Predictability of water resources with global climate models. Case of Northern Tunisia

Abstract

The objective of the research is to explore the predictability of water resources directly with GCMs by analysing long-term effects of climate change on Northern Tunisia’s blue and green water. Hydrologic impacts rely on a rainfall-runoff lumped model using outputs of CMIP6 GCMs within the framework of the ssp2-45 scenario. Among the 30 CMIP6 models, the composite cnrm-esm2-1 and fgoals-g3 best restore observed runoff from 1995 to 2014 and give the best GCM. Hydrologic projections 2015–2100 show significant drops in rainfall (9%), runoff (21%), groundwater recharge (15%), as well as for green water (6%). The results show that the use of raw GCMs predictions on large basins is possible and provides precisions comparable to what is produced when using Regional Climate Models in medium size basins.

Full text

Comptes Rendus
Géoscience
Sciences de la Planète
Mustapha Besbes and Jamel Chahed
Predictability of water resources with global climate models. Case of Northern
Tunisia
Published online: 12 June 2023
https://doi.org/10.5802/crgeos.219
Part of Special Issue: Geo-hydrological Data & Models
Guest editors: Vazken Andréassian (INRAE, France),
Valérie Plagnes (Sorbonne Université, France), Craig Simmons (Flinders University, Australia)
and Pierre Ribstein (Sorbonne Université, France)
This article is licensed under the
Creative Commons Attribution 4.0International License.
http://creativecommons.org/licenses/by/4.0/
Les Comptes Rendus. Géoscience — Sciences de la Planète sont membres du
Centre Mersenne pour l’édition scientifique ouverte
www.centre-mersenne.org
e-ISSN : 1778-7025

Comptes Rendus
Géoscience — Sciences de la Planète
Published online: 12 June 2023
https://doi.org/10.5802/crgeos.219
Geo-hydrological Data & Models / GDM - Géo-hydrologie, données et modèles
Predictability of water resources with global climate
models. Case of Northern Tunisia
Mustapha Besbes aand Jamel Chahed ∗,b
aUniversity of Tunis El Manar and Tunisian Academy of Sciences, Letters and Arts,
Tunisia
bUniversity of Tunis El Manar, National Engineering School of Tunis, Tunisia
E-mails: besbesmustapha9@gmail.com (M. Besbes), jamel.chahed@enit.utm.tn
(J. Chahed)
Abstract. The objective of the research is to explore the predictability of water resources directly with
GCMs by analysing long-term effects of climate change on Northern Tunisia’s blue and green water.
Hydrologic impacts rely on a rainfall-runofflumped model using outputs of CMIP6 GCMs within
the framework of the ssp2-45 scenario. Among the 30 CMIP6 models, the composite cnrm-esm2-1
and fgoals-g3 best restore observed runofffrom 1995 to 2014 and give the best GCM. Hydrologic
projections 2015–2100 show significant drops in rainfall (9%), runoff(21%), groundwater recharge
(15%), as well as for green water (6%). The results show that the use of raw GCMs predictions on
large basins is possible and provides precisions comparable to what is produced when using Regional
Climate Models in medium size basins.
Keywords. Climate change, Global climate model, Earth system model, Tunisia, Blue water, Green
water.
Published online: 12 June 2023
“The effects of climate change during the 21st Century are relatively well predicted with respect
to temperature, but their hydrological effects are much less certain” [de Marsily, 2008].
1. General introduction
Tunisia uses 80% of its national blue water resource
for irrigation, which has almost reached its maxi-
mum water allocation. As for rain-fed agriculture, the
green water, it plays an essential role in national food
security: 65% of the national agricultural production
and 80% of agricultural exports. However, blue and
green water resources are still insufficient to meet
∗Corresponding author.
the population’s food needs, and Tunisia must import
foodstuffs, that contain virtual water [Allan, 1998, Re-
nault and Wallender, 2000, Hoekstra, 2003, Oki et al.,
2003]. Previous works on water prospects in Tunisia
have shown [Chahed et al., 2008, Besbes et al., 2014,
2019b] that it is necessary to evolve the water para-
digm towards a holistic vision that expands water re-
sources to include green and virtual water; and con-
sequently redefine the issue of Water Security in cli-
mate change context.
Global Climate Models (GCMs) or Global gen-
eral circulation models are currently the best tools
ISSN (electronic) : 1778-7025 https://comptes-rendus.academie- sciences.fr/geoscience/

2Mustapha Besbes and Jamel Chahed
to anticipate the future impacts of climate change.
Models with biophysical and biogeochemical pro-
cesses, called Earth System Models (ESMs), include
land and ocean carbon cycle, atmospheric chemistry,
and other biogeochemical cycles [Watanabe et al.,
2011, Collins et al., 2011]. Climate Change effects
[Collins et al., 2011, IPCC, 2021] are fairly well pre-
dicted as far as temperature is concerned. But GCMs
have some difficulty in accurately estimating pre-
cipitation, a highly heterogeneous, non-smooth, and
spatially discontinuous field [Hughes and Guttorp,
1994, Mandal et al., 2016]. However, to be accurate,
hydrological models, which transform precipitation
into runoff, need unbiased precipitation data [Man-
dal et al., 2016, Maraun and Widmann, 2018, Bruyère
et al., 2014].
Much work has recently been devoted to pro-
ducing aggregated predictions obtained by applying
weightings to the results of a given ensemble of GCMs
[McSweeney et al., 2015, Knutti et al., 2017, Herger
et al., 2018, Laurent et al., 2021, Zhang et al., 2022].
Ensembles improve the results by smoothing out the
uncertainties and biases of single models’ predic-
tions [Wang et al., 2021]. The other way to improve
climate models’ predictions is sought by increasing
GCMs resolution or downscaling to build Regional
Climate Models (RCMs). Nevertheless, Somot et al.
[2018] argue that the reference datasets used for the
evaluation of GCMs are often not suitable for RCMs,
and some are run using specific higher-resolution
data [Fathalli et al., 2019, Somot et al., 2018, Zhang
et al., 2022]. Kim et al. [2020] and Cos et al. [2022] in-
dicate that climate model errors propagate through
the RCMs to become a fundamental source of un-
certainty. The issue of climate models’ uncertain-
ties is especially concerning for impact studies at
the watershed scale, for which unbiased precipita-
tion is needed at small grids, requiring the imple-
mentation of RCMs. Many bias reduction methods
have been applied in order to make climate models’
outputs suitable for impact studies [Foughali et al.,
2015, Dakhlaoui et al., 2022, Switanek et al., 2022].
Studies of climate change impacts with RCMs
have been developed on medium-scale catchments
in the Mediterranean basin [Deidda et al., 2013,
Terink et al., 2013]. Some studies use RCMs outputs
to run hydrological models and simulate observed
runoff. In studies on Northern Tunisia, Ensemble
RCMs provide inputs of hydrological models (tem-
perature and precipitation): four RCMs from the Eu-
ropean program ENSEMBLES [Foughali et al., 2015]
and eleven RCMs from the Euro-Cordex Project
[Dakhlaoui et al., 2022]. Otherwise, GCMs outputs
are also used directly in predicting environmen-
tal impacts, in particular on large areas or basins,
[Ramirez-Villegas et al., 2013, Farsani et al., 2019,
Wang et al., 2021, Shokouhifar et al., 2022, Hamed
et al., 2022, Li et al., 2022].
Despite uncertainties, Climate Models reveal sig-
nificant impacts on the Mediterranean basin which
would be among the most vulnerable regions in the
world, with on average a sharp decline in rainfall
and higher evapotranspiration due to the tempera-
ture increase. Countries on the southern and east-
ern shores are likely to be the most severely affected,
resulting in an increase in food dependence [Bes-
bes et al., 2010, de Marsily and Abarca-del Rio, 2016,
de Marsily, 2020]. Data analysis for Tunisia shows
an increase in mean temperature of 1.2 °C over the
past century; while for the same period, no signifi-
cant change in mean rainfall is detected even though
an increase in variability (highest standard deviation)
is observed for the latest period [King et al., 2007].
With growing concerns about impacts on the future
of water resources in Tunisia, more specific climate
change modeling and predictions are intended, to
prepare adaptation and remediation measures [AFD-
MA, 2021].
In the present research, we analyze the long-term
effects of climate change using the predictions from
CMIP6 on Northern Tunisia’swater resources, includ-
ing blue and green water. The region represents the
essential source of surface water, which gives it the
qualifier “water tower” of Tunisia. It is also the cereal
region of the country, mainly cultivated in rain-fed:
it is its “attic”. Based on hydrological modeling, the
analysis aims at determining the foreseeable climate
change effect on the overall water resources of the
northern region of Tunisia. The hydrologic model, a
lumped rainfall-runoffmodel, is first calibrated and
validated based on historical simulations, extensively
confronted with field data. The simulations carried
out by 30 GCMs that participated in the CMIP6 ex-
ercise provide annual chronological mean temper-
ature and precipitation series for each sub-region
(governorate). Among the various combinations of
radiative forcing defined in the CMIP6 experiences,
we consider the medium IPCC emission scenario

