The Diamniadio Urban Pole (PUD), launched in 2014, is located in Senegal’s Rufisque Department, about 30 kilometers east of Dakar. It spans the municipalities of Diamniadio, Bargny, Sendou, and Sébikhotane. It serves as the gateway to the capital, Dakar, and is situated at the intersection of National Highways 1 and 2. It is bounded to the north by the Niayes urban area, to the south by the Atlantic Ocean, to the east by the Diass urban area, and to the west by the Dakar metropolitan area [8][9]. It is a priority government project aimed at creating a sustainable and resilient city, with the goal of relieving traffic congestion in Dakar. The PUD covers an area of 1,644 hectares. It is divided into four districts and is expected to eventually accommodate 300,000 residents [8].

The project is based on public-private partnerships and is intended to serve as a showcase for a modern, forward-looking Senegal. The PUD is designed as a dense and compact city, an urban model for a new city based on sustainability and resilience. It is presented as [10][11].

Figure 1 shows the geographic location of the PUD.

Figure 1. Geographic Location of the PUD.

The Diamniadio Urban Center is located in the coastal Sahelian climate zone, characterized by specific patterns of wind, precipitation, temperature, relative humidity, and sunshine [8]. According to ANACIM, average annual temperatures range from 21 to 28˚C. The highest average maximum temperature (28.65˚C) is observed in October, while the average minimum temperature (21.1˚C) is recorded in February. In Dakar, maximum temperatures are generally observed between September and October, and minimum temperatures between January and May. Between 1970 and 2021, the average annual rainfall in the Dakar region showed significant interannual variability, with values ranging from 116.7 mm to over 660 mm. The Diamniadio Urban Center (PUD) is subdivided into three (3) watersheds, as illustrated in Figure 2.

Figure 2. Digital elevation model of the PUD.

SWAT (Soil and Water Assessment Tool) is a physics-based semi-distributed model that simulates, in a simplified manner, the water cycle in a watershed at various time scales (hourly, daily, monthly, or annually). It was developed by the USDA-ARS. SWAT was designed to assess hydrological components-including water resources, sediment production, nutrients, and pesticides-under various scenarios and at different time scales. Its operation relies on a large number of [6]:

input data: topography, soils, land use, management practices, precipitation, temperatures, sunshine, etc.;output data: streamflow, runoff volumes, soil moisture, sediment and nutrient fluxes, groundwater recharge, etc.

Figure 3 shows the hydrological structure of the SWAT model.

Figure 3.Hydrological structure of SWAT [6].

The SWAT model divides the watershed into several sub-watersheds, which are in turn subdivided into Hydrologic Response Units (HRUs) based on cropping practices, land use, and soil types. SWAT thus enables the estimation of hydrological components such as surface runoff, evapotranspiration, peak runoff discharge, sediment production, and groundwater flow [12].

The water balance simulated by SWAT is based on Equation (1) [13].

where SW_t is the soil water content at time t (mm), SW_o is the amount of water available to the plant in the soil at the initial time (mm), R_day is the precipitation at time i (mm), Q_surf is surface runoff at time i (mm), E_a is evapotranspiration at time i (mm), W_seep is percolation at timei (mm), and Q_gw is the base flow from the groundwater table at time i (mm).

The SWAT model simulates infiltration and surface runoff using two (2) methods: the Soil Conservation Service (SCS) Curve Number (CN) method and the method of Green and Ampt [14][15]. For this study, the Curve Number method is used; its expression is given by Equation (2).

where:

Q s u r f : amount of runoff (mm);

R d a y : daily precipitation (mm);

S: retention parameter (mm) defined by Equation (3).

where:

CN: Curve Number (Curve number for the day).

