Located in northeastern Benin, between the geographical coordinates 9˚52' and 11˚51' North latitude and 2˚34' and 3˚46' East longitude, the Sota River basin at the Coubéri outlet is a sub-basin of the Niger River and covers a total area of 13432.5 km² (Figure 1).

The basin experiences a Sudano-Sahelian climate, characterized by a unimodal rainfall regime with a single rainy season from April to October. The mean annual rainfall is 936 mm [8], and average temperatures range between 19˚C and 37˚C. Administratively, the basin spans seven municipalities: Bembèrèkè, Gogounou, Kalalé, Kandi, Malanville, Nikki, and Ségbana. The population of the Sota basin is estimated at 870,511 inhabitants, with an economy primarily based on agricultural activities and livestock farming.

Figure 1.Location of the study area and gauging stations.

The hydroclimatic data used originate from multiple sources. Daily precipitation data were obtained from nine rain gauges installed within the Sota Basin, while temperature (minimum and maximum) and potential evapotranspiration data came from the Kandi synoptic station. All observational data were provided by the Benin Meteorological Agency for the period 1970-2020. Daily discharge data at the Coubéri gauging station were sourced from the database of the General Directorate of Water (DG-Eau) for the period 1953-2017. Additionally, daily precipitation, minimum temperature, and maximum temperature from three regional climate model (RCM) were used (Table 1). These models are part of the CORDEX-Africa (AFR-44) program, which provides data at a spatial resolution of 0.44˚ and covers a geographical domain extending from 24.64˚W to 60.28˚E in longitude and from 45.76˚S to 42.24˚N in latitude [23]. The RCP4.5 and RCP8.5 scenarios were considered for the RCMs’ outputs.

Table 1. Regional climate models (RCMs) and their driving global circulation models (GCMs).

2.3.1. Characterizing Low Flows

The characterization of low flows in this study is based on the QMNA which is the lowest monthly flow observed in a given hydrological year. It is derived by calculating the monthly mean discharges and then extracting the minimum value for each year within the study period. This index (Equation (1)) is widely used in hydrology to quantify the intensity of low-flow events and capture their variability [24]-[26].

where Q m , j i is the daily discharge for dayj of month m in year i; n_m is the number of days in month m.

To detect potential changes in the QMNA time series, three statistical tests were applied:

the Pettitt test, used to identify breakpoints in the series,the Mann-Kendall test, employed to detect monotonic trends, and the Sen’s slope estimator, applied to quantify the magnitude of any detected trend.

In addition, a frequency analysis of the QMNA series was performed using four probability distributions: Weibull, Gumbel, Gamma, and Lognormal. The goodness-of-fit of these distributions was evaluated based on the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC), which allow for the selection of the most appropriate statistical model to represent the low-flow distribution. The best-performing distribution is the one associated with the lowest AIC and BIC values [27][28].

2.3.2. Climate Data Pre-Processing

Regional climate model outputs typically exhibit systematic biases that may affect hydrological simulations, thereby necessitating the application of correction methods prior to their use [29][30].

Accordingly, during the pre-processing phase, data from the RCMs were subjected to bias correction to reduce systematic deviations from observed data. To this end, the Distribution Mapping (DM) method, implemented in the CMhyd software package [29], was employed. This method adjusts the empirical distribution function of simulated data to that of observations according to Equation (2):

where x s i m denotes the raw simulated value, F s i m its cumulative distribution function, F o b s − 1 the inverse of the observed distribution, and x c o r r the bias-corrected value.

The quality of the bias-corrected series was subsequently evaluated using Taylor diagrams, which synthesize the correlation coefficient, standard deviation, and normalized root mean square error between corrected simulations and observations [31].

2.3.3. Discharge Modelling

Hydrological modeling was performed using the HEC-HMS (Hydrologic Engineering Center-Hydrologic Modeling System) software, widely employed to simulate the hydrological response of watersheds [32][33]. The conceptual framework adopted consists of a combination of sub-models tailored to the characteristics of the Sota River basin:

Canopy method: Simple CanopySurface method: Simple SurfaceLoss method: Soil Moisture Accounting (SMA)Transform method: SCS Unit HydrographBaseflow method: Constant Monthly

The model was calibrated and validated over the historical periods 1998-2006 and 2010-2016, respectively, using observed discharge data from the Coubéri gauging station. Simulation quality was assessed using the Nash-Sutcliffe Efficiency (NSE) and Root Mean Square Error (RMSE) statistical metrics [34].

Once satisfactory model performance was established, bias-corrected climate projections (RCP4.5 and RCP8.5) derived from the three RCMs were utilized as forcing data to simulate future discharges over the 2030-2080 period.

The percentage change in future mean annual discharge relative to the historical mean is given by the Equation (3):

where Q i is the future mean annual change of a given year, and Q h i s t ¯ the mean annual over the historical period.

2.3.4. Determination of Ecological Flow, Irrigation Water Requirements, and Low-Flow Augmentation Discharge

The objective of this phase was to determine, based on historical and future daily discharge time series, the capacity of a reservoir to augment low flows through controlled releases during the dry season, thereby ensuring ecological flow requirements and satisfying irrigation water demands.

Ecological flow ( Q e c o ) is defined as the discharge required to maintain aquatic ecosystem functioning at an acceptable level [35].

Following the Tennant method [36], three ecological flow scenarios were analyzed:

a minimalist scenario for aquatic ecosystem survival, wherein ecological flow corresponds to 10% of mean annual discharge, representing the threshold below which ecosystem survival is compromised; an intermediate scenario wherein ecological flow equals 30% of mean annual discharge, considered “excellent” for ecological health; and a third optimal scenario wherein environmental flow corresponds to 60% of mean annual discharge, associated with optimal conditions for ecosystem health.

The mean annual discharge used to establish the minimal, intermediate, and optimal flow thresholds according to the Tennant method was computed from the historical flow data.

Data on irrigation water requirements were extracted from the Water Master Plan for the Beninese portion of the Niger River basin [37]. Two major crops were considered: rice and tomato, with a total projected irrigated area of 7895 ha. For these two crops, the dry season production cycle (December to March) requires water demands of 11,817 m³/ha for rice and 11,600 m³/ha for tomato. The average of these two water requirements (which are close) was adopted due to the lack of detailed land use [37].

Monthly irrigation water requirements were converted to irrigation discharge Q i r r ( m ) . Thus, for each month m of the low-flow period, the target discharge Q t ( m ) , to meet irrigation and ecosystem needs, is given by:

The discharge required for low-flow augmentation ( Q r e g u l a t i o n ( m ) ) is therefore the difference between the target monthly discharge and the projected mean monthly discharge ( Q p r o j ( m ) ). It follows that:

and the volume of water required for low-flow augmentation ( V r e g u l a t i o n ( m ) )

where N d a y s ( m ) represents the number of days per month during the low-flow period and 86,400 is a constant to convert discharge in m³/s to volume in m³.

Consequently, for each ecological flow scenario, the active storage volume V u required for low-flow augmentation is derived as:

V u represents water to be supplied at the Couberi outlet, not the volume of a specific reservoir.

Afterward, the comparison between low-flow augmentation discharge and projected mean monthly discharge enables quantification of their relationship as a percentage.

The evolution of the Pettitt’s test U-statistic applied to the QMNA indices exhibits an increasing trend from 1954 to 1971, characterized by a progressive rise in statistical values up to the first breakpoint (Figure 2). This is followed by a gradual decline in the test statistic, reaching a marked minimum in 2003, before subsequently increasing again through 2016. These discontinuities correspond respectively to the major Sahelian drought of the early 1970s documented in several studies [8][22][38] and to a more recent period towards a recovery.

Figure 2. Evolution of the Pettitt test U statistic for the QMNA series.

However, Mann Kendall and Sen’s slope tests applied to the sub-periods delineated by these breakpoints indicate no statistically significant trends (Table 2), suggesting pronounced interannual variability in low flows without a clear long-term trajectory.

Table 2. Statistics from the Mann-Kendall test and Sen’s slope estimator.

From a frequency analysis perspective, the fitting of statistical distributions to observed values demonstrates that all four distributions somehow fit the QMNA series (Figure 3), although the Weibull distribution provides a superior fit to the data in terms of AIC and BIC information criteria (Table 3). This performance is consistent with previous studies on extreme flows in Sahelian regions, where the Weibull distribution is frequently favored for modeling low flow quantiles [11][39].

Figure 3.Goodness-of-fit of distribution functions to the QMNA series.

Table 3.BIC and AIC values for the distribution functions fitted to the QMNA series.

Furthermore, quantile analysis for different return periods confirms the inverse relationship between return period and low-flow severity, thereby highlighting the basin’s increased vulnerability to severe low flows during extreme events (Table 4).

Table 4.Characteristic low-flow values for different return periods.

Raw and bias-corrected precipitation and temperature simulations from the three RCMs are compared with observations using Taylor diagrams (Figure 4).

Figure 4.Taylor diagram showing bias correction performance for RCM data.

The Distribution Mapping technique proved particularly effective for correcting temperature data biases, ensuring better representation of thermal conditions, whereas precipitation showed less precise corrections. Specifically, the CCLM bias corrected data are the closest to the observations, with very similar standard deviation and high correlation coefficient (exceeding 0.95). The corrected data from the RCA4 and REMO models are somewhat more distant from observations with slightly lower standard deviations than those of observations, but correlation remains highly satisfactory (greater than or equal to 0.85), reflecting notable improvement compared to the raw data.