Mustapha Besbes and Jamel Chahed 3
SSP2-4-5, designed to prolong current trends. The
analysis of these simulations on the hydrological
model and comparison of their outputs to observed
runoffdata will lead to selecting the GCMs, the re-
sults of which fit best with observations. The cou-
ple “selected GCM-hydrologic model” is then applied
in a prospective projection (2015–2100), where GCM
outputs (temperature, precipitation) serve as inputs
for the hydrologic Model, the outputs of which are
surface runoff, aquifer recharge, and actual evapo-
transpiration.
GCM-hydrologic model coupling is not exten-
sively used in the literature, considering that the bias
induced by large dimensions of GCMs meshes is
likely to introduce unacceptable uncertainties at the
catchment areas scale. However, the downscaling al-
ternative to GCMs usually comes with bias correc-
tion, the implementation of which is long and not
devoid of uncertainty. Our objective is to explore to
what extent, the direct GCM-hydrologic models’ cou-
pling, more consistent and more accessible to inter-
pretation, can be relevant for hydrological studies.
It is to answer the following question: is it possible
to predict Water Resources only with GCMs, without
downscaling, and what would be the resulting uncer-
tainties?
2. Northern Tunisia and its climate
2.1. Setting up the context: Tunisia
Tunisia has a surface area of 165,000 km2and a popu-
lation of 12 M (2020). 5 M ha are cultivated area, 11%
of which can be irrigated; 4.5 M ha are rangeland and
grassland; 1.6 M ha are forests and steppes; 4.7 M ha
are unproductive lands (wetlands, deserts, urban ar-
eas . . . ). Agricultural needs are covered by domestic
production at 70%; the country is a net importer of
food products. Due to random climatic conditions,
the level of cereal self-sufficiency ranges from 15 to
60% depending on annual precipitation. Tunisia has
average rainfall estimated at 37 km3/y. According to
Besbes et al. [2019b], the total hydraulic resources
(blue water) are 4.8 km3/y: 2.7 km3/y are surface
runoff, of which 85% runs offthe northern Tunisia
basin. In 2010, groundwater withdrawals were esti-
mated at 2 km3, representing an average operating
rate of 95% of renewable groundwater resources esti-
mated at 2.1 km3/y, and many aquifer systems are se-
verely overexploited. The water resource of rain-fed
agriculture (green water), available for evaporation
and consumption by plants, related to arable land
(5 million ha), is estimated on average at 13 km3/y,
which increases to 19 km3/y when incorporating
rangelands. The total blue water withdrawals were
estimated at 3.0 km3in 2015, of which 21% by mu-
nicipalities, 76% by irrigation, and 3% by industries
[Besbes et al., 2019b].
Extending over seven degrees of latitude, Tunisia
offers a succession of hydro-climatic types: (a) hu-
mid and sub-humid in the North, (b) semi-arid and
arid in central Tunisia, and (c) hyper-arid or Saha-
ran for the entire South. These climatic types are ar-
ticulated with the three major hydrological provinces
(Figure 1): In the North, the reliefs, oriented SW–NE,
frame the country’s most fertile plains and make this
area the wettest in the country. The main river, Oued
Medjerda, fully controlled by a series of large storage
dams, crosses the region from west to east to unclog
the North of the Tunis Gulf.
From North to South, this variety results in a range
of rainfall regimes from more than 1000 mm/y in the
North, up to 50 mm/y at the southern tip of the coun-
try, with an overall average of 250 mm/y. The annual
average temperature for the country is 19.5 °C, which
varies by region and season between 12 °C in winter
and 32 °C in summer.
2.2. The study area: hydro-meteorological series
and data sources
The study area, Northern Tunisia, forms an en-
semble of three large hydrologic basins: Med-
jerda, Far North—Ichkeul and Cap Bon—Miliane
(Figure 1), extending over 29,000 km2that is 18%
of the area of Tunisia. With an average rainfall esti-
mated at 510 mm/y, Northern Tunisia receives 40%
of the national average rainfall, provides more than
80% of surface water, 40% of groundwater resources,
and a large part of green water, used in particular
to grow cereals. The region covers 11 governorates
among the 24 in the country, namely those of Tunis,
Ariana, BenArous, Manouba, Bizerte, Nabeul, Beja,
Jendouba, Le Kef, Siliana, Zaghouan (Figure 2).
Monthly precipitation and temperature data at the
national level for the 24 Tunisian governorates cov-
ering the period 1901–2020 can be downloaded from
the Climate Change Knowledge Portal (CCKP) devel-
oped by the World Bank Group [WBG, 2022], which
data source is the Climatic Research Unit (CRU)

4Mustapha Besbes and Jamel Chahed
Figure 1. Tunisia, the three major hydrological
regions, with the Northern Tunisia study area
limit [from Besbes et al., 2019b].
gridded time series, presented on a 0.5° grid. These
monthly time series are developed by interpolation
of values derived from observational data on temper-
ature and rainfall in thousands of reference weather
stations worldwide [Harris et al., 2020]. For Tunisia
in particular, the original series of monthly rain-
fall depth are developed, criticized, and analyzed by
the Tunisian National Institute of Meteorology [INM,
2022]. These data have been reanalyzed and inte-
grated into the global networks respectively of NOAA
Figure 2. The governorates map of Tunisia.
in its global inventory [NOAA, 2022], and WMO
[WMO, 2022]. All these databases are available to the
public. Among the 7500 reference climate stations
the rainfall data of which are published by the WMO,
NOAA, and CRU, with historical series of more than
100 years, Tunisia has 15 rainfall reference stations,
8 of which cover (directly or in the immediate vicin-
ity) the Northern Tunisia basin over periods exceed-
ing one hundred years with monthly rainfall series.
Moreover, the European Centre of Medium-Range
Weather Forecasts (ECMRWF) provides the last anal-
ysis ERA5 of climate variables, covering the Earth on
a 30 km grid. ERA5, which combines global histori-
cal observations into estimates by improved model-
ing and data processing systems [Jiao et al., 2021], has

Mustapha Besbes and Jamel Chahed 5
Table 1. Statistical parameters of CRU and
ERA5 vs. DGRE precipitation series for the
study area
Source of precipitation
series (1970–2020)
CRU ERA5 DGRE
Mean value mm/y 686 504 512
Standard deviation mm/y 136 108 107
Correlation coefficient 0.43 0.44 1
Root mean square error
RMSE mm/y
217 110
Average relative bias 24.9% −1.4%
been shown to be among the best-performing reanal-
ysis products [Hersbach et al., 2020, Tarek et al., 2020,
Jiang et al., 2021]. For all governorates of Northern
Tunisia, the CCKP provides also historical series of
precipitation 1970–2020 labeled ERA5 [WBG, 2022].
Finally, the official Tunisian institution for water
resources management, the General Directorate of
Water Resources (DGRE, Ministry of Agriculture) has,
among other missions, to maintain and develop na-
tional hydro-meteorological networks for Precipita-
tion, Runoff, and Groundwater. The produced time
series are regularly reviewed and updated. The lat-
est update constitutes the main reference source for
our case study [DGRE, Prointec-Comete, 2019] and
will be considered as the historical reference ob-
servations series. However, these observations show
some missing values. In order to complete missing
data over 1970–2020, we use the global series CRU
and ERA5. Table 1 presents the statistical parameters
of these series in comparison with DGRE, which in-
dicate a better similarity between ERA5 and DGRE
reference observations, so that missing observations
could be filled from ERA5 series.
3. Historical trends and possible future scenar-
ios
3.1. Observed series and current trends
Northern Tunisia was subject to a significant increase
in temperature over the past 50 years, with average
increases of 0.5 °C per decade, a total increase of
2.5 °C since 1970, including over the northern re-
gions (Figure 3). Annual rainfall in northern Tunisia is
variable but its long-term trend has been stable since
1970 (Figure 3).
3.2. Possible climate future scenarios
The climate change community has established a
scenarios framework to facilitate integrated inter-
comparisons of GCMs projections. The new frame-
work adopted in CMIP6 combines the Representative
Concentration Pathways (RCPs) defined within the
CMIP5 with Shared Socioeconomic Pathways (SSPs)
in a Scenario Matrix Architecture [O’Neill et al., 2016,
Riahi et al., 2017].
In Figure 4, each cell of the matrix indicates a
combination of socioeconomic development path-
way (i.e., an SSP) and climate outcome based on
a particular forcing pathway. Dark blue cells indi-
cate scenarios that will serve as the basis for climate
model projections in Tier 1 of Scenario MIP; light
blue cells indicate scenarios in Tier 2 (additional sce-
narios). For example, the trend scenario SSP2 is com-
patible with the forcing scenario RCP4.5 (4.5 W/m2).
Scenario SSP2-4.5 represents the medium part of the
range of future forcing pathways and combines in-
termediate societal vulnerability with an intermedi-
ate forcing level as its land use and aerosol pathways
are not extreme relative to other SSPs.
The present research aims to explore the pre-
dictability of water resources using climate out-
puts derived from CMIP6 ESMs’ simulations related
to scenario SSP2-4.5. Within this scenario, social,
economic, and technological trends do not shift
markedly from historical patterns, which makes it a
dynamics-as-usual scenario. Projections assume a
slowdown in economic growth with average growth
rates in the second half of the century [Riahi et al.,
2017]. We will further investigate this scenario. In
doing so, we explore a reference case that should
allow an evaluation of the approach and open the
way to comparative explorations of various scenarios
[Fricko et al., 2017].
3.3. Climate change, climate models and projec-
tions
According to global projections, the temperature
will continue to increase in Tunisia throughout the
end of the century. A significant reduction in annual
precipitation, with a trend of increase in the intensity

6Mustapha Besbes and Jamel Chahed
Figure 3. Mean annual temperature (°C) and precipitation (mm/y) in Northern Tunisia (Average values
of Northern Tunisia obtained by weighting the governorates values according to their areas); data sources
are: DGRE, Prointec-Comete [2019] and WBG [2022].
Figure 4. SSP-RCP scenario matrix illustrating MIP (Model Intercomparison Project) simulations
[adapted from O’Neill et al., 2016].
of heavy rainfall events, is expected under high emis-
sions scenarios. This will affect the water resources of
the country, with the majority of projections indicat-
ing a progressive decrease in runoffand groundwater
recharge [Bargaoui et al., 2014, Döll, 2009, Oueslati
et al., 2012, Nasr et al., 2008, Schewe et al., 2014,
Slama et al., 2020].

