The Casamance River, 320 km long, rises in eastern Casamance, around Fafacourou, 50 km northeast of Kolda, and crosses it from east to west, flowing into the Atlantic Ocean. It is a small coastal river whose watershed is almost entirely within the territory of Senegal [5] . The Casamance River watershed is located in the southern part of Senegal, between latitudes 12˚20' and 13˚21' North and longitudes 14˚17' and 16˚47' West (Figure 1). This area has a tropical Sudano-Guinean climate, hot and humid with an average annual temperature of 27˚C [9] . Only a small southern part of the basin extends as far as Guinea Bissau.

The region’s climate is Sudano-Guinean, with rainfall from June to October, peaking in August and September, and a dry season from November to May [10] . Average rainfall ranges from 700 to 1300 mm. The lowest average monthly temperatures are recorded between December and January, ranging from 25˚C to 30˚C, while the highest temperatures are recorded between March and September, with variations of 30˚C to 40˚C.

Rainfall and evapotranspiration data (determined using the formula proposed by OUDIN et al. [11] for rainfall-runoff models), which reflect the climatological phenomena of the model, and flow measurements, which reveal the hydrological functioning of the catchment, are used in this study.

- Rainfall data

Before studying a long series of rainfall data, it is necessary to check whether it corresponds to a homogeneous whole. For example, it is important to check whether there is any heterogeneity due to the use of faulty equipment over one or more periods of time: a rain gauge with a non-compliant ring surface and test tube, a rain gauge with buckets that tilt before or after the correct volume, etc. Last but not least, there is the possibility of human error: forgotten observations, operating errors, poor data entry, etc. [12] .

Rainfall data vary in quality and duration from station to station. This makes it necessary to select reference stations. The selection criteria are based on three fundamental factors: the size of the sample, their proximity to the study area (i.e. their geographical position) and the quality of the data (weakness of gaps in the different series actually observed) [13] .

For our study, we have rainfall data from the ANACIM (French National Civil Aviation and Meteorology Agency) stations in Ziguinchor, Sedhiou and Kolda. Of these three stations, the Kolda station was selected on the basis of its location in the upstream Casamance basin and its spatial and temporal representativeness.

The homogeneity of station data is checked using KhronoStat software. KhronoStat includes various homogeneity tests (homogeneity being understood here in the sense of the absence of breaks in the series): MANN WHITNEY test modified by AN PETTITT, TA BUISHAND U statistic, AFS LEE and SA HEGHINIAN procedure, P. BOIS control ellipse and HUBERT segmentation method. For all these tests, the null hypothesis H0 corresponds to the absence of a break at the 1% threshold. These tests are particularly sensitive to a change in mean, and if the null hypothesis of series homogeneity is rejected, they provide an estimate of the break date [14] . In this study, homogeneity is verified using the BUISHAND test, the PETTITT test, the Bayesian method of Lee and HEGHINIAN and the HUBERT segmentation method.

- Potential evapotranspiration

Potential evapotranspiration (PTE), an essential input to the model, expresses climatic demand and is used in the model’s production function. PTE values were estimated using the formula proposed by Oudin et al. based on temperature data from the Kolda station. These values were calculated on a daily time step and then scaled monthly.

- Flow rates

Flows are used to assess the quality of the GR2M model for the Casamance watershed. The observed flows come from the Kolda station and are provided by the DGPRE (Water Resources Management and Planning Department).

The GR2M (Génie Rural à 2paramètres Mensuel) model is a two-parameter global rainfall-runoff model. Its development was initiated at Cemagref in the late 1980s, with a view to applications in the field of water resources and low-water levels [15] .

Its structure, although empirical, resembles conceptual reservoir models, with a procedure for monitoring the state of humidity in the reservoir, which seems to be the best way of taking account of past conditions and ensuring continuous operation of the model. Its structure (Figure 2) combines a production reservoir and a routing reservoir, as well as an opening to the outside world other than the atmospheric environment. These three functions simulate the hydrological behavior of the basin.

The model has two optimizable parameters:

X1: production reservoir capacity (mm)

X2: groundwater exchange coefficient (-)

As input data to the model, we have monthly rainfall (mm), ETP in mm and observed flows in (mm/month). As output, the model provides flows that need

to be compared with observed flows in order to judge the model’s robustness for the basin under study.

Model validation is verified by comparing calculated and observed flows using a quality criterion. The best-known and most widely used criterion for conceptual models is the Nash and Sutcliff (1970) criterion, expressed by equation (1) below.