Furthermore, the amount of infiltrated water corresponds to the difference between precipitation and the amount of runoff. SWAT offers three methods for determining evapotranspiration, including the Penman-Monteith method (Equation 4), which was used in this study.

where 𝐸₀ is the potential evapotranspiration (mm·day⁻¹), Δ is the slope of the vapor pressure curve at a given air temperature (kPa·˚C⁻¹), H_net is the net radiation (MJ·m⁻²·day⁻¹), G is the heat flux transmitted to the ground (MJ·m⁻²·day⁻¹), ρ_air is the density of air (kg·m⁻³), C_p is the specific heat capacity of air (MJ·kg⁻¹·˚C⁻¹), e is the saturated vapor pressure of air or specific humidity (kPa), γ is the psychrometric constant (kPa·˚C⁻¹), r_c is the stomatal resistance (s·m⁻¹), r_a is the aerodynamic resistance (s·m⁻¹), and L_v is the latent heat of vaporization of water (MJ·kg⁻¹).

The SWAT model covers several areas of study, including climate, hydrology, erosion, crop growth, and water management and quality [16][17]. In this study, only the hydrology component is considered. It was configured to directly provide monthly flow values in m³/s. These values correspond to the monthly flows entered directly into SWAT-CUP without any additional processing. The model was set up using ArcSWAT 2012.10 7.26. The watersheds and hydrographic networks were delineated using the DTM, with a minimum threshold of 11.1 ha for watersheds WS1 and WS2 and a minimum threshold of 18.1 ha for watershed WS3.

The sensitivity analysis, performed prior to calibration, involves determining which model parameters most influence the output data. It can be performed manually by changing one parameter at a time or automatically using SWAT-CUP. The parameters are then ranked in descending order of sensitivity, from the highest absolute value to the lowest. The procedure consists of six steps [14]:

step 1: Use the default values for all parameters;step 2: Set all parameters except the one under study;step 3: Run the model with the minimum value of the parameter under study and save the output data;step 4: Run the model with the maximum value of the parameter under study and record the output data;step 5: Repeat steps 2 through 4 for all parameters under study;step 6: Use Equation 9 to calculate the sensitivity of each parameter for each term in the hydrologic balance.

where i is the name of the parameter being tested, C_i is the sensitivity index of the parameter being tested, and each parameter has six sensitivity indices; P consists of two values ( P min and P max ) for each parameter.

These minimum and maximum values are considered the input values for the parameter under test.

P_moy: corresponds to the average of the minimum and maximum values of each parameter to be tested;

∆𝑃: corresponds to the difference between the minimum and maximum values of each parameter to be tested;

F: consists of two values (𝐹_min and 𝐹_max) for each parameter;

𝐹_min and 𝐹_max are the respective output values of P_min and P_max ;

𝐹_moy: the average of 𝐹_min and 𝐹_min;

∆𝐹: the difference between 𝐹_min and 𝐹_max.

Table 1presents the parameters typically considered in a sensitivity analysis.

Table 1.Parameters for sensitivity analysis [18].

The parameters selected for calibration were chosen based on their hydrological relevance and the results of the global sensitivity analysis performed in SUFI-2. Their importance was assessed based on the absolute value of the t-stat and the p-value. A parameter with a high absolute t-stat value and a low p-value is considered more sensitive when the statistical significance threshold is set at 0.05 or less. This classification serves as the basis for prioritizing parameters during calibration.

Calibration involves adjusting the parameters selected following the sensitivity analysis to achieve a better match between the simulated flow rates and the regionalized reference flow rates, using quality criteria. Since the three watersheds covered by the PUD do not have hydrometric stations at their outlets, the data series used for calibration were obtained by regionalizing the observed discharge rates in the neighboring Panthior watershed, using the DAR method described in Section I.6. Validation then allows us to assess the calibrated model’s ability to reproduce hydrological behavior over an independent time period. Calibration can be performed manually or using SWAT-CUP [19]. In this study, the time periods were selected based on data availability. The simulation covers the period from 1981 to 2024. Calibration was performed using SWAT-CUP 2012 version 5.1.6 for the period 1986-1988 based on 36 regionalized monthly reference flows, and validation for the years 1991 and 1992 included 24 regionalized monthly reference flows. The period from 1981 to 1983 represents the model’s warm-up phase (NYSKIP: 3). The years following 1992—specifically 1993-2024—are then used to analyze the simulated hydrological dynamics, including soil moisture, recharge, and so on. No monthly data are missing for the periods under consideration; therefore, no interpolation or data replacement procedures were applied. SWAT-CUP (Calibration and Uncertainty Procedures) is a standalone software program associated with SWAT that is used for calibration, validation, uncertainty analysis, and sensitivity analysis. It includes five algorithms: SUFI-2, PSO, GLUE, ParaSol, and MCMC [20]. In this study, the SUFI-2 algorithm was considered. The Nash-Shutckiffe criterion (NSE) was selected as the objective function and maximized during calibration. The procedure consisted of 5 successive iterations, each comprising 500 simulations for each watershed. The total number of simulations was 2,500 per watershed, or 7,500 simulations for the three watersheds. Each iteration involved a readjustment based on the results obtained, while adhering to the physically permissible limits in SWAT. Convergence was assessed based on the stabilization of the objective function and the factors p and r.

Performance evaluation aims to verify the model’s ability to accurately reproduce observations for its application. It is based on simulated results and several quality criteria presented in Table 2 [21], [22].

Table 2. Criteria for evaluating the performance of the SWAT model [14].

Q o b s t : Observed; Q s i m t : Simulated; Q o b s ¯ : Observed mean; Q s i m ¯ : Simulated mean.

Table 2 presents the following metrics:

NSE: the Nash-Sutcliffe efficiency measure assesses the model’s ability to reproduce the observed values;R²: the coefficient of determination reflects the quality of the fit between the simulated and observed values;PBIAS: the percentage bias indicates the model’s tendency to underestimate or overestimate the observed values (depending on the convention used, a positive PBIAS indicates an underestimation, and a negative PBIAS indicates an overestimation);RMSE: the root mean square error measures the average deviation between the simulated and observed values;RSR: The ratio of the RMSE to the standard deviation of the observations corresponds to the mean squared error normalized by the standard deviation of the observations.

The data used include climate, topography, soil, hydrometric, and land-use data. The daily climate data, obtained from NASA, cover the period from 1981 to 2024 for the coordinate point 14.7329˚N and 17.2072˚W. These data, derived from MERRA-2, correspond to the grid cell centered at 14.5˚N and 17.5˚W, with a native grid resolution of 0.5˚ latitude by 0.625˚ longitude. The data include precipitation, maximum and minimum temperatures, relative humidity, and wind speed; only solar radiation is associated with a 1˚ × 1˚ radiative cell centered at 14.5˚N and 17.5˚W, derived from the SRB and CERES radiative products. Solar radiation data are missing only for the period 1981-1983 and were estimated using the Hargreaves method (Equation (10)) [28].

where R S represents solar radiation (MJ m⁻² day⁻¹), R a represents extraterrestrial radiation (MJ m⁻² day⁻¹), T max and T min represent the maximum and minimum temperatures (˚C), and k R s represents the adjustment coefficient. The adjustment coefficient is taken to be 0.19 since the PUD is located in a coastal region. Following the procedures used to determine the solar radiation data, the NASA POWER time series were not subject to any local bias correction due to the lack of a sufficiently complete ground-based meteorological time series. The absence of local correction and the spatial resolution of the data are considered sources of uncertainty. The topographic data consist of Digital Elevation Models (DEMs) derived from the USGS SRTM. The land-use data come from 1992 Landsat 4 - 5 TM C2 L2 satellite images, while the soil data are derived from the FAO’s digital soil maps. Land cover was determined through unsupervised classification of the Landsat image, which was divided into 73,843 spatial units of pixels across 6 classes. Thematic accuracy was assessed using a confusion matrix based on 100 validation points, which yielded an overall accuracy of 91% and a Kappa coefficient of 0.84. These correspond to the Panthior hydrometric station in Bounga Bambara (code 1384099005), located at 14˚45'00"N, 17˚10'12"W, which monitors a watershed covering an area of 88.5 km² with observed discharge rates in m³/s. Panthior station is not located at the outlets of the PUD basins. It belongs to a neighboring and hydrologically distinct basin. It is used as a reference due to the lack of flow measurements. The observed flows at Panthior were extrapolated to WS1, WS2, and WS3 using the drainage-area ratio (DAR) method. The Drainage-Area Ratio (DAR) method allows for the estimation of the discharge of an ungauged basin based on observations available in a neighboring basin, taking into account geographical proximity and comparable hydroclimatic conditions (Equation (11)) [29].