Mustapha Besbes and Jamel Chahed 7
Figure 5. Observed and predicted annual Temperatures on Northern Tunisia for the CMIP6 active
Models, Scenario ssp2.45, period 1970–2100 [Data from WBG, 2022].
The study carried out by AFD-MA [2021] analyzes
the effects of RCP4.5 and RCP8.5 climate scenarios
on Tunisia’s water resources and food security. Pro-
jections on the 2050 and 2100 horizons indicate a
rise of the arid and semi-arid climatic stages towards
the North and a quasi-disappearance of the humid
stage by 2100. They indicate also a rise in winter tem-
peratures, which affects crop development stages
and yields, and increases water stress with a national
cereal production decrease between 16% and 38%
depending on the scenarios and horizons [AFD-MA,
2021].
The 2021 IPCC Report [IPCC, 2021] assesses
results from climate models participating in the
CMIP6, which include new and better representa-
tions of physical, chemical, and biological processes,
as well as higher resolution. Observed warming is
within the very likely range of the CMIP6 ensemble.
However, some differences from observations re-
main, for example in regional precipitation patterns.
Analysis of models’ outputs includes both historical
period and future projections, as well as predictabil-
ity of the climate system on various temporal and
spatial scales.
The models used in CCKP-CMIP6 compilation
are the following: ACCESS-CM2; ACCESS-ESM1-5;
AWI-CM-1-1-MR; BCC-CSM2-MR; CAMS-CSM1-0;
CANESM5; CESM2; CMCC_CM2-SR5; CMCC-ESM2;
CNRM-CM6-1; CNRM-ESM2-1; EC-EARTH3; EC-
EARTH3-VEG; FGOALS-G3; GFDL_ESM4; HADGEM3
-GC31-II; INM-CM4-8; INM-CM5-0; IPSL_CM6A_LR;
KACE-1-0-g; KIOST-ESM; MIROC-ES2I; MIROC6;
MPI_ESM1-2-HR; MPI-ESM1-2-LR; MRI-ESM2;
NESM3; NORESM2-LM; NORESM2-MM; TAIESM1;
UKESM1-0-II.
Specific projections of Temperatures and Precip-
itations for Northern Tunisia: CMIP6 historical sim-
ulations are performed over different periods from
1850 to 2014. Figures 5 and 6 present the hydro-
climatic outputs respectively temperatures and pre-
cipitation, resulting from the simulations relating
to the ssp2.45 scenario. These are driven from the
26 models whose outputs on Northern Tunisia are
exploitable and include retrospective simulations
over the historical reference period (1995–2014) and
prospective projections up to the end of the century
(2015–2100). The figures also show data on the ref-
erence period (1970–2020): the observed tempera-
tures (mostly reanalyzed ERA5 values) and the ob-
served precipitations (from DGRE database). We can
see that many models provide series with negative
bias vis-a-vis historical temperature and precipita-
tion series, which is well attested by the Ensemble
distribution (aggregated prediction obtained by ap-
plying weightings to the results of the 26 GCMs) and
in particular its median series (P50).

8Mustapha Besbes and Jamel Chahed
Figure 6. Observed and predicted annual Precipitation on Northern Tunisia for the CMIP6 active Models,
Scenario ssp2.45, period 1970–2100 [Data from WBG, 2022, DGRE, Prointec-Comete, 2019].
4. Modeling climate change and water re-
sources
4.1. A model to convert precipitation into actual
evapotranspiration, runoff, and groundwa-
ter recharge
From projections of temperature (T) and precipi-
tation (P) over the next decades developed by var-
ious global climate models, we propose, with a
hydrological model, to quantify the hydrological im-
pacts of Climate Changes on Northern Tunisia and
predict the future of blue and green water resources
at various horizons of the 21st century.
The hydrological model is a production function
type with two reservoirs: a balance reservoir and a
transfer one. System inputs are precipitation and po-
tential evapotranspiration; outputs are actual evap-
otranspiration, infiltration, and runoff. In semi-arid
regions such as Tunisia, this type of model is oper-
ational, tested, and validated with a daily time step,
but given the availability of climatic data series and
the lengths of periods to simulate, we had to work
on a monthly basis. On the water budget, such a time
step extension results in errors that are all the greater
as the climate aridity is higher [Besbes, 1978]. This
constraint led us to restrict our study area to the lim-
its of Northern Tunisia climatic region. Executed in
sequence, the set of equations governing the water
cycle in the model reservoirs (Figure 7) are, all quan-
tities expressed in monthly water heights:
Water balance reservoir:
(i) AET =Min (Rt−1+P; PET);
(ii) R=Min ((Rt−1+P−AET);Rmax)
Transfer reservoir:
(iii) F=Max ((Rt−1+P−AET −Rmax); θ);
(iv) I=Min (F,Imax);
(v) RU =F–I
where: AET is the Actual Evapo-Transpiration; tthe
time, here the month; Pthe precipitation; PET the
Potential Evapo-Transpiration; Rthe soil reserve;
Rmax the maximum soil reserve capacity; Fthe to-
tal Flow; Ithe infiltration (effective infiltration, or
groundwater recharge); Imax the maximum infiltra-
tion capacity; RU the runoff. The two parameters
Rmax and Imax need to be calibrated using the ob-
served data.
The Potential evapotranspiration is given, for tem-
perate and Mediterranean regions, by the monthly
TURC formula [Alexandris et al., 2008, Bonnet et al.,
1970, McMahon et al., 2013]:
PET =0.4 ×(T/(T+15)) ×(IG +50).
Tis the temperature in Celsius degrees.
IG is the global solar radiation in cal/cm2/day
(over the study region and on annual average,
IG varies from 350 cal/cm2/day in Bizerte to

Mustapha Besbes and Jamel Chahed 9
Figure 7. Schematic diagram of the hydrologi-
cal model.
410 cal/cm2/day in Siliana, with a maximum in
July and a minimum in January [ANME (National
Agency for Energy Management), 2023]).
The study area is limited to the 11 administrative
districts in the northernmost region of the country.
For each of these districts, the CCKP [WBG, 2022]
provides monthly historical series of temperature
and precipitation, which are transformed into the
corresponding series of infiltration (groundwater
recharge), runoff, and AET using the monthly hydro-
logical model. The monthly outputs of the model are
first integrated and converted into annual output se-
ries. The model is run separately for each of the gov-
ernorates considered as lumped isolated watersheds.
The general assessment is then carried out on the 11
governorates: individual annual series are integrated
on space for the whole districts to build a yearly
model over the entire study area. The simulated se-
ries are finally compared to observed hydrometric
data.
The historical runoffdata series are produced and
edited by DGRE (Tunisian Ministry of Agriculture)
in the Hydrological Yearbooks, which have been
published regularly since the early 1980s. [DGRE,
2020]. The Yearbooks, which cover the three ma-
jor hydrological basins making up the study area
(Far North-Ichkeul, Medjerda, Cap Bon-Miliane,
Figure 1), report on the measurements taken at the
main hydrometric stations with variable frequen-
cies but are most often formatted into daily values,
then aggregated into monthly and annual values.
These values do not consider the flow exchanges
between Algeria and Northern Tunisia, estimated
today at 100 Mm3/year on average in favor of Tunisia
[BPEH, 2019], nor the weak diffuse coastal runoff
not monitored with hydrometric stations. In the re-
gion’s water budget, these two quantities work in
opposite directions. Compared to the mean basin
runoff(2.3 km3/y), we consider as a first hypothesis
that these two quantities (# 4%) are confined to error
margins.
4.2. The hydrological model calibration process
The first step of model calibration consists to define
a calibration period for which we search the optimal
value of the following two parameters: Rmax the max-
imum soil reserve capacity, and Imax the maximum
infiltration capacity. Recall that for each governorate
these parameters represent spatially averaged char-
acteristics of the whole district and are unable to be
directly compared with field measurements. The sec-
ond step is validation; it consists to define a valida-
tion period for which we run the model with the al-
ready calibrated parameters Rmax and Imax. The cal-
ibration period extends over 17 hydrological years,
from 01/09/1985 to 31/08/2002; the validation ex-
tends over the 17 following years from 01/09/2002 to
31/08/2019 (Figure 8).
The calibration process aims to reproduce as well
as possible the observed annual runoffvolumes
[DGRE, 2020]. As for the second major term of the hy-
drological resources, annual groundwater recharge,
there are no direct observations or precise measure-
ments. We must therefore refer to the most recent
published field estimates [Besbes et al., 2019b].
Calibrating the model parameters:
(i) The maximum soil reserve capacity Rmax:
the first calibration tests showed that the
monthly time step operation is incompat-
ible with the potential evapotranspiration
displayed in the study region, yet one of
the wettest in Tunisia. Any significant Soil
Reserve Value will automatically feed into
the month’s AET, leaving virtually nothing
for runoffand groundwater recharge. It was
only by setting Rmax to zero that the model
was able to function normally, calculating
quite plausible water budgets. Under simi-
lar climatic conditions, this type of model
works very well daily, the monthly model is
supposed to integrate the daily balances of
the month. We succeeded in obtaining plau-
sible results directly at the monthly model by
introducing the artifice of Rmax =0.

10 Mustapha Besbes and Jamel Chahed
Figure 8. Modeled and observed yearly runofffor calibration and validation series (km3/y).
(ii) The maximum infiltration capacity Imax: the
dominant value is 10 mm/month, except in
the governorates of Beja and Jendouba where
Imax =8 mm.
We assess the model calibration quality, as well as
validation, by estimating: (i) the average relative bias
between computed and observed runoffof the series,
(ii) the correlation coefficient between modeled and
observed annual runoffs; (iii) Furthermore and more
specifically, our model estimates also the groundwa-
ter recharge: we then calculate the average deviation
between computed and field estimated recharge. Fi-
nally, we draw a scatter plot of modeled vs. observed
yearly runoff, respectively for calibration and valida-
tion series.
The hydrologic model calibration is based on
a double aggregation: (a) a spatial aggregation of
the outputs obtained for the eleven governorates of
the study region, (b) a temporal aggregation of the
monthly outputs (runoff, groundwater recharge)
to build yearly series that can be compared to
observed or estimated data. The results are as
follows:
(i) The average relative bias between computed
and observed annual runofffor calibration
and validation yearly series establishes at re-
spectively 22% and −5%.
(ii) The correlation coefficients between mod-
eled and observed annual runoffs are respec-
tively 0.86 and 0.74.
(iii) The average computed annual recharge for
the whole series 1985–2019 is 0.68 km3/y. The
average field estimated annual recharge is
0.67 km3/y [Besbes et al., 2019b]. The aver-
age deviation between computed and field-
estimated recharge is 0.01 km3/y that is a rel-
ative mean deviation of 1.5%.
4.3. Selection of CMIP6 GCMs compatible with
historical reference series
We have seen, Figures 5 and 6, that considering the
historical reference period 1995–2014, most models
exhibit a negative bias vis-à-vis historical tempera-
ture and precipitation series. We must therefore pro-
ceed to the choice of the models which are the most
suitable to correctly reproduce climatic and hydro-
logic series for the period 1995–2014, considered as
the historical reference period and for which there
are simulations by all of the CMIP Models.
The selection process of the most relevant GCMs
for predictive simulations follows two steps:
(1) The First level of selection criteria con-
cerns the predictability of the historical

Mustapha Besbes and Jamel Chahed 11
Table 2. Statistical parameters of the precipitation series generated by active GCMs on 1995–2014,
compared to the observed reference DGRE series
Mean mm/y Standard
deviation mm/y
Mean bias/DGRE
mm/y
Mean relative
bias
DGRE observed series 554.1 115.4
access-cm2 334.9 69.7 −219.2 −40%
access-esm1-5 251.7 58.7 −302.4 −55%
bcc-csm2-mr 259.9 65.9 −294.2 −53%
canesm5 206.4 44.2 −347.6 −63%
cmcc-esm2 277.1 57.9 −277.0 −50%
cnrm-cm6-1 652.2 141.6 98.1 18%
cnrm-esm2-1 630.0 121.5 75.9 14%
ec-earth3 277.7 62.7 −276.4 −50%
ec-earth3-veg 256.0 74.4 −298.1 −54%
fgoals-g3 504.6 62.6 −49.5 −9%
gfdl-esm4 364.9 72.8 −189.2 −34%
inm-cm4-8 269.3 62.9 −284.7 −51%
inm-cm5-0 339.8 66.6 −214.3 −39%
ipsl-cm6a-lr 421.1 111.6 −133.0 −24%
kace-1-0-g 324.8 71.5 −229.3 −41%
kiost-esm 318.4 69.2 −235.7 −43%
miroc6 451.0 87.1 −103.1 −19%
miroc-es2l 297.8 50.2 −256.3 −46%
mpi-esm1-2-hr 272.7 66.1 −281.4 −51%
mpi-esm1-2-lr 154.3 43.0 −399.8 −72%
mri-esm2-0 359.3 69.8 −194.8 −35%
nesm3 170.5 72.1 −383.6 −69%
noresm2-lm 182.8 34.5 −371.3 −67%
noresm2-mm 253.5 50.9 −300.6 −54%
taiesm1 252.1 62.7 −302.0 −55%
Ensemble-P50 280.4 14.3 −273.7 −49%
temperature and precipitation series, which
is evaluated by the degree of conformity of
the chronological graphs and by the quality
of the statistical parameters of the series.
This selection was limited to precipitation
(resp. Figures 9a and 9b) with criteria based
on the statistical series parameters. Indeed
and concerning the temperature, vis-à-vis
observed values, the 30 models present se-
ries with almost identical biases, and relative
biases of 5% with a standard deviation of
0.06%, which means that there is no signifi-
cant difference between the series resulting
from the models, which would have allowed
a selection. As for precipitation (Figure 9b
and Table 2), few series come close to the ref-
erence one (DGRE). According to criteria that
minimize differences with DGRE, it thus ap-
pears that five GCMs give precipitation close
to the observations, in particular with very

12 Mustapha Besbes and Jamel Chahed
Figure 9. Historical reference and predicted series (1995 to 2014) for annual temperature (a) and precip-
itation (b).
Table 3. Statistical parameters of runoffand groundwater recharge series generated by active GCMs on
1995–2014, compared to observed reference DGRE series
Data DGRE cnrm-cm6-1 cnrm-esm2-1 fgoals-g3 miroc6 ipsl-cm6a-lr
Runoff
Mean km3/y 2.87 3.93 3.64 1.99 1.49 1.30
Std deviation km3/y 1.59 2.55 2.17 1.23 1.45 1.49
Mean Bias vis-à-vis DGRE 1.06 0.76 −0.88 −1.38 −1.58
Mean Relative Bias 37% 27% −31% −48% −55%
Recharge
Mean km3/y 0.67 0.93 0.91 0.65 0.55 0.46
Std deviation km3/y 0.00 0.32 0.35 0.21 0.28 0.26
Mean Bias vis-à-vis DGRE 0.26 0.24 −0.02 −0.12 −0.21
Mean Relative Bias 39% 36% −3% −18% −31%
low average relative biases (Table 2); these
are the models cnrm-cm6-1, cnrm-esm2-1,
fgoals-g3, ipsl-cm6a-lr, miroc6.
(2) The second level of selection criteria con-
cerns the outputs of the hydrological model
and their ability to reproduce the reference
historical surface runoffand groundwater
recharge. This reproducibility is assessed by
the conformity of computed chronological
series toward the statistical parameters of
the reference series. Regarding groundwater
recharge, the average field estimated value
is 0.67 km3/y [Besbes et al., 2019b]. As for
runoff, each of the CMIP6 GCMs generates,
over the reference period 1995–2014, rain-
fall series used as input to the hydrolog-
ical model to produce the corresponding
runoffseries, which is compared to the series
observed by DGRE [2020]. The hydrological
results (Table 3) confirm in the first analysis
those obtained on precipitation (Table 2),
namely the five models: cnrm-cm6-1, cnrm-
esm2-1, fgoals-g3, ipsl-cm6a-lr, miroc6,
which give the best results.
But more precisely, and more particularly with
regard to runofffor which there are series of reliable
observations, the statistical parameters of Table 3
clearly indicate that the outputs of three models
among the five selected still remain far from the ref-
erence data, either too optimistic or too pessimistic,
and would therefore not guarantee sufficient relia-
bility on the predictions. These are the models: cnrm
cm6-1, ipsl-cm6a-lr, and miroc6. On the other hand,
the cnrm-esm2-1 and fgoals-g3 models give results
very close to the historical reference data, particu-

Mustapha Besbes and Jamel Chahed 13
larly for the runoffthat they globally frame, with a
respective bias of +0.76 and −0.88 km3/y. We do not
find this symmetry at the groundwater recharge level,
where fgoals-g3 reconstitutes the expected result on
its own. However, given that we do not have direct
measurements of the groundwater recharge but only
an estimate from expertise, we will give greater credit
to the models that best predict runoff.
5. Predicting blue and green water according
to the scenario ssp2-45
5.1. Construction of a global climate model spe-
cific to Northern Tunisia
Among the CMIP6 models, the cnrm-esm2-1 and the
fgoals-g3 are therefore those that best restore the se-
ries of historical runoffover Northern Tunisia; but
this restitution is imperfect: the cnrm slightly over-
estimates the sought solution and the fgoals slightly
underestimates it. We will consider that a suitable
model would be a composition of the two mod-
els over the computational grid, a composite model,
with a weighting αsuch that the composite precipita-
tion Prc is written: Prc =α×cnrm+(1−α)×fgoals. The
same weighting is applied to temperature. To find the
weighting that minimizes the difference with obser-
vations, we vary αbetween 0 and 1 on the rainfall se-
ries of the two models (cnrm and fgoals) for the refer-
ence historical period 1995–2014. We obtain the re-
sults presented in Table 4, where we see that there
is no solution that simultaneously minimizes both
the deviations in runoffand in recharge, but we can
consider that the best compromise would be reached
with α=0.5. In what follows, we will call this model
the “cnrm-fgoal” model.
5.2. Projection of blue water resources with the
cnrm-fgoal model
The hydrological behavior of Northern Tunisia is
simulated for the next successive situations: (i) the
known period 1995–2014 considered as the histori-
cal reference period with observed precipitation and
runoffseries, and estimated groundwater recharge,
(ii) the forecasting projection perspective covering
the period 2015–2100. The three Global Climate
Models: cnrm-esm2-1, fgoals-g3 and their compos-
ite “cnrm-fgoal”, are used to first simulate the hydro-
logic history of the basin for the reference period,
and then to predict the hydrological impacts related
to the ssp2-45 scenario. Figures 10 and 11 report the
obtained results. The explanations and comments on
these figures will be included in the summary Table 5.
However, concerning both runoffand groundwater
recharge, we can already observe that, after a partic-
ularly wet period that extends until 2030, the region
will experience a series of dry cycles that will last un-
til the end of the 21st century.
5.3. Projection of green water resources
5.3.1. The green water model: formulation and vali-
dation
The average Actual Evapotranspiration (AET) car-
ried out by the hydrological model described above
represents the ultimate potential of Green Water
(GW) resources that can be extracted by plants from
the soil. Only part of this potential is productive
and can be converted into food production. Indeed:
(i) Only part of arable land available for rainfed crops
is cultivated, (ii) Evapotranspiration does not oc-
cur throughout the year, especially for annual crops
such as cereals whose active period is limited in
time, (iii) Cultivated area is not always entirely har-
vested due to possible crop failure during low rain-
fall episodes. The Green Water Model (GWM) is for-
mulated on the basis of hydrological model outputs,
which provide AET. To determine the productive part
of GW resource, we define the rate of conversion β
of the annual AET, which represents the maximum
green water potential of the whole region under study
(GWP), into annual GW actually used by the crops, so
that we have GW =β.GWP, where GWP =A.AET, A is
the yearly cultivated area. GW here is assumed to be
the green water useful volume, which is proportional
to agricultural production.
As for the hydrological model, the GWM is for-
mulated for each governorate and adjusted for ce-
reals cultivated area (8640 km2on average) based
on sub-national data from the Ministry of Agricul-
ture [MARH, 1998–2012] and national data from FAO-
STAT [FAO, 2017a,b]. These data sources provide re-
gional statistics on agricultural production from 1985
to 2020. The model outputs by governorate are inte-
grated over the entire study region and adjusted on
observations.
Figure 12 shows the relationship between the an-
nual production of cereals (CPr) in Million Tons

14 Mustapha Besbes and Jamel Chahed
Figure 10. ssp2-45 Runoffsimulation with cnrm-esm2-1, fgoals-g3, and the selected cnrm-fgoal model
for historical observation period and projections.
Figure 11. ssp2-45 Groundwater recharge simulation with cnrm-esm2-1, fgoals-g3, and the selected
cnrm-fgoal model for historical observation period and projections.
Table 4. The weighting effect on runoffand groundwater recharge results, from 100% cnrm-esm2-1
(α=1) to 100% fgoals-3g (α=0)
Alpha 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
Bias on runoffkm3/y 0.76 0.55 0.34 0.14 −0.04 −0.22 −0.37 −0.52 −0.65 −0.77 −0.88
Bias on recharge km3/y 0.24 0.23 0.20 0.18 0.16 0.13 0.10 0.10 0.04 0.01 −0.02
according to cereals’ Green Water Potential (GWP)
assimilated to the total volume of AET. The slope
of the least squares regression line provides the
specific increase in production per unit of GWP,
which amounts to 0.58 kg/m3i.e. a specific volume
of GW of 1.71 m3/kg. This value compares quite well
with the water productivity of cereals in arid coun-
tries [Zimmer, 2013].
The specific volume of GW is then used to cal-
culate the equivalent GW of annual cereal produc-
tion and deduce the corresponding beta conversion
rates. These are fitted using linear regression over the
period 1985–2019 and presented in Figure 13, where
the regression line shows a significant increase in the
conversion rates of GWP into useful GW, from 0.5 to
0.7, reflecting crop yield improvement. The conver-

Mustapha Besbes and Jamel Chahed 15
Figure 12. Cereal production in Northern Tunisia related to Green Water Potential AET (period 1985–
2019).
sion function beta is used to generate retrospective
GW amounts (Figure 13) compared to observed ce-
reals production. Figure 13 shows that the GW vol-
umes produced by the model manage to capture the
relationship between climatic conditions and cereal
production, mainly grown in rain fed conditions.
During this historical reference period, GW follows
the evolution of the beta factor, indicating signifi-
cant yield evolution despite the stability of cereal sur-
faces, [Attiaoui and Boufateh, 2019]. Cereals Yields
improvement is not specific to the study area; it is
found throughout Tunisia as well as in other Maghreb
countries (Algeria and Morocco). Voluntary agricul-
tural policies, as well as a number of incentives and
favorable farming practices explain that in the case
of wheat, for example, yields have doubled in twenty
years [Besbes et al., 2019a].
5.3.2. Simulation of the green water future with the
selected models:
To make the predictive results comparable with
the current situation, prospective simulations cover-
ing the period 2020–2100 were carried out with initial
conditions, those prevailing in the 2010s, including
crop yield and cultivated areas. For instance, the av-
erage maximum value of beta over 2011–2020 (0.68)
will be kept constant throughout the prospective
study period. Thus, this exercise makes it possible
to examine the impacts of regional parameters pre-
dicted by Global Climatic Models on green water re-
sources. For both blue water and green water, all cal-
culations are carried out at the level of each of the
eleven governorates, and then aggregated at the level
of the entire region (Figure 14 and Table 5). Calcu-
lus uses outputs of the two GCMs cnrm-esm21 and
fgoals-g3, as well as their composite cnrm-fgoal pre-
viously adjusted as part of the hydrological model.
5.4. Results of water resources prediction and
summary table
Table 5 is a synthetic composition intended to guide
the reader. It summarizes precipitation, runoff,
recharge, and green water calculated from the com-
posite model “cnrm-fgoal”, coupled to the hydrolog-
ical and the green water models, implemented over
the observation and the projections periods. Here
are the main points and findings:
(i) The decade 2011–2020 represents the his-
torical reference state of the hydro-climatic
system; it is the initial state of the projections,
calculated with the model by reference to the
observed state and to which any variation is
related.

16 Mustapha Besbes and Jamel Chahed
Figure 13. Checking the Green Water Model over the period 1985–2019, by comparison with cereals’
water footprint (km3/y).
Figure 14. ssp2-45 useful Green Water simulation with cnrm-esm2-1, fgoals-g3, and cnrm-fgoal for
historical observation period (1995–2014) and projections (2015–2100).
(ii) We have chosen to present the general
results of the study in successive se-
quences of 20 years, i.e. four sequences
to represent the period simulated in the
projections.
(iii) A first reading of the table highlights the great
fragility of blue water in relation to changes
in precipitation and temperature, which is
not the case for less sensitive green water. In a
first analysis, this difference is probably due
to the priority given to AET when modeling
the water budget.
This being so, we can note the following main
results expressed in Table 5:
(iv) Precipitation: starting from 16.33 km3/y in
the 2010s, precipitation decreases to a min-
imum of around 14 km3/y in the 2050s and
then slightly rises to 14.7 km3/year at the end
of the century; this represents an average de-
cline of 9% over the 21st Century.
(v) Temperatures: during the century, the aver-
age annual temperature increased by nearly
2 °C, ranging from 17.56 °C to 19.44 °C.

Mustapha Besbes and Jamel Chahed 17
Table 5. Summary table in Northern Tunisia for overall modeling with the GCM “cnrm-fgoal”, the
hydrological model and the green water model
Item/period 2011–2020 2021–2040 2041–2060 2061–2080 2080–2100 Summary
2021–2100
Average observed precipitation (Pr) km3/y 15.03
Average precipitation cnrm-fgoal km3/y 16.23 15.56 14.53 14.07 14.69 14.71
Difference with initial Pr % 0% −4% −11% −13% −10% −9%
Average reference temperature ERA5 °C/y 18.30
Average temperature by cnrm-fgoal °C/y 17.56 17.95 18.57 19.00 19.44 18.74
Difference with initial temperature % 0% 2% 6% 8% 11% 7%
Average observed runoffkm3/y 3.08
Average runoffhydrologic model km3/y 2.15 1.99 1.88 1.35 1.60 1.70
Difference with initial runoff% 0% −7% −13% −37% −26% −21%
Average estimated recharge km3/y 0.67
Average recharge hydrologic model km3/y 0.72 0.72 0.59 0.57 0.57 0.61
Difference with initial recharge % 0% 0% −18% −21% −21% −15%
Average green water harvested km3/y 2.58
Average green water model km3/y 2.67 2.60 2.44 2.45 2.53 2.50
Difference with initial GW % 0% −3% −9% −8% −5% −6%
Total actual evapotranspiration AET km3/y 13.37 12.85 12.06 12.15 12.52 12.39
Difference with initial AET % 0% −4% −10% −9% −6% −7%
(vi) With regard to runoff, it decreases on
the composite model cnrm-fgoal, from
2.15 km3/y on average in the 2010s to 1.35 in
the 2070s, i.e. a drop of 37%, then increases to
form a century average value of 1.70 km3/y.
i.e. an average drop of 21%.
(vii) During the century, aquifers recharge went
from 0.72 to 0.57 km3/y, i.e. a drop of 21%.
(viii) With regard to Green Water, the predictive
simulations for 2100 show evolutions slightly
below the initial level, with a decrease of up
to −6% on average.
6. Discussion
Finally, the following points should be pointed out in
the form of a discussion:
(1) On the green water model: over the projected
period 2021–2100, the temperature increases
steadily. This warming results in higher
water requirements for plants: (i) higher
evapotranspiration demand, (ii) longer and
warmer growth periods. The first factor is
already effective in terms of the results ob-
tained; indeed, the decrease in GW is small
compared to that of precipitation. The sec-
ond factor is not taken into account by our
green water model; one can imagine that
such an additional withdrawal of green wa-
ter would lead to even greater decreases in
runoffand recharge [Mankin et al., 2019].
(2) On the development of scenario ssp2.45: at
the end of the 2060s (Table 5), we observe
an important reduction in runofffollowed
by a rise. Similar results are observed in the
simulations obtained from the outputs of
other GCMs we analyzed, i.e. cnrm cm6-1
and miroc6. We previously attributed this
phenomenon to the priority given to evap-
otranspiration by the hydrological model.
This is partly true, but there is another rea-
son which produces a perhaps even greater
effect. Indeed: the SSP2 narrative assumes
moderate global population growth with a
demographic transition in the second half
of the century. Population stabilization com-
bined with sustained income growth would
be accompanied by an improvement in agri-
cultural productivity as well as in carbon
and energy intensity. On the other hand, the

18 Mustapha Besbes and Jamel Chahed
increasing rigor of climate change mitiga-
tion is expected to decrease dependence on
fossil fuels, in particular by optimizing the
energy mix [Fricko et al., 2017]. All these
factors are likely to reduce the effects of
human activities that climate models seem
to capture.
(3) On other studies dealing with long-term wa-
ter resources prediction in Northern Tunisia:
(i) Precipitation forecasts with the “cnrm-
fgoal” model indicate a slight change
over the first half of the century, then
a huge decrease with a maximum drop
around the sixties (−13%). Then follows
a slow recovery, bringing the average
variation 2021–2100 to (−9%). These
values are within orders of magnitude
of previous predictions with RCM for
Northern Tunisia. Under scenario RCP
4.5: Deidda et al. [2013] and Bird et al.
[2016] predict a reduction of around
(−17%) for 2040–2070. For the same
period, Dakhlaoui et al. [2022] find an
average value of (−8%) in the region,
increasing to (−18%) for 2070–2100.
Terink et al. [2013] announce a (−5% to
−10%) for 2040–2050 and the AFD-MA
study [AFD-MA, 2021] predicts, for all
of Tunisia, a precipitation drop of (−6%
to −9%) by 2050 and (−9% to −18%) by
2100.
(ii) Concerning runoffand under scenario
RCP 4.5, Dakhlaoui et al. [2022] predict
runoffreductions of (−10% to −30%)
for 2040–2070 and −20% to −38% for
the period 2070–2100, figures which are
close to those of our present simula-
tions, which expect reductions around
−37% by 2070 and −26% by 2090.
(4) Consequences of green water predictions for
cereal production: In line with the results of
most studies in the Maghreb countries and
the Mediterranean basin, rainfall is consid-
ered the crucial factor in cereals production
[Zimmer, 2013]. This suggests that a precipi-
tation decline would have harmful repercus-
sions on cereals production. According to the
policy-free scenario (RCP8.5), the AFD-MA
[2021] study predicts a decline of more than
a third of National agricultural production
by 2100. Based on statistical approaches, At-
tiaoui and Boufateh [2019] estimate that na-
tionally, a 1% decrease in rainfall would re-
sult in (0.92%) decrease in cereals produc-
tion in the short run and (1.295%) in the long
run. These would be less marked in the more
watered cereal-growing regions of the north.
The authors further suggest that the increase
in temperature would even have a positive
effect on cereals production in these colder
regions of Tunisia. Our green water model
provides comparable orders of magnitude of
the impact of climate change on cereals pro-
duction. The predictive simulations for 2100
show an average decrease of around 6% in
green water (e.g., cereals production) for an
increase in temperature of 7% and a decrease
in precipitation of 9%. It is as if the effects
of the increase in temperature and the de-
crease in precipitation compensate for each
other to produce smaller changes in the pre-
dicted AETs and thus in crops productions,
albeit the conversion rates and cultivated ar-
eas, which are considered constant.
In sum, the results of the simulations pre-
dict an increase in temperature accompa-
nied by a long-term decrease in precipita-
tion. The implications of these predictions
on Green Water are less significant than on
Blue Water. This is an essential finding of this
research that reinforces the previous ana-
lyzes showing the weight and the paramount
role of green water in the national water
budget [Besbes et al., 2019b]. These find-
ings, their deepening, monitoring and pos-
sibly updating constitute essential informa-
tion to guide the development of adaptation
strategies compatible with the future poten-
tial of water resources in all their forms and
for each region.
7. Conclusion
The simulations carried out by the GCMs that partici-
pated in the CMIP6 exercise provide temperature and
precipitation series for the study area. We considered
in this study the IPCC medium scenario SSP2-4-5,
designed to prolong current trends. Analysis of the

Mustapha Besbes and Jamel Chahed 19
thirty GCMs used in the research led to the selection
of only two GCMs (cnrm-esm2.1, fgoals-g3) whose
outputs: (i) are the closest to the observed historical
temperature and precipitation series but imperfectly,
(ii) applied as inputs to the hydrological model, gen-
erate outputs that frame the observed runoffseries.
The modeling systems’ runoffoutputs are weighted
to fit observations and build a composite best-fit
model: the “cnrm-fgoal” weighted at 0.5–0.5 for the
prospective hydrological simulations until 2100.
Inputs and outputs of the cnrm-fgoal composite
model indicate slight alterations of water resources
until the forties, then a huge alteration from the
middle of the 21st century and a maximum drop
in the sixties, with a significant drop in precipita-
tion (−13%), runoff(−37%), groundwater recharge
(−21%) and Green Water (−8%). Subsequently, we
note a recovery in all indicators, so that the average
variation over most of the 21st century, from 2021
to 2100, will be a decrease of −9% for precipitation,
−21% for runoff,−15% for recharge, and −6% for
Green Water.
To our initial question, stated as follows: can
we predict Water Resources only with GCMs, with-
out downscaling, the results obtained in the present
study show that the use of the raw predictions of orig-
inal climate models for hydrological modeling pur-
poses on large basins is possible. Simulations at these
scales appear to provide relative precisions compara-
ble to what has been produced in Northen Tunisia by
Regional Models applied to small/moderate basins
after inputs bias corrections.
The direct implementation of climate predictions
for hydrological studies is fundamentally based on a
rigorous selection of suitable climate models for the
region, carried out based on an exhaustive analysis of
the historical simulations of the climatic models and
their comparison with observed data on the same pe-
riods. This requires that reliable and complete hydro-
logical data are available on spatiotemporal scales
compatible with those of climate simulations.
Conflicts of interest
Authors have no conflict of interest to declare.
References
AFD-MA (2021). French Development Agency-
Ministry of Agriculture. Tunisia. Contribution to
the elements of the preparatory phase of the Na-
tional Adaptation Plan for food security. Summary.
Dec. 2021.
Alexandris, S., Stricevic, R., and Petkovic, S. (2008).
Comparative analysis of reference evapotranspira-
tion from the surface of rainfed grass in central Ser-
bia, calculated by six empirical methods against
the Penman–Monteith formula. Eur. Water, 21/22,
17–28.
Allan, J. A. (1998). Moving water to satisfy uneven
global needs: trading water as an alternative to en-
gineering it. ICID J., 47(2), 1–8.
ANME (National Agency for Energy Management)
(2023). Solar Photovoltaic. Solaire Photovoltaique
|ANME. accessed on 21/03/2023.
Attiaoui, I. and Boufateh, T. (2019). Impacts of cli-
mate change on cereal farming in Tunisia: a panel
ARDL–PMG approach. Environ. Sci. Pollut. Res., 26,
13334–13345.
Bargaoui, Z., Tramblay, Y., Lawin, E. A., and Servat,
E. (2014). Seasonal precipitation variability in re-
gional climate simulations over northern basins of
Tunisia. Int. J. Climatol., 34(1), 235–248.
Besbes, M. (1978). Estimation of Groundwater
Recharge. A Regional Effective Infiltration Model.
Doc.th. Pierre & Marie Curie University, Paris.
Besbes, M., Chahed, J., and Hamdane, A. (2014).
Sécurité Hydrique de la Tunisie, Gérer l’eau en Con-
ditions de Pénurie. Ed. L’Harmattan, Paris.
Besbes, M., Chahed, J., and Hamdane, A. (2019a).
Food and water management in Northwest Africa.
In The Oxford Handbook of Food, Water and Soci-
ety, page 426. Oxford University Press, New York.
Besbes, M., Chahed, J., and Hamdane, A. (2019b). Na-
tional Water Security: Case Study of an Arid Coun-
try: Tunisia. Springer International Publishing,
Cham, Switzerland.
Besbes, M., Chahed, J., Hamdane, A., and De Marsily,
G. (2010). Changing water resources and food sup-
ply in arid zones: Tunisia. In Schneider-Madanes,
G. and Courel, M. F., editors, Water and Sustain-
ability in Arid Regions. Springer, Berlin.
Bird, D. N., Benabdallah, S., Gouda, N., Hummel, F.,
Koeberl, J., La Jeunesse, I., Meyer, S., Prettenthaler,
F., Soddu, A., and Woess-Gallasch, S. (2016). Mod-
elling climate change impacts on and adaptation
strategies for agriculture in Sardinia and Tunisia
using AquaCrop and value-at-risk. Sci. Total Env-
iron., 543, 1019–1027.

20 Mustapha Besbes and Jamel Chahed
Bonnet, M., Delarozière, O., Jusserand, C., and Roux,
P. (1970). In Calcul des bilans d’eau mensuels et
annuels par les méthodes de Tornthwaite et de Turc.
BRGM, France. 70SGN107HYD.
BPEH (2019). National Water Sector Report. Office
of Planning and Hydraulic Balances. BPEH. The
Ministry of Agriculture, Tunisia.
Bruyère, C. L., Done, J. M., Holland, G. J., and
Fredrick, S. (2014). Bias corrections of global mod-
els for regional climate simulations of high-impact
weather. Clim. Dyn., 43, 1847–1856.
Chahed, J., Hamdane, A., and Besbes, M. (2008). A
comprehensive water balance of Tunisia: blue wa-
ter, green water and virtual water. Water Int., 33(4),
415–424.
Collins, W. J., Bellouin, N., Doutriaux-Boucher, M.,
Gedney, N., Halloran, P., Hinton, T., et al. (2011).
Development and evaluation of an Earth-System
model–HadGEM2. Geosci. Model. Dev., 4(4), 1051–
1075.
Cos, J., Doblas-Reyes, F., Jury, M., Marcos, R., Breton-
nière, P. A., and Samsó, M. (2022). The Mediter-
ranean climate change hotspot in the CMIP5 and
CMIP6 projections. Earth Syst. Dyn., 13(1), 321–
340.
Dakhlaoui, H., Hakala, K., and Seibert, J. (2022). Hy-
drological impacts of projected climate change on
Northern Tunisian headwater catchments—an en-
semble approach addressing uncertainties. In Cli-
mate Change in the Mediterranean and Middle
Eastern Region, pages 499–519. Springer, Cham.
de Marsily, G. (2008). Eau, changements climatiques,
alimentation et évolution démographique. Rev. Sci.
Eau./J. Water Sci., 21(2), 111–128.
de Marsily, G. (2020). Will we soon run out of water?
Ann. Nutr. Metab., 76(1), 10–16.
de Marsily, G. and Abarca-del Rio, R. (2016). Wa-
ter and food in the twenty-first century. In Re-
mote Sensing and Water Resources, pages 313–337.
Springer, Cham.
Deidda, R., Marrocu, M., Caroletti, G., Pusceddu, G.,
Langousis, A., Lucarini, V., Puliga, M., and Sper-
anza, A. (2013). Regional climate models’ perfor-
mance in representing precipitation and temper-
ature over selected Mediterranean areas. Hydrol.
Earth Syst. Sci., 17(12), 5041–5059.
DGRE (1980–2020). Hydrological Yearbooks of
Tunisia. General Directorate of Water Resources
(DGRE), The Ministry of Agriculture, Tunisia.
DGRE, Prointec-Comete (2019). Carte des Ressources
en eau de la Tunisie. CRET. Phase III: Elaboration
de la carte des écoulements Superficiels. Direction
Générale des Ressources en eau (DGRE), Ministère
de l’agriculture, Tunis.
Döll, P. (2009). Vulnerability to the impact of cli-
mate change on renewable groundwater resources:
a global-scale assessment. Environ. Res. Lett., 4(3),
article no. 035006.
FAO (2017a). AQUASTAT: FAO’s global water informa-
tion system, the Land and Water Division. http://
www.fao.org/nr/water/aquastat/main/index.stm,
consulted on 11 Oct 2022.
FAO (2017b). FAOSTAT: food and agriculture data.
Statistics Division. http://www.fao.org/faostat/en/
#data,k consulted on 11 Oct 2022.
Farsani, I., Farzaneh, M. R., Besalatpour, A. A., Salehi,
M. H., and Faramarzi, M. (2019). Assessment of
the impact of climate change on spatiotemporal
variability of blue and green water resources under
CMIP3 and CMIP5 models in a highly mountain-
ous watershed. Theor. Appl. Climatol., 136, 169–
184.
Fathalli, B., Pohl, B., Castel, T., and Safi, M. J. (2019).
Errors and uncertainties in regional climate simu-
lations of rainfall variability over Tunisia: a multi-
model and multi-member approach. Clim. Dyn.,
52(1), 335–361.
Foughali, A., Tramblay, Y., Bargaoui, Z., Carreau, J.,
and Ruelland, D. (2015). Hydrological modeling in
Northern Tunisia with regional climate model out-
puts: performance evaluation and bias-correction
in present climate conditions. Climate, 3(3), 459–
473.
Fricko, O., Havlik, P., Rogelj, J., Klimont, Z., Gusti, M.,
Johnson, N., et al. (2017). The marker quantifi-
cation of the Shared Socioeconomic Pathway 2: a
middle-of-the-road scenario for the 21st century.
Glob. Environ. Change, 42, 251–267.
Hamed, M. M., Nashwan, M. S., and Shahid, S. (2022).
A novel selection method of CMIP6 GCMs for ro-
bust climate projection. Int. J. Climatol., 42(8),
4258–4272.
Harris, I., Osborn, T. J., Jones, P., et al. (2020). Version
4 of the CRU TS monthly high-resolution gridded;
multivariate climate dataset. Sci. Data, 7, article
no. 109.
Herger, N., Abramowitz, G., Knutti, R., Angélil, O.,
Lehmann, K., and Sanderson, B. M. (2018). Select-