Nash ( Q ) = 100 [ 1 − ∑ i ( Q i , o b s − Q i , c a l ) 2 ∑ i ( Q i , o b s − Q o b s ¯ ) 2 ] (1)

The simulation is bad if the Nash criterion < 60%, good if it is >70% and perfect if it and equal to 100%.

Climate projection data are modeled data from the global climate model compilations of the Coupled Model Intercomparison Projects (CMIP), overseen by the World Climate Research Program [16] .

The data presented are derived from the sixth phase of the CMIPs. The CMIPs form the database for the IPCC assessment reports. For our study, the approach applied is to represent interannual and monthly variations in temperature, precipitation and ETP under the two scenarios SSP2-4.5 and SSP5-8.5 defined as follows:

The SSP2-4.5 scenario: As an update of the RCP4.5 scenario, SSP2-4.5, with an additional radiative forcing of 4.5 W/m² by 2099, represents the average trajectory of future greenhouse gas emissions. This scenario assumes that climate protection measures are taken.

The SSP5-8.5 scenario: With an additional radiative forcing of 8.5 W/m² by 2100, this scenario represents the upper limit of the range of scenarios described in the literature. It can be understood as an update of the CMIP5 RCP8.5 scenario, now combined with socio-economic trajectories.

The rainfall data series from the Kolda station, studied using KhronoStat software, reveals a random time series for the BUISHAND test that is accepted at the 99% confidence level and rejected at the 95% and 90% confidence levels (Figure 3). The PETTITT test, on the other hand, shows no break in the series studied at the various confidence levels. The Bayesian method defines the breakpoint at 1991, as does the HUBERT segmentation. This breakpoint corresponds in fact to a decrease in rainfall during that year. This observation confirms the earlier study by BODIAN [14] , who placed the breakpoint for the Kolda station between 1990-1997.

The Casamance watershed at Kolda was delineated with ARCGIS as shown in Figure 4. An area of 3448.38 km² was obtained after delimitation. This will be used for flow conversions and also for the GR2M software.

- Model parameters and efficiency criteria

To calibrate and validate the GR2M model, a series of simulations was carried out to determine the optimal model parameters X1 and X2, which will be used to simulate flows for the 2099 horizon.

The calibration and validation results shown in Table 1 below are considered to be good, since the NASH criteria obtained are greater than 60%.

- Hyetogram and hydrograph obtained during calibration and validation

Figure 5 below shows the hyetogram of monthly rainfall and the monthly hydrographs simulated and observed during the calibration period (1981-1986) and the validation period (1987-1993).

A similarity between observed and simulated flows for the calibration and validation periods has been observed, so the GR2M model applies perfectly to the Casamance watershed and can be used for future simulations.

- Temperatures

The evolution of climate forcings to 2099 in the multi-model ensemble predicts a temperature increase of 2.5˚C compared with the SSP2-4.5 scenario, and an increase of 5˚C for the SSP5-8.5 scenario, as shown in Figure 6. These temperatures will be used to calculate ETP for 2099.

- Rainfall

The evolution of climate forcings by 2099 in the multi-model ensemble predicts an increase in precipitation until 2039, followed by a gradual decrease of up to −18 mm by 2099, compared with the SSP2-4.5 scenario. This decrease intensifies

with the SSP5-8.5 scenario, with a drop of −65 mm in August by 2099 (see Figure 7).

- Evapotranspiration (ETP)

Using temperature data, we were able to project ETP to 2099. The results obtained according to the different scenarios are shown in Figure 8 and Figure 9 below.

With the rise in temperatures forecast by the two scenarios, evapotranspiration increases by up to 206 mm by 2099 in the most pessimistic scenario, SSP5-8.5. These ETPs exceed those of the 1981-2021 reference period, whose maximum value is 193 mm.

- Runoff trends to 2099

Using the parameters previously calculated by the GR2M software, we projected the evolution of runoff at Kolda for the 2099 horizon (see Figure 10 above).

Runoff on the Casamance River at Kolda will fall sharply with the SSP2-4.5 and SSP5-8.5 scenarios. From 2040 onwards, there will be no flow on the Casamance River at Kolda during non-rainy periods. In the rainy period to 2099, maximum flows will not exceed 2.5 m³/s for the SSP2-4.5 scenario and 1 m³/s for the SSP5-8.5 scenario.