where Q r e f ( t ) is the regionalized reference discharge for basin i, Q p ( t ) is the observed monthly discharge, A i is the area of the PUD basins, and A p is the area of the Panthior basin. The time series obtained in this way preserve the temporal dynamics of the flows observed at Panthior, while their amplitudes are adjusted in proportion to the areas of the target basins. They are therefore referred to in this study as regionalized reference flows. The application of the DAR method assumes a relative homogeneity of specific discharge between the Panthior watershed and the PUD watersheds. Differences in area, slope, soil type, land use, and hydrological behavior nevertheless introduce uncertainty. The results therefore do not constitute an independent local hydrometric validation. The 1982 data were compiled but were not used in the performance criteria because they were from the start-up period.

All of the data used are presented in Table 3.

Table 3.Model input data.

In 1992, land cover in the PUD watersheds included water bodies, trees, flooded vegetation, croplands, bare soil, and savannas (Figure 4).

To integrate land-use data into SWAT, the classes were assigned the following codes: water bodies (WATR), trees (FRST), flooded vegetation (WETF), crops (AGRL), bare soil (BARR), and pastures (PAST).

Figure 4.Land cover—1992.

The soils in the PUD watersheds consist mainly of Cambrian Lithosols with vertic characteristics and arenosols. They thus include expansive clay and sandy formations (Figure 5).

Figure 5.Soil types in the PUD.

Slopes within the PUD are mostly gentle to moderate. The steepest slopes are found in watersheds 1 (WS1) and 3 (WS3), which are therefore more susceptible to runoff and erosion. In contrast, watershed 2 (WS2) is less steep than the other two. Figure 6 shows the slopes of the PUD watersheds.

Figure 6.Slopes of the PUD’s watersheds.

The SWAT model can be used to study numerous processes, including hydrology, climate, soil erosion, crop growth, and water management and quality. In this study, only the hydrology component is considered. Figure 7 presents the main results of the SWAT simulation: watersheds and sub-watersheds, longest hydraulic paths, hydrographic networks, and outlets.

Figure 7.Characteristics of the PUD’s watersheds obtained using SWAT.

The hydrological modeling performed using SWAT provides information on the three watersheds of the PUD, including the area of the sub-watersheds (SWS) and the hydrologic response unit (HRU) numbers. These results are presented in Table 4.

Table 4. SWAT results for the areas and HRU numbers of the PUD sub-watersheds.

The areas of WS1, WS2, and WS3 are 1259.02 ha, 328.42 ha, and 852.45 ha, respectively, for a total of 2439.89 ha, which exceeds the 1644 ha specified in the PUD. This discrepancy is due to the inclusion of certain upstream portions of the watersheds located outside the PUD that contribute to runoff. The watersheds were modeled in their entirety to preserve hydrological continuity. Using the areas of the various watersheds, it is possible to determine the conversion factors for the flow rates of the Panthior River relative to those of the PUD watersheds. Table 5 shows the conversion factors for the Panthior River’s flow rates.

Table 5.Conversion factors for the Panthior River’s flow rates.

The results of the sensitivity analysis, calibration, and validation are presented in the following sections.

3.1.1. Sensitivity Analysis

Sensitivity analysis helps identify the most influential model parameters prior to calibration. In this study, six parameters (soil, sub, hru, rte, gw, and mgt) were selected for sensitivity analysis. The SUFI-2 algorithm was used in SWAT-CUP for each of the three watersheds. The SUFI-2 algorithm was used in SWAT-CUP, with approximately 500 simulations for each of the three watersheds. For each parameter, SUFI-2 provides a t-statistic and p-value, as shown in Table 6. A high absolute value of t indicates high sensitivity, while a low p-value indicates statistically significant sensitivity. The results show that CN2 is the most influential parameter for all three watersheds. Other parameters, notably ALPHA_BN for Watershed 1 (WS1), ESCO, REVAP, CH_N2, and SOL_BD for Watershed 2 (WS2), as well as GWQMN, GW_REVAP, REVAPMN, ESCO, CH_N1, ALPHA_BNK, and GW_DELAY for Watershed 3 (WS3), also exhibit significant sensitivity. The prefix “V” indicates that the initial value of the parameter is replaced by a new one, while “R” corresponds to a relative change in the value. Figures 8-10 present the results of the overall sensitivity analysis of the SWAT-CUP parameters for watersheds WS1, WS2, and WS3.

Table 6.t-statistics and p-values for the parameters following the sensitivity analysis.

Figure 8. t-statistics and p-values for the sensitive parameters of WS1.

Figure 9.t-statistics and p-values for the sensitive parameters of WS2.

Figure 10. t-statistics and p-values for the sensitive parameters of WS3.

3.1.2. Model Calibration and Validation

Following the sensitivity analysis, the model was calibrated using SWAT-CUP based on observed discharge data from 1986 to 1988. Approximately 2500 simulations were conducted to adjust twenty parameters in order to reduce the discrepancies between observed and simulated discharges and achieve satisfactory performance. The final values of the calibrated parameters are presented in Table 7.

These values were then used to verify the model’s performance over an independent period. The validation thus covers the period 1991-1992. The performance results obtained during calibration and validation are presented in Table8. Performance was evaluated using the NSE, R², PBIAS, RSR, RMSE, p-factor, and r-factor metrics. The NSE assesses the overall quality of the simulation, while R² measures the consistency between observed and simulated flows. PBIAS indicates the model’s tendency to overestimate or underestimate flows. The RSR evaluates the normalized residual errors, and the RSME measures the mean error between simulated and observed discharge values. The p and r factors are used to assess calibration uncertainties in SWAT-CUP. For a calibration to be considered satisfactory, the p factor must be as close to 1 as possible, and the r factor as low as possible.

Table 7.Parameter values after calibration with SWAT-CUP.

Table 8. Statistical performance criteria for calibration and validation.

The calibration and validation results obtained with SWAT-CUP, as well as the 95PPU uncertainty envelope, are presented in Figure 11 for WS1, in Figure 12for WS2, and inFigure 13 for WS3.

Figure 11. Comparison of regionalized and simulated reference discharge rates with the 95PPU Uncertainty Band—calibration and validation of WS1.

Figure 12.Comparison of regionalized and simulated reference discharge rates with the 95PPU uncertainty band—calibration and validation of WS2.

Figure 13.Comparison of regionalized and simulated reference discharge rates with the 95PPU uncertainty band—calibration and validation of WS3.

The results from the hydrological modeling of the PUD using SWAT indicate that the model generally reproduces the hydrological behavior of the three studied watersheds well. The analysis is based on the delineation of watersheds and sub-watersheds, hydrographic networks, and hydrological response units.

The sensitivity analysis conducted using SWAT-CUP/SUFI-2 shows that parameter CN2 is the most sensitive across the three watersheds. This result is consistent with several SWAT applications in West Africa, notably in the Pra and White Volta basins in Ghana [30] and the Sherigu basin straddling Ghana and Burkina Faso [31]. This result highlights the importance of surface runoff, land use, and impervious surfaces in the PUD’s hydrological response. Other parameters related to groundwater, soils, evapotranspiration, and channels also influence the results, depending on the watershed.

The model’s performance ranges from satisfactory to excellent. The NSE values range from 0.82 to 0.97, while the R² values range from 0.83 to 0.98, indicating good agreement between simulated and observed discharge values. The low RSR and RMSE values also indicate that simulation errors remain limited [21].

WS1 exhibits good stability between calibration and validation, while WS2 achieves its best performance during validation. The WS3 also shows good results; however, its p-factor indicates that the 95PPU uncertainty band does not adequately represent the predictive uncertainty. The low coverage may be related to uncertainties associated with the regionalization of Panthior discharge rates, the lack of direct observations at the outlets, and the parameter ranges used. These results do not support the conclusion that the model has been validated for WS3. Thus, the watershed requires further adjustment of its sensitive parameters [32].

Overall, the SWAT model proves to be suitable for simulating the hydrological behavior of the PUD. The results allow for an examination of the influence of urbanization and impervious surfaces on runoff. Despite the satisfactory performance achieved, the model remains subject to uncertainties related to the quality and availability of observed rainfall data. Nevertheless, it can serve as a decision-support tool for stormwater management and sustainable planning of the PUD.