This disparity is common in statistical correction approaches, as precipitation is often more challenging to model accurately due to its high spatial and temporal variability.

The observed and simulated hydrographs from the HEC-HMS model for the calibration and validation periods (Figure 5) show a good reproduction of seasonal and interannual dynamics. Simulation quality is confirmed by NSE values (0.75 during calibration and 0.70 during validation). However, the model exhibits intensity biases, with a tendency to underestimate low flow at the beginning of the rainy season (e.g. 1998, 1999, 2010 and 2014) and to overestimate flood peaks in some years (2002, 2011, and 2014) and to occasionally underestimate them (2004 and 2013).

Figure 5. Hydrographs of observed and simulated discharges during calibration and validation periods.

Figure 6. Hydrographs of future projected discharges.

Bias-corrected climate data from the RCMs served as inputs to the HEC-HMS model to simulate future discharges. Inspection of these discharges over the historical period (1970-2005) shows that all three datasets adequately simulate the seasonal pattern of observed flow, with low flows between December and May and flood peaks in September (Figure 6). Irrespective of the RCM considered, low flows are well simulated while flood peaks are underestimated under the CCLM and RCA4 models.

Figure 7.Interannual variability of future discharges under RCP 4.5 and RCP 8.5 scenarios.

Projections from the CCLM and REMO models indicate a general declining trend in future annual discharges, with reductions ranging from −29% to −34% relative to the reference period (Figure 7). This trend toward markedly lower annual flows corroborates projections of increasing water stress reported for other West African basins [15][33], posing significant challenges for meeting agricultural, domestic, and ecological water demands. In contrast, projections from the RCA4 RCM suggest an increase in annual discharge of up to +12% (RCP4.5) and +31% (RCP8.5). This inter-model discharge variability stems from well-documented uncertainties in projecting the frequency and magnitude of extreme rainfall events, which propagate in simulated flows [1][40].

Table 5 presents low-flow augmentation requirements according to the three ecological flow scenarios considered in this study.

Table 5. Discharge and active storage volume for low-flow augmentation according to ecological flow maintenance scenarios, climate models, and climate scenarios considered in this study.

a. RCP 4.5 b. RCP 8.5.

Results reveal substantial differences among RCMs regarding future low-flow augmentation requirements. The CCLM and REMO models project higher requirements, with augmentation discharges ranging from 6 to 20 m³/s depending on the augmentation level. Conversely, RCA4 indicates lower requirements ranging from 5 to 14 m³/s, reflecting less severe low flows in its projections. Nevertheless, our results align with the work of [41] on several rivers with environmental flow estimated at 20 m³/s.

The augmentation discharges obtained in this study correspond to a percentage of mean annual discharge varying between 13% and 30% for the minimalist scenario, 17% and 46% for the intermediate scenario, and 32% and 91% for the optimal scenario. These proportions fall within the ranges observed in other basins with similar characteristics. For instance, the International Water Management Institute [42] reports that 60-80% of the mean annual flow is needed to maintain a river in a natural state. Assessing the sustainability of the lower Limpopo River basin, [43] found that environmental flows correspond to approximately 50%, 39%, 27%, and 14% of the mean annual flow for excellent, medium, poor, and degraded scenarios, respectively. [44] using five methods including Tennant’s method, indicated that 46% - 71% of the mean annual flow is necessary to meet environmental needs during low-flow periods. In contrast, [45] estimated environmental flows at 20% - 30% of the mean annual flow during low-flow months for the Amu Darya River, and [46] estimated the minimum ecological flow for the Hulan River at 10% of the natural flow.

In terms of storage volume required for augmentation, results from this study range from 55 to 73 × 10⁶ m³/year for minimal augmentation, 116 to 164 × 10⁶ m³/year for moderate augmentation, and 220 to 321 × 10⁶ m³/year for optimal augmentation. According to the Water Master Plan for the Niger basin in Benin, the mean annual volume for the Sota basin at Coubéri is estimated at 946.1 × 10⁶ m³/year [37], implying that the augmentation volumes estimated in this work represent less than one-third of the total annual available water resource. Therefore, the tested regulation strategies remain hydrologically realistic, but the primary challenge lies in the seasonal variability of flows.

Moreover, the analysis of the drying climate projections (CCLM and REMO) indicates that wet-season water availability is sufficient to fully meet the storage requirements for both minimal and intermediate flow augmentation scenarios across all years. For the optimal flow augmentation scenario, storage capacity cannot be met in a small subset of exceptionally dry years: in up to 12% of years under the CCLM_RCP4.5 projection and fewer than 6% of years in the other model-scenario combinations. From an operational perspective, targeting the intermediate flow regulation scenario would be the prudent management strategy during these extreme dry years.

Overall, these results indicate that low-flow augmentation is largely feasible even under drying climate conditions, with only limited risk of failing to meet the most ambitious (optimal) scenario.