The study was conducted in the Senegalese part of the Senegal River Basin, located between 14˚ and 17˚N and between 13˚ and 17˚W. The Senegal River Basin is the largest and most water-abundant river basin in the country. Its Senegalese part covers approximately 76,618 km², representing more than one-third of the national territory [22].

The quantitative assessment of irrigation water requirements focuses on the five sub-units of the Senegal River Valley Management and Planning Unit shown in Figure 1.These sub-units, referred to as sub-UGP, comprise the Senegal Delta, the Middle Senegal Valley, Lower Ferlo, Thiangol-Louguéré and Upper Ferlo. They cover approximately 63,352 km², calculated from the spatial boundaries used in this study, and contain the main irrigated areas in the Senegalese part of the basin. The Falémé sub-unit, which drains the southeastern edge of the basin and belongs to the Eastern Senegal Management and Planning Unit, was excluded because it contains little irrigated land.

The study area is characterised by a marked south-to-north rainfall gradient. Precipitation is highest in the southeastern part of the basin and decreases towards the northern and northwestern regions, where annual totals can fall below 200 mm [23][24]. The rainfall regime is unimodal, with precipitation mainly concentrated between June and October and a long dry season during the remainder of the year [24][25]. Low and highly irregular rainfall constrains rainfed agriculture and reinforces the role of irrigation in agricultural production [8][21].

Figure 1.Location of the study area in the Senegalese part of the Senegal River Basin. The map shows the five sub-units of the Senegal River Valley Management and Planning Unit retained for the assessment of irrigation water requirements. It also shows irrigated areas, water bodies, synoptic stations operated by the National Agency for Civil Aviation and Meteorology, ANACIM, and stations located within irrigated perimeters.

Daily climate data covering the 1984-2025 period were obtained from two gridded climate products: AgERA5 and NASA POWER. AgERA5 is an agrometeorological dataset derived from the ERA5 reanalysis, with a spatial resolution of 0.1˚, equivalent to approximately 10 km [26]. NASA POWER meteorological data are primarily derived from the MERRA-2 reanalysis and have a spatial resolution of 0.5˚ × 0.625˚ [13][27]. The variables used include minimum and maximum air temperature, relative humidity, wind speed, solar radiation and precipitation.

Observational data were obtained from the eight meteorological stations presented in Table 1. These comprise five synoptic stations operated by the National Agency for Civil Aviation and Meteorology, ANACIM, and located in open environments, as well as three stations installed within irrigated perimeters in the Saint-Louis region. The observation periods vary among stations. Each climate product was therefore compared with observations over the station-specific common period. All variables were harmonised beforehand in accordance with the FAO-56 requirements.

Table 1.Characteristics of the ground-based meteorological stations used to evaluate reference evapotranspiration (ET₀).

Reference evapotranspiration was estimated using the FAO-56 Penman-Monteith equation [10]:

where Δ is the slope of the saturation vapour pressure curve in kPa·˚C⁻¹, R n is the net radiation in MJ m⁻²·day⁻¹ and G is the soil heat flux density in the same unit. T is the mean air temperature in ˚C, γ is the psychrometric constant in kPa·˚C⁻¹, U 2 is the wind speed at 2 m in m·s⁻¹ and e s − e a is the vapour pressure deficit in kPa. All parameters were calculated following the standard FAO-56 procedure [10].

AgERA5 wind speed, available at 10 m, was adjusted to 2 m using the logarithmic wind profile recommended by FAO-56. A conversion factor of 0.748 was applied. NASA POWER wind speed was directly available at 2 m.

At the daily time step, soil heat flux was assumed to be negligible and set to zero, in accordance with the FAO-56 recommendations [10]. For the monthly ET₀ calculated from the ANACIM synoptic station data and used for validation, soil heat flux was estimated from the change in monthly mean air temperature.

Effective precipitation was estimated using the FAO and AGLW method implemented in CROPWAT [28]:

where P e is the monthly effective precipitation and P is the total monthly precipitation, both expressed in millimetres. Negative values of P e were set to zero. For the calculation of irrigation water requirements, monthly effective precipitation was then distributed to the daily time step in proportion to the precipitation observed on each day. The FAO/AGLW method is defined at the monthly scale because effective precipitation represents the fraction of monthly rainfall stored in the root zone and available to the crop, a quantity that is not meaningful at the daily scale. To feed the daily requirement chain while preserving this monthly balance, the monthly effective precipitation was distributed over the days of the month in proportion to the observed daily rainfall, so that the daily values sum to the monthly effective total and follow the actual temporal distribution of rