Evaluation of Gridded Rainfall Products in Three West African Basins

: In recent years, accessing rainfall data from ground observation networks maintained by national meteorological services in West Africa has become increasingly challenging. This is primarily due to high acquisition costs and the often sparse distribution of rainfall gauges across the region, which limits their use in hydrological studies and related research. At the same time, the rising availability of precipitation products derived from satellite/earth observations, reanalysis datasets, and in situ measurements presents exciting prospects for hydrological applications. Nonetheless, these datasets constitute indirect measurements, necessitating rigorous validation against ground-based rainfall data. This study comprehensively assesses twenty-three gridded rainfall products, including sixteen from satellites, six from reanalysis data, and one from in situ measurements, across the Senegal, Gambia, and Casamance River basins. Performance evaluation is conducted across distinct climatic zones, both pre-and post-resampling against observed rainfall data gathered from forty-nine rainfall stations over a six-year period (2003–2008). Evaluation criteria include the Kling– Gupta Efficiency (KGE) and Percentage of Bias (PBIAS) metrics, assessed at daily, monthly, and seasonal time steps. The results reveal distinct performance levels among the evaluated rainfall products. RFE, ARC2, and CPC notably yield the highest KGE scores at the daily time step, while GPCP, CHIRP, CHIRPS, RFE, MSWEP, ARC2, CPC, TAMSAT, and CMORPHCRT demonstrate superior performance at the monthly time step. During the rainy season, these products generally exhibit robustness. However, rainfall estimates derived from reanalysis datasets (ERA5, EWEMBI, MERRA2, PGF, WFDEICRU, and WFDEIGPCC) perform poorly in the studied basins. Based on the PBIAS metric, most products tend to underestimate precipitation values, while only PERSIANN and PERSIANNCCS lead to significant overestimations. Spatially, optimal performance of the products is observed in the Casamance basin and the Sudanian and Sahelian climatic zones within the Gambia and Senegal basins. Conversely, in the Guinean zone of the Gambia and Senegal Rivers, the rainfall products displayed the poorest performance.


Introduction
The United Nations' World Population Prospects Report [1] highlights a steady population growth trend in West Africa.Projections indicate a population increase from 391 million in 2019 to 796 million by 2050, reaching approximately 1.5 billion by the end of the century.This demographic increase is expected to lead to a significant rise in water demand across various sectors, including domestic supply, industry, and intensified agriculture [2,3].CHIRP v.2, CHIRPS v.2, PERSIANN-CDR, MSWEP v2.2, TAMSAT v3, ERA5, JRA-55 Adj, MERRA-2 PRECTOT, MERRA-2 PRECTOTCORR, WFDEI-CRU, WFDEI-GPCC, CPC v.1, CRU TS v. 4.00, and GPCC) to determine their reliability in simulating flows in different basins in West and Central Africa.
However, most of these studies focus primarily on hydrological validation after spatial interpolation rather than direct/pixel-by-pixel evaluation against in situ rainfall data, and little is known of their performance across climatic zones.Additionally, some studies evaluated only a limited number of products, and the value of emerging datasets such as IMERG must be evaluated and compared across multiple regions and climates.Considering the importance of the Senegal, Gambia, and Casamance River basins in West Africa and the sparse research on the performance of gridded precipitation products in the region, this paper aims to evaluate the performance of twenty-three gridded rainfall products at daily, monthly, and seasonal time steps across these three river basins.Using a large number of products, this study also builds upon an exceptional database of ground-based rainfall observations from four countries and explores the performance of gridded datasets before and after spatial resampling across different time scales.

