4). This map together with all the other products was integrated
into a DAS which can support the insurance company to verify
and process the claims coming from framers. The system helps
assessors to compare the yield at each point with its average at
village, district and township levels. It further allows detailed
study of the behaviour of various indicators affecting the yield
in the course of growing season (Figure 6).
Table 1. Comparison of field and model results for several
villages in the pilot area.
field reported
yield (ton/ha)
modelled yield
(ton/ha)
difference
Mean 3.57 3.61 -0.04
Median 3.6 3.59 0.01
Mode 3.5 3.3 0.20
Stand.deviation 0.44 0.33 0.11
Minimum 2.4 2.8 -0.38
Maximum 4.3 4.7 -0.40
Count 59 59
Mean bias error of model versus field
results -0.04
Variation of errors with respect to mean
field results (CV) 0.11
Root mean square error of model results
(RMSE) 0.40
y = 0.648x + 1.2271
R2 = 0.24
0.0
1.0
2.0
3.0
4.0
5.0
0.0 1.0 2.0 3.0 4.0 5.0
Modele d yield
Measured yield
Figure 5. Modelled against measured yield in 59 sample points.
The one to one line is presented for comparison.
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. 34, Part XXX
CONCLUDING REMARKS
The results of this pilot study have demonstrated that the
combination of remote sensing and biophysical models can
provide valuable detailed and quantified data for agricultural
insurance banks. The introduction of these products into a
Damage Assessment System (DAS) has resulted in a powerful
tool to quickly assess damage to yields that were caused by
different natural hazards. The quality of actual and potential
yields maps was found to be good and the next steps will be to
apply the system for larger areas and to further test and improve
the DAS. This should finally result in an operational system that
supports agricultural insurance bank to quickly judge claims
from its clients.
REFERENCES
Bastiaanssen, W.G.M. and S. Ali, 2003. A new crop yield
forecasting model based on satellite measurements applied
across the Indus Basin, Pakistan. Agric. ,Ecol. Environ. 94, pp.
321-340.
ESYS Consulting, 2000. Earth Observation responses to Geo-
Information Market Drivers. pp 34
Sharifi, M.A. 2007. Development of a support system to assess
the impacts of natural hazard on rice cultivation: the case of
sume-sara, Gilan province, Iran. Project Report. pp 87
WaterWatch, 2006. Natural Hazard Damage Assessment of
rice yield in Gilan Province, Iran. Final report, pp. 44.
World Bank, 2005. Managing Agricultural Production Risk,
Innovations in Developing Countries. Washington DC, pp. 114.
Zwart, S.J., W.G.M. Bastiaanssen, 2004. Review of measured
crop water productivity values for irrigated wheat, rice, cotton
and maize. Agric. Water Manage. 69, pp. 115-133.
Zwart, S.J., W.G.M. Bastiaanssen, 2007. SEBAL for detecting
spatial variation of water productivity and scope for
improvement in eight irrigated wheat systems. Agric. Water
Manage. 89, pp. 287-296.
Figure 6. an output of DAS showing the trend of biomass production in the growing season as well its
comparison with its average at different administration levels.