Mustapha Besbes and Jamel Chahed 21
ing a climate model subset to optimise key ensem-
ble properties. Earth Syst. Dyn., 9, 135–151.
Hersbach, H., Bell, B., Berrisford, P., Hirahara, S.,
Horányi, A., and Muñoz Sabater, J. (2020). The
ERA5 global reanalysis. Q. J. R. Meteorol. Soc.,
146(730), 1999–2049.
Hoekstra, A. Y. (2003). Virtual Water Trade: Proceed-
ings of the International Expert Meeting on Virtual
Water Trade, Delft, The Netherlands, 12–13 Decem-
ber 2002. Value of Water Research Report, Series
No. 12. IHE, Delft, The Netherlands.
Hughes, J. P. and Guttorp, P. (1994). A class of stochas-
tic models for relating synoptic atmospheric pat-
terns to regional hydrologic phenomena. Water Re-
sour. Res., 30(5), 1535–1546.
INM (2022). Banque de Données Climatologiques.
INM | Institut National de la Météorologie (me-
teo.tn). Institut National de la Météorologie, Tunis.
IPCC (2021). In Masson-Delmotte, V., Zhai, P., Pi-
rani, A., Connors, S. L., Péan, C., Berger, S., Caud,
N., Chen, Y., Goldfarb, L., Gomis, M. I., Huang, M.,
Leitzell, K., Lonnoy, E., Matthews, J. B. R., May-
cock, T. K., Waterfield, T., Yelekçi, O., Yu, R., and
Zhou, B., editors, Climate Change 2021: The Phys-
ical Science Basis. Contribution of Working Group
I to the Sixth Assessment Report of the Intergovern-
mental Panel on Climate Change. Cambridge Uni-
versity Press, Cambridge, UK and New York, NY,
USA.
Jiang, Q., Li, W., Fan, Z., He, X., Sun, W., Chen, S., Wen,
J., Gao, J., and Wang, J. (2021). Evaluation of the
ERA5 reanalysis precipitation dataset over Chinese
Mainland. J. Hydrol., 595, article no. 125660.
Jiao, D., Xu, N., Yang, F., et al. (2021). Evaluation
of spatial-temporal variation performance of ERA5
precipitation data in China. Sci. Rep., 11, article
no. 17956.
Kim, S., Eghdamirad, S., Sharma, A., and Kim, J. H.
(2020). Quantification of uncertainty in projections
of extreme daily precipitation. Earth Space Sci.,
7(8), article no. e2019EA001052.
King, L., Nasr, Z., Almohamad, H., and Maag, C. C.
(2007). le Climat. In MARH, GTZ, Gopa, Exaconsult:
Stratégie nationale d’adaptation de l’agriculture
tunisienne et des écosystèmes aux changements cli-
matiques. Janvier, Tunis. Ch. 7.2.
Knutti, R., Sedláˇcek, J., Sanderson, B. M., Lorenz, R.,
Fischer, E. M., and Eyring, V. (2017). A climate
model projection weighting scheme accounting for
performance and interdependence. Geophys. Res.
Lett., 44, 1909–1918.
Laurent, A., Fennel, K., and Kuhn, A. (2021). An
observation-based evaluation and ranking of his-
torical Earth system model simulations in the
northwest North Atlantic Ocean. Biogeosciences,
18(5), 1803–1822.
Li, X., Tan, L., Li, Y., Qi, J., Feng, P., Li, B., Liu, D. L.,
Zhang, X., Marek, G. W., Zhang, Y., Liu, H., Srini-
vasan, R., and Chen, Y. (2022). Effects of global
climate change on the hydrological cycle and
crop growth under heavily irrigated management–
A comparison between CMIP5 and CMIP6. Com-
put. Electron. Agric., 202, article no. 107408.
Mandal, S., Breach, P. A., and Simonovic, S. P.
(2016). Uncertainty in precipitation projection un-
der changing climate conditions: a regional case
study. Am. J. Clim. Change, 5(1), 116–132.
Mankin, J. S., Seager, R., Smerdon, J. E., Cook, B. I.,
and Williams, A. P. (2019). Mid-latitude freshwa-
ter availability reduced by projected vegetation re-
sponses to climate change. Nat. Geosci., 12(12),
983–988.
Maraun, D. and Widmann, M. (2018). Cross-
validation of bias-corrected climate simulations is
misleading. Hydrol. Earth Syst. Sci., 22(9), 4867–
4873.
MARH (1998–2012). Annuaires des statistiques agri-
coles. Ministère de l’Agriculture et des Ressources
Hydrauliques, Tunisia.
McMahon, T. A., Peel, M. C., Lowe, L., Srikanthan, R.,
and McVicar, T. R. (2013). Estimating actual, po-
tential, reference crop and pan evaporation using
standard meteorological data: a pragmatic synthe-
sis. Hydrol. Earth Syst. Sci., 17, 1331–1363.
McSweeney, C. F., Jones, R. G., Lee, R. W., and Row-
ell, D. P. (2015). Selecting CMIP5 GCMs for down-
scaling over multiple regions. Clim. Dyn., 44, 3237–
3260.
Nasr, Z., Almohammed, H., Gafrej Lahache, R., Maag,
C., and King, L. (2008). Drought Modelling un-
der climate change in Tunisia during the 2020 and
2050 periods. Option Méditerr. Séries A, 80, 365–
369.
NOAA (2022). Global Historical Climatol-
ogy Network monthly (GHCNm). https:
//www.ncei.noaa.gov/data/ghcnm/v4beta/doc/
ghcn-m_v4_prcp_inventory.txt.
Oki, T., Sato, M., Kawamura, A., Miyake, M., Kanae, S.,

22 Mustapha Besbes and Jamel Chahed
and Musiake, K. (2003). Virtual water trade to Japan
and in the world. In Hoekstra, A. Y., editor, Value of
Water Research Report, Series No.12. IHE, Delft, The
Netherlands.
O’Neill, B. C., Tebaldi, C., Van Vuuren, D. P., Eyring,
V., Friedlingstein, P., Hurtt, G., et al. (2016). The
scenario model intercomparison project (Scenari-
oMIP) for CMIP6. Geosci. Model Dev., 9(9), 3461–
3482.
Oueslati, I., Lili-Chabaane, Z., Shabou, M., Zribi, M.,
Ben Issa, N., Chakroun, H., Galafassi, D., Rathwell,
K., Hoff, H., and Pizzigalli, C. (2012). Methodology
to Analyse the actual and the future effect of wa-
ter scarcity on the available water resources in Mer-
guellil watershed. Geophys. Res. Abstr., 14, article
no. EGU2012-9366-1.
Ramirez-Villegas, J., Challinor, A. J., Thornton, P. K.,
and Jarvis, A. (2013). Implications of regional im-
provement in global climate models for agricul-
tural impact research. Environ. Res. Lett., 8(2), arti-
cle no. 024018.
Renault, D. and Wallender, W. W. (2000). Nutritional
water productivity and diets: from crop per drop,
towards nutrition per drop. Agric. Water Manage.,
45, 275–296.
Riahi, K., Van Vuuren, D. P., Kriegler, E., Edmonds, J.,
O’neill, B. C., Fujimori, S., and Tavoni, M. (2017).
The shared socioeconomic pathways and their en-
ergy, land use, and greenhouse gas emissions im-
plications: an overview. Glob. Environ. Change, 42,
153–168.
Schewe, J., Heinke, J., Gerten, D., Haddeland, I., Ar-
nell, N. W., Clark, D. B., Dankers, R., Eisner, S.,
Fekete, B., Colón-González, F. J., Gosling, S. N.,
Kim, H., Liu, X., Masaki, Y., Portmann, F., Satoh, Y.,
Stacke, T., Tang, Q., Wada, Y., Wisser, D., Albrecht,
T., Frieler, K., Piontek, F., Warszawski, L., and Kabat,
P. (2014). Multi-model assessment of water scarcity
underclimate change. Proc. Natl. Acad. Sci. USA,
111(9), 3245–3250.
Shokouhifar, Y., Lotfirad, M., Esmaeili-Gisavandani,
H., and Adib, A. (2022). Evaluation of climate
change effects on flood frequency in arid and semi-
arid basins. Water Suppl., 22(8), 6740–6755.
Slama, F., Gargouri-Ellouze, E., and Bouhlila, R.
(2020). Impact of rainfall structure and climate
change on soil and groundwater salinization. Clim.
Change, 163(1), 395–413.
Somot, S., Ruti, P., Ahrens, B., Coppola, E., Jordà, G.,
Sannino, G., and Solmon, F. (2018). Editorial for the
Med-CORDEX special issue. Clim. Dyn., 51(3), 771–
777.
Switanek, M., Maraun, D., and Bevacqua, E. (2022).
Stochastic downscaling of gridded precipitation to
spatially coherent subgrid precipitation fields us-
ing a transformed Gaussian model. Int. J. Clima-
tol., 42(12), 6126–6147.
Tarek, M., Brissette, F. P., and Arsenault, R. (2020).
Evaluation of the ERA5 reanalysis as a potential
reference dataset for hydrological modelling over
North America. Hydrol. Earth Syst. Sci., 24, 2527–
2544.
Terink, W., Immerzeel, W. W., and Droogers, P. (2013).
Climate change projections of precipitation and
reference evapotranspiration for the Middle East
and Northern Africa until 2050. Int. J. Climatol.,
33(14), 3055–3072.
Wang, Z., Zhan, C., Ning, L., and Guo, H. (2021). Eval-
uation of global terrestrial evapotranspiration in
CMIP6 models. Theor. Appl. Climatol., 143(1), 521–
531.
Watanabe, S., Hajima, T., Sudo, K., Nagashima, T.,
Takemura, T., Okajima, H., Nozawa, T., Kawase, H.,
Abe, M., Yokohata, T., Ise, T., Sato, H., Kato, E.,
Takata, K., Emori, S., and Kawamiya, M. (2011).
MIROC-ESM 2010: model description and basic re-
sults of CMIP5-20c3m experiments. Geosci. Model
Dev., 4(4), 845–872.
WBG (2022). The World Bank Group, Cli-
mate Change Knowledge Portal. https:
//climateknowledgeportal.worldbank.org/
download-data.
WMO (2022). Climate Explorer: Starting point. KNMI
Climate Explorer https://climexp.knmi.nl/.
Zhang, M. Z., Xu, Z., Han, Y., and Guo, W. (2022). Eval-
uation of CMIP6 models toward dynamical down-
scaling over 14 CORDEX domains. Clim. Dyn.,
pages 1–15.
Zimmer, D. (2013). L’empreinte eau. Les faces cachées
d’une ressource vitale. Charles Léopold Meyer,
Paris.