Study Area
This study focuses on the main hydrological systems of Senegal: the Casamance, Gambia, and Senegal River basins (refer to Figure 1).The Senegal River spans latitudes 10 • 30 ′ to 17 • 30 ′ north and longitudes 7 • 30 ′ to 16 • 30 ′ west, with a length of approximately 1,800 km [46].Originating from the Bafing, Bakoye, and Falémé tributaries, sourced in the Fouta Djalon massif in Guinea, the basin covers an expansive area of 395.000 square kilometers [47].This basin extends across the high plateau region of northern Guinea, the western part of Mali, the southern regions of Mauritania, and the northern part of Senegal [48].The Gambia basin occupies the latitudinal range between 11 • 22 ′ and 14 • 40 ′  north and longitudinally between 11 • 13 ′ and 16 • 42 ′ west [49].The Gambia River stretches across 1180 km [50] and drains an extensive area of 77.054 square kilometers, shared among Guinea, Senegal, and The Gambia [51].The Casamance basin lies between 12 • 20 ′ and 13 • 21 ′ north latitude and 14 • 17 ′ and 16 • 47 ′ west longitude in the southern region of Senegal.Spanning approximately 300 km in length, the basin covers an area of 20.150 square kilometers [52].While predominantly situated within Senegalese territory, it marginally extends into parts of the Republic of The Gambia to the north and Guinea-Bissau to the south [52][53][54].
The altitude varies significantly across these basins, ranging from 0 to 83 m for the Casamance basin, 0 to 1530 m for the Gambia basin, and for the Senegal basin (Figure 1a).Climatically, total rainfall plays a pivotal role in classification [40,46].Over the period 1940-2004, the average annual rainfall (Figure 1b), according to Vintrou's [55] classification, indicates that the Casamance basin is situated within the Sudanian and Guinean climatic zones (extreme south), whereas the Gambia and Senegal basins encompass the Sahelian (extreme north), Sudanian, and Guinean (extreme south) climatic zones.Based on these climatic zones, the unimodal rainfall pattern is characterized by a wet season lasting three to six months and a dry season spanning nine to six months [16,47,50].The altitude varies significantly across these basins, ranging from 0 to 83 m for the Casamance basin, 0 to 1530 m for the Gambia basin, and for the Senegal basin (Figure 1a).Climatically, total rainfall plays a pivotal role in classification [40,46].Over the period 1940-2004, the average annual rainfall (Figure 1b), according to Vintrou's [55] classification, indicates that the Casamance basin is situated within the Sudanian and Guinean climatic zones (extreme south), whereas the Gambia and Senegal basins encompass the Sahelian (extreme north), Sudanian, and Guinean (extreme south) climatic zones.Based on these climatic zones, the unimodal rainfall pattern is characterized by a wet season lasting three to six months and a dry season spanning nine to six months [16,47,50].

Data
Two sources of data are used in this study: observed rainfall data and gridded rainfall products.

Observed Rainfall Data
The daily rainfall data utilized in this study originate from the national meteorological services of Senegal, Mali, Guinea, and The Gambia and are archived within the databases of the Senegal River Basin Organization (OMVS) and the Gambia River Basin Organization (OMVG).These datasets encompass observations from thirty-four stations within the Casamance basin, twenty-nine stations within the Gambia basin, and seventyfour stations within the Senegal basin.Figure 2 depicts the observed rainfall data collected across these three catchments.The temporal coverage of the datasets varies, spanning from 1950 to 2012 for the Casamance basin, 1950 to 2014 for the Gambia basin, and 1950 to 2019 for the Senegal basin.
Upon analysis of Figure 2, it becomes evident that the quality and continuity of these datasets differ among the basins.Among the one hundred and thirty-seven stations analyzed, only ten exhibit less than a 5% data gap, while most stations lack recent data.To

Data
Two sources of data are used in this study: observed rainfall data and gridded rainfall products.

Observed Rainfall Data
The daily rainfall data utilized in this study originate from the national meteorological services of Senegal, Mali, Guinea, and The Gambia and are archived within the databases of the Senegal River Basin Organization (OMVS) and the Gambia River Basin Organization (OMVG).These datasets encompass observations from thirty-four stations within the Casamance basin, twenty-nine stations within the Gambia basin, and seventy-four stations within the Senegal basin.Figure 2 depicts the observed rainfall data collected across these three catchments.The temporal coverage of the datasets varies, spanning from 1950 to 2012 for the Casamance basin, 1950 to 2014 for the Gambia basin, and 1950 to 2019 for the Senegal basin.
Upon analysis of Figure 2, it becomes evident that the quality and continuity of these datasets differ among the basins.Among the one hundred and thirty-seven stations analyzed, only ten exhibit less than a 5% data gap, while most stations lack recent data.To ensure consistency and reliability in the analysis, a reference period spanning from 2003 to 2008 was selected for the statistical evaluation of the rainfall products (Figure 2).This timeframe offers the advantage of possessing complete and concurrent data across all forty-nine selected rainfall stations, comprising five stations for the Casamance basin, eighteen for the Gambia basin, and twenty-six for the Senegal basin (refer to Figure 1b).Furthermore, this period aligns with the recording timeframe of the majority of available gridded rainfall products (see Table 1).
to 2008 was selected for the statistical evaluation of the rainfall products (Figure 2).This timeframe offers the advantage of possessing complete and concurrent data across all forty-nine selected rainfall stations, comprising five stations for the Casamance basin, eighteen for the Gambia basin, and twenty-six for the Senegal basin (refer to Figure 1b).Furthermore, this period aligns with the recording timeframe of the majority of available gridded rainfall products (see Table 1).Read in the "Data Sources" column: IS, In situ; S, Satellite; R, Reanalysis; A, Analysis; and in the "Start of Operation" column, P, Present.
By evaluating a diverse range of products, this study aims to provide comprehensive insights into their performance across Senegal's main hydrological systems.The coordinates of the forty-nine selected stations (refer to Figure 1b) serve as reference points for extracting gridded rainfall data.Table 1 provides a summary of the characteristics of each product, along with corresponding references for further details on their development and methodologies.

Methods
Considering the variations in spatial and temporal resolutions among the products (as outlined in Table 1), the initial step involved temporal and spatial resampling.Subsequently, a comprehensive point-by-pixel assessment was conducted utilizing two statistical criteria: Kling-Gupta Efficiency (KGE) and Percentage Bias (PBIAS).

Temporal Resampling of Products
The gridded precipitation products under assessment exhibit diverse temporal resolutions (refer to Table 1).To facilitate a comprehensive evaluation, this study assesses these products at daily, monthly, and seasonal intervals.Consequently, temporal aggregation is performed to align the products with these specific time steps, including daily, monthly, and seasonal aggregations.
For seasonal assessments, only the rainy season is considered, with its duration varying across climatic zones.Drawing from previous studies focused on characterizing the rainy season in West Africa [28,50,85], specific periods have been selected for this study.These periods are delineated as follows: three months for the Sahelian zone (July to September), five months for the Sudanian zone (June to October), and six months for the Guinean zone (May to October).

Spatial Resampling of Products
Spatial resampling serves to establish a uniform resolution for all selected gridded precipitation data, allowing for a comparative assessment against the initial datasets.Given the differing spatial resolutions among the data products (as detailed in Table 1), it is imperative to initially evaluate them at their native resolutions.This ensures a precise comparison between observed rainfall data and the corresponding pixel values at the same station.
Subsequently, an appropriate grid is defined for all products to standardize the spatial resolution.To assess the impact of this new grid on data performance, grids for each spatial resolution are extracted and overlaid onto the selected stations.Appendix A illustrates this overlay, depicting the alignment between rainfall stations and different grids.The objective is to identify the resolution that accommodates the fewest stations within the same grid.
Analysis of Appendix A reveals that only the 0.25 • × 0.25 • , 0.1 • × 0.1 • , and 0.05 • × 0.05 • grids contain fewer than three stations per grid.Among these options, the 0.1 • × 0.1 • resolution is the most common among the products (refer to Table 1).Consequently, this resolution is chosen for resampling the products, ensuring consistency in spatial representation across all datasets.
In the literature, various interpolation methods are available for spatially distributing precipitation data [86,87].These methods prove invaluable when rainfall estimates from multi-sensor source data, initially on one spatial grid, require transformation to another grid with a different spatial resolution [86].Among the commonly employed interpolation techniques are bilinear [88,89], first-order conservative [90,91], distance-weighted average [88,92,93], and nearest neighbor [94,95] interpolations.These methods, integrated into the SCRIP toolbox (Spherical Coordinate Remapping and Interpolation Package) [88], are applied in this study.Specifically, the algorithms utilized are Remapbil for bilinear interpolation, Remapcon for first-order conservative interpolation, Remapdis for distanceweighted average interpolation, and Remapnn for nearest neighbor interpolation.These algorithms can be executed using the Climate Data Operators (CDO) software v2.2.1, accessible from the Linux terminal.
In this work, the aforementioned interpolation methods are employed to resample the products to a spatial resolution of 0.1 • × 0.1 • .Consequently, products such as ARC2, RFE, MSWEP, IMERGDE, IMERGDF, and IMERGDL, already possessing this spatial resolution, are not subject to resampling.

Evaluation of Gridded Precipitation Products
In this study, the Kling-Gupta Efficiency (KGE) [96] and Percentage Bias (PBIAS) are utilized as metrics to assess the robustness of the products both before and after resampling (refer to Table 2).The KGE is an enhanced version of the Nash criterion (NSE) [97], which incorporates evaluations of correlation (r), bias (β), and variability (α) [96].KGE values range from −∞ to 1, where a value of 1 indicates a perfect fit to the in situ data [47,98,99].
PBIAS (%) On the other hand, the PBIAS represents the percentage measure of the average difference between two variables [47].PBIAS values can range from −∞ to +∞, where positive or negative values indicate overestimation or underestimation of the estimated variable, respectively.Values close to 0 suggest a good performance of the evaluated variable.KGE values of 0.5 and PBIAS values of 0 are commonly employed as reference thresholds to differentiate between the best and worst performances of the evaluated variable [47,100,101].In this study, these thresholds are defined to facilitate a more comprehensive analysis of the results obtained.The red line depicted on the graphs signifies this threshold for each evaluation criterion.
For this study, the functions of these criteria, developed within the hydroGOF package under R [102], are employed.
The KGE (Kling-Gupta Efficiency) is composed of three variables: r, the correlation coefficient; β, the biases; and α, the variability between the two evaluated variables.The PBIAS (Percentage Bias) is defined as the percentage difference between the estimated precipitation (PEst) and the observed precipitation (PObs).

Results
The study presents results based on different time steps, both on a global scale covering all three basins and at the individual hydrosystem level for a more detailed analysis.These include (i) daily, (ii) monthly, and (iii) seasonal performances of gridded precipitation products.
Furthermore, minimal discrepancy is observed among the resampling methods.Consequently, only the outcomes of the bilinear method are presented.Figures containing all the results are presented in Appendix B, Figure A1 for the daily time step, Figure A2 for the monthly time step, and Figure A3 for the seasonal time step.

Product Performance at Daily Time Intervals
In Figure 3, the comprehensive daily performance of the products before and after spatial resampling is illustrated.Analysis of the KGE values reveals that the gridded precipitation estimates exhibit suboptimal performance on a daily scale.Prior to spatial resampling, only RFE, ARC2, CPC, and MSWEP products demonstrated KGE values exceeding 0.5, while CMORPHCRT exhibited an average KGE of 0.5.Notably, spatial resampling marginally enhanced performance, with CPC, CMORPHCRT, and TAMSAT showing improvements of 3%, 4%, and 2%, respectively.
These include (i) daily, (ii) monthly, and (iii) seasonal performances of gridded precipita-tion products.
Furthermore, minimal discrepancy is observed among the resampling methods.Consequently, only the outcomes of the bilinear method are presented.Figures containing all the results are presented in Appendix B, Figure A1 for the daily time step, Figure A2 for the monthly time step, and Figure A3 for the seasonal time step.

Product Performance at Daily Time Intervals
In Figure 3, the comprehensive daily performance of the products before and after spatial resampling is illustrated.Analysis of the KGE values reveals that the gridded precipitation estimates exhibit suboptimal performance on a daily scale.Prior to spatial resampling, only RFE, ARC2, CPC, and MSWEP products demonstrated KGE values exceeding 0.5, while CMORPHCRT exhibited an average KGE of 0.5.Notably, spatial resampling marginally enhanced performance, with CPC, CMORPHCRT, and TAMSAT showing improvements of 3%, 4%, and 2%, respectively.Moreover, all these products exhibited a tendency to slightly underestimate observed rainfall both before and after resampling, as indicated by the PBIAS values.However, products such as IMERGDF, PERSIANNCCS, PERSIANN, IMERGDE, and IMERGDL displayed low KGEs.Notably, PERSIANNCCS and PERSIANN demonstrated substantial overestimations of observed rainfall.Additionally, reanalysis data exhibited limited robustness, consistently providing underestimated values compared to ground data.Among the resampling methods, the nearest neighbor method exhibited the poorest performance.
From a spatial perspective, Figures 4 and 5 depict the spatial distribution of daily KGE values before and after spatial resampling, respectively.It is evident that RFE and ARC2 exhibit reasonable performance in replicating ground rainfall patterns in the Casamance and Senegal basins prior to spatial resampling.However, the performance of these products displays spatial variability within these basins.Moreover, all these products exhibited a tendency to slightly underestimate observed rainfall both before and after resampling, as indicated by the PBIAS values.However, products such as IMERGDF, PERSIANNCCS, PERSIANN, IMERGDE, and IMERGDL displayed low KGEs.Notably, PERSIANNCCS and PERSIANN demonstrated substantial overestimations of observed rainfall.Additionally, reanalysis data exhibited limited robustness, consistently providing underestimated values compared to ground data.Among the resampling methods, the nearest neighbor method exhibited the poorest performance.
From a spatial perspective, Figures 4 and 5 depict the spatial distribution of daily KGE values before and after spatial resampling, respectively.It is evident that RFE and ARC2 exhibit reasonable performance in replicating ground rainfall patterns in the Casamance and Senegal basins prior to spatial resampling.However, the performance of these products displays spatial variability within these basins.
In the Casamance basin, RFE and ARC2 demonstrate robustness across all climatic zones, with KGE values ranging from 0.42 to 0.79.Meanwhile, MSWEP exhibits average performance in the Casamance basin.In contrast, in the Senegal River basin, RFE and ARC2 only demonstrate robustness in the Sahelian and Sudanian domains, with KGE values below 0.75.
Furthermore, the performance of CPC notably improves after spatial resampling across all three hydrosystems.Notably, the performance of gridded rainfall products varies across different climatic zones within each basin.Optimal results are observed in the Casamance basin, particularly in the Sudanian zone of the Gambia and the Sahelian and Sudanian domains of the Senegal River basin.

Product Performance at Monthly Intervals
Analysis of the KGE values depicted in Figure 6 reveals that gridded precipitation exhibits greater robustness at the monthly time step compared to the daily time step.Specifically, GPCP, CHIRPS, RFE, MSWEP, ARC2, CPC, CHIRPS, TAMSAT, and CMORPHCRT data demonstrate KGE values ≥ 0.84 at their respective initial resolutions, with relatively low error percentages.

Product Performance at Monthly Intervals
Analysis of the KGE values depicted in Figure 6 reveals that gridded precipitation exhibits greater robustness at the monthly time step compared to the daily time step.Specifically, GPCP, CHIRPS, RFE, MSWEP, ARC2, CPC, CHIRPS, TAMSAT, and CMORPHCRT data demonstrate KGE values ≥ 0.84 at their respective initial resolutions, with relatively low error percentages.Spatial resampling resulted in a 4% improvement in KGE values for CHIRPS and a 1% improvement for CHIRP, while CPC, TAMSAT, and CMORPHCRT experienced a 2% deterioration in KGE values.Notably, the GPCP product maintained the same performance as before spatial resampling.However, IMERGDF, PERSIANNCCS, PERSIANN, IMERGDE, and IMERGDL products displayed poor performance at the monthly time step.Analysis of estimation errors indicates that PERSIANNCCS and PERSIANN tend to overestimate observed rainfall.Similar results were observed for the reanalysis products.
Figures 7 and 8 display the spatial distribution of monthly KGE values before and after spatial resampling.These figures illustrate that ARC2, CHIRP, CHIRPS, CMORPHCRT, CPC, GPCP, MSWEP, RFE, and TAMSAT exhibit good agreement with observed data across all climatic domains.However, their performance varies depending on climate zones, basins, and resampling methods.Spatial resampling resulted in a 4% improvement in KGE values for CHIRPS and a 1% improvement for CHIRP, while CPC, TAMSAT, and CMORPHCRT experienced a 2% deterioration in KGE values.Notably, the GPCP product maintained the same performance as before spatial resampling.However, IMERGDF, PERSIANNCCS, PERSIANN, IMERGDE, and IMERGDL products displayed poor performance at the monthly time step.Analysis of estimation errors indicates that PERSIANNCCS and PERSIANN tend to overestimate observed rainfall.Similar results were observed for the reanalysis products.
Figures 7 and 8 display the spatial distribution of monthly KGE values before and after spatial resampling.These figures illustrate that ARC2, CHIRP, CHIRPS, CMORPHCRT, CPC, GPCP, MSWEP, RFE, and TAMSAT exhibit good agreement with observed data across all climatic domains.However, their performance varies depending on climate zones, basins, and resampling methods.In the Casamance basin, CHIRP, RFE, MSWEP, ARC2, CHIRPS, TAMSAT, CPC, CMORPHCRT, and GPCP products demonstrate robust performance both before and after spatial resampling, with KGE values ranging from 0.33 to 0.94.The most favorable outcomes were observed in the Guinean zone.
Similarly, in the Gambia basin, these products robustness both before and after spatial resampling, with KGE values fluctuating between −0.8 and 0.94.However, performance varies across different zones, with the Sudanian zone showing the best results.
In the Senegal basin, only CHIRP maintains satisfactory performance both before and after spatial resampling, with KGE values ranging from 0.47 to 0.91.Post-resampling, CHIRPS, GPCP, and TAMSAT exhibit favorable performance across the basin, while In the Casamance basin, CHIRP, RFE, MSWEP, ARC2, CHIRPS, TAMSAT, CPC, CMOR-PHCRT, and GPCP products demonstrate robust performance both before and after spatial resampling, with KGE values ranging from 0.33 to 0.94.The most favorable outcomes were observed in the Guinean zone.
Similarly, in the Gambia basin, these products exhibit robustness both before and after spatial resampling, with KGE values fluctuating between −0.8 and 0.94.However, performance varies across different zones, with the Sudanian zone showing the best results.
In the Senegal basin, only CHIRP maintains satisfactory performance both before and after spatial resampling, with KGE values ranging from 0.47 to 0.91.Post-resampling, CHIRPS, GPCP, and TAMSAT exhibit favorable performance across the basin, while CMOR-PHCRT and CPC show improved performance primarily in the Sahelian and Sudanian climatic zones.
Overall, the most notable performances are observed in the Sahelian and Sudanian climatic zones across all products.

Seasonal Product Performance
The results regarding the performance of gridded precipitation estimates during the rainy season are depicted in Figure 9. Analysis of this figure reveals that, during the rainy season, the products exhibit performances akin to those obtained at the monthly time step.
Conversely, PERSIANN, IMERGDF, EWEMBI, IMERGDE, and IMERGDL yielded the weakest results.PBIAS analysis indicates that PERSIANN and PERSIANNCCS tend to overestimate in situ rainfall during the wet season.Furthermore, during the wet season, the conservative-first and distance-weighted-average methods exhibited the best performance.
climatic zones across all products.

Seasonal Product Performance
The results regarding the performance of gridded precipitation estimates durin rainy season are depicted in Figure 9. Analysis of this figure reveals that, during the season, the products exhibit performances akin to those obtained at the monthly time Before spatial resampling, RFE, ARC2, MSWEP, CPC, CHIRP, CMORPH CHIRPS, and GPCP data provide reasonably satisfactory estimates of ground rainfal lowing spatial resampling, the KGE values of GPCP, CHIRPS, CHIRP, and CMORPH improved by 1%, 6%, 9%, 11%, and 19%, respectively, while that of CPC deteriorat 4%.Additionally, estimation errors for these products decreased.
Conversely, PERSIANN, IMERGDF, EWEMBI, IMERGDE, and IMERGDL yi the weakest results.PBIAS analysis indicates that PERSIANN and PERSIANNCCS to overestimate in situ rainfall during the wet season.Furthermore, during the wet se the conservative-first and distance-weighted-average methods exhibited the best p mance.
Figures 10 and 11 present the spatial distribution of KGE metrics during the season, both pre-and post-spatial resampling.It is evident that the products exhi favorable performance during this period correspond to those identified at the mo temporal scale, namely, ARC2, CHIRP, CHIRPS, CMORPHCRT, CPC, GPCP, MS and RFE.However, it is noteworthy that the performance of these products varies a different hydrological basins.Figures 10 and 11 present the spatial distribution of KGE metrics during the rainy season, both pre-and post-spatial resampling.It is evident that the products exhibiting favorable performance during this period correspond to those identified at the monthly temporal scale, namely, ARC2, CHIRP, CHIRPS, CMORPHCRT, CPC, GPCP, MSWEP, and RFE.However, it is noteworthy that the performance of these products varies across different hydrological basins.
In the Casamance basin, ARC2, CHIRP, CHIRPS, CMORPHCRT, CPC, GPCP, MSWEP, and RFE consistently demonstrated robust performance across all climatic zones, with the Guinean zone displaying particularly notable performance.Conversely, within the Gambia basin, these products exhibited divergent performance patterns.Only ARC2 and RFE displayed superior performance across all climatic zones, while CHIRP, CHIRPS, CMORPHCRT, CPC, GPCP, and MSWEP showcased enhanced performance specifically within the Sudanian zone.In the Casamance basin, ARC2, CHIRP, CHIRPS, CMORPHCRT, CPC, GPCP, MSWEP, and RFE consistently demonstrated robust performance across all climatic zones, with the Guinean zone displaying particularly notable performance.Conversely, within the Gambia basin, these products exhibited divergent performance patterns.Only ARC2 and RFE displayed superior performance across all climatic zones, while CHIRP, CHIRPS, CMORPHCRT, CPC, GPCP, and MSWEP showcased enhanced performance specifically within the Sudanian zone.
Within the Senegal basin, ARC2, CMORPHCRT, CPC, and RFE exhibited superior performance in the Sahelian zone.Additionally, MSWEP showcased satisfactory performance throughout the basin, with optimal performance observed within the Guinean zone.Within the Senegal basin, ARC2, CMORPHCRT, CPC, and RFE exhibited superior performance in the Sahelian zone.Additionally, MSWEP showcased satisfactory performance throughout the basin, with optimal performance observed within the Guinean zone.

Discussion
The analysis of the point-to-pixel comparison between the observed rainfall and gridded rainfall data reveals notable variability in the performance of gridded rainfall datasets contingent upon the temporal resolution and climatic domain.Notably, there is a discernible decrease in performance observed at the daily time step across the board.Beck et al. [60] and Satgé et al. [35] have suggested that such performance discrepancies stem, in part, from discrepancies between the time steps utilized for rain gauge adjustments and the temporal misalignment between rain gauge and satellite reporting times.ARC2, CPC, and RFE exhibit outstanding performance in rainfall estimation across the Casamance, Gambia, and Senegal basins.Notably, ARC2 and RFE benefit from adjustments based on observed rainfall data, while CPC relies solely on in situ measurements.This differential treatment could elucidate their superior performance.However, the reliability of in situ data also plays a crucial role in shaping the accuracy of corrected gridded rainfall estimates [35].The absence of reliable in situ measurements hampers local adjustments to satellite-derived estimates, potentially explaining the inefficacy of certain products corrected to the daily time step within the hydrological systems under scrutiny.
The findings resonate with those of Dembélé and Zwart [36] in Burkina Faso, who reported similar outcomes for ARC2 and RFE.
At the monthly temporal scale, a suite of gridded rainfall products including GPCP, CHIRP, CHIRPS, RFE, MSWEP, ARC2, TAMSAT, and CMORPHCRT demonstrate favorable agreement with observed data.Notably, the performance of CHIRPS and MSWEP resonates with findings from Satgé et al. [35] in West Africa, and similarly, Didi et al. [37] underscore the robustness of CHIRPS across five West African nations.This superior performance at monthly scales, compared to daily resolutions, can be attributed to the aggregation effect wherein errors at sub-monthly intervals, partly attributed to temporal misalignment between rain gauge and satellite reporting times, tend to offset one another upon aggregation [36].
At the seasonal scale, particularly during the rainy season, gridded rainfall datasets (RFE, ARC2, MSWEP, CPC, CHIRP, CMORPHCRT, CHIRPS, and GPCP) adequately esti-mate ground-level precipitation within the studied basins.Nonetheless, it is noted that gridded rainfall estimates are susceptible to seasonal variations, with the consistency of rainfall volumes during winter potentially mitigating errors, while the statistical weight of errors during dry seasons may exacerbate performance discrepancies [35].
From a climatic standpoint, the spatial distribution of Kling-Gupta Efficiency (KGE) values elucidates varying product performances across climatic zones within each basin.Notably, better performances are observed in the Casamance basin overall, the Sudanian zone of the Gambia basin, and the Sahelian and Sudanian climatic zones of the Senegal basin.This disparity in performance can be rationalized by the proximity of these climatic zones to the Atlantic Ocean, facilitating the circulation of colder air masses compared to inland areas.Conversely, the Guinean region in the Gambia and Senegal River basins exhibits poorer performance, attributed to its mountainous terrain and high cloud cover [30,103], which may lead to an overestimation of rainfall due to clouds "not precipitating".
Furthermore, the precipitation bias (PBIAS) analysis indicates that PERSIANN and PERSIANNCCS tend to overestimate observed rainfall, while reanalysis such as ERA5, EWEMBI, MERRA2, PGF, WFDEICRU, and WFDEIGPCC consistently underestimate values.This underestimation or overestimation is often associated with variations in complex topography, with altitude playing a significant role in influencing error magnitude [92].
Regarding spatial resampling, its impact on improving gridded precipitation product results is generally limited, with some data even experiencing performance degradation post-resampling.However, the efficacy of resampling methods varies across products and spatial-temporal scales, with bilinear (Remapbil) exhibiting the most promising results, followed by first-order conservative (Remapcon) and distance-weighted (Remapdis) methods.
The rain gauge stations used in this study are generally located near the main rivers.Thus, there are no stations with data for the period studied in certain parts of the catchment areas used in this work.This situation makes it impossible to obtain an idea of the performance of the data evaluated in these areas.The poor representativeness of the stations used to evaluate gridded rainfall data mainly concerns the Gambia and Senegal River basins.For example, no stations were used in the north-eastern part of the Senegal basin.This limits the robustness of gridded rainfall data in this area.

Conclusions
This study endeavors to assess the efficacy of twenty-three gridded rainfall products across West Africa, particularly within the Casamance, Gambia, and Senegal River basins.The methodology encompasses an evaluation of data performance pre-and postresampling, juxtaposed with observed rainfall data from forty-nine stations over a span of six years (2003)(2004)(2005)(2006)(2007)(2008) at daily, monthly, and seasonal intervals, utilizing Kling-Gupta Efficiency (KGE) and Percentage Bias (PBIAS) metrics for assessment.
The findings indicate that RFE, ARC2, and CPC demonstrate superior precipitation estimation at the daily time step, while GPCP, CHIRP, CHIRPS, RFE, MSWEP, ARC2, TAMSAT, and CMORPHCRT exhibit greater robustness at monthly intervals.Moreover, RFE, ARC2, MSWEP, CPC, CHIRP, CMORPHCRT, CHIRPS, and GPCP display optimal performance during the rainy season.Spatial analysis of KGE values underscores varying product performance across climatic zones, with the most favorable outcomes observed throughout the Casamance basin and the Sudanian and Sahelian climatic zones of the Gambia and Senegal basins, while subpar performances are typically noted in the Guinean area of the Gambia and Senegal Rivers.
PBIAS results reveal tendencies for PERSIANN and PERSIANNCCS to overestimate precipitation, whereas ERA5, EWEMBI, MERRA2, PGF, WFDEICRU, and WFDEIGPCC consistently underestimate values across spatial and temporal scales.Additionally, resampling efforts exhibit limited efficacy in enhancing product performance, although the Remapbil, Remapcon, and Remapdis methods are recommended for resampling gridded precipitation products within Senegal's primary hydrological systems.
In conclusion, the most efficient products, contingent upon temporal and spatial scales, serve as viable alternatives for diverse hydrological studies in the absence of observed rainfall data within Senegal's main hydrosystems.This study elucidates the strengths and limitations of each product across all examined scales, thereby facilitating informed decision-making for future users.Moreover, the findings provide valuable insights for data-producing organizations and institutions to enhance the accuracy of rainfall estimation algorithms within the study area.However, future endeavors may benefit from hydrological assessments to gauge the products' efficacy in simulating flows.

Figure 1 .
Figure 1.Location of the Casamance, Gambia, and Senegal River basins: (a) spatial distribution of altitudes and rainfall stations; (b) spatial distribution of selected rainfall stations and mean annual rainfall over the period 1940-2004.

Figure 1 .
Figure 1.Location of the Casamance, Gambia, and Senegal River basins: (a) spatial distribution of altitudes and rainfall stations; (b) spatial distribution of selected rainfall stations and mean annual rainfall over the period 1940-2004.

Figure 2 .
Figure 2. Inventory of daily rainfall data from stations selected for the three catchments: (a) Casamance, (b) Gambia, and (c) Senegal.The black lines on the graphs indicate the period used to evaluate the gridded rainfall products.

Figure 2 .
Figure 2. Inventory of daily rainfall data from stations selected for the three catchments: (a) Casamance, (b) Gambia, and (c) Senegal.The black lines on the graphs indicate the period used to evaluate the gridded rainfall products.

Figure 3 .
Figure 3. Overall daily performance of products before and after spatial resampling.BSR is "Before Spatial Resampling".

Figure 3 .
Figure 3. Overall daily performance of products before and after spatial resampling.BSR is "Before Spatial Resampling".

Figure 4 .
Figure 4. Spatial distribution of daily KGE values for products before spatial resampling.

Figure 5 .
Figure 5. Spatial distribution of daily KGE values for products after spatial resampling using the bilinear method (Remapbil).In the Casamance basin, RFE and ARC2 demonstrate robustness across all climatic zones, with KGE values ranging from 0.42 to 0.79.Meanwhile, MSWEP exhibits average performance in the Casamance basin.In contrast, in the Senegal River basin, RFE and ARC2 only demonstrate robustness in the Sahelian and Sudanian domains, with KGE values below 0.75.Furthermore, the performance of CPC notably improves after spatial resampling across all three hydrosystems.Notably, the performance of gridded rainfall products varies across different climatic zones within each basin.Optimal results are observed in the Casamance basin, particularly in the Sudanian zone of the Gambia and the Sahelian and Sudanian domains of the Senegal River basin.

Figure 4 .
Figure 4. Spatial distribution of daily KGE values for products before spatial resampling.

Figure 4 .
Figure 4. Spatial distribution of daily KGE values for products before spatial resampling.

Figure 5 .
Figure 5. Spatial distribution of daily KGE values for products after spatial resampling using the bilinear method (Remapbil).In the Casamance basin, RFE and ARC2 demonstrate robustness across all climatic zones, with KGE values ranging from 0.42 to 0.79.Meanwhile, MSWEP exhibits average performance in the Casamance basin.In contrast, in the Senegal River basin, RFE and ARC2 only demonstrate robustness in the Sahelian and Sudanian domains, with KGE values below 0.75.Furthermore, the performance of CPC notably improves after spatial resampling across all three hydrosystems.Notably, the performance of gridded rainfall products varies across different climatic zones within each basin.Optimal results are observed in the Casamance basin, particularly in the Sudanian zone of the Gambia and the Sahelian and Sudanian domains of the Senegal River basin.

Figure 5 .
Figure 5. Spatial distribution of daily KGE values for products after spatial resampling using the bilinear method (Remapbil).

Figure 6 .
Figure 6.Overall monthly performance of products before and after spatial resampling.BSR is "Before Spatial Resampling".

Figure 6 .
Figure 6.Overall monthly performance of products before and after spatial resampling.BSR is "Before Spatial Resampling".

Hydrology 2024 , 22 Figure 7 .
Figure 7. Spatial distribution of monthly KGE values for products before spatial resampling.Figure 7. Spatial distribution of monthly KGE values for products before spatial resampling.

Figure 7 .
Figure 7. Spatial distribution of monthly KGE values for products before spatial resampling.Figure 7. Spatial distribution of monthly KGE values for products before spatial resampling.

Figure 7 .
Figure 7. Spatial distribution of monthly KGE values for products before spatial resampling.

Figure 8 .
Figure 8. Spatial distribution of monthly KGE values for products after spatial resampling using the bilinear method (Remapbil).

Figure 8 .
Figure 8. Spatial distribution of monthly KGE values for products after spatial resampling using the bilinear method (Remapbil).

Figure 9 .
Figure 9. Overall rainy season performance of products before and after spatial resampling."Before Spatial Resampling".

Figure 9 .
Figure 9. Overall rainy season performance of products before and after spatial resampling.BSR is "Before Spatial Resampling".

Hydrology 2024 , 22 Figure 10 .
Figure 10.Spatial distribution of rainy season KGE values for products before spatial resampling.

Figure 10 .
Figure 10.Spatial distribution of rainy season KGE values for products before spatial resampling.

Figure 10 .
Figure 10.Spatial distribution of rainy season KGE values for products before spatial resampling.

Figure 11 .
Figure 11.Spatial distribution of rainy season KGE values for products after spatial resampling using the bilinear method (Remapbil).

Figure 11 .
Figure 11.Spatial distribution of rainy season KGE values for products after spatial resampling using the bilinear method (Remapbil).

Figure A2 .
Figure A2.Overall monthly performance of products before and after spatial resampling.

Figure A3 .
Figure A3.Overall rainy season performance of products before and after spatial resampling.

Table 1 .
Characteristics of the twenty-three gridded precipitation products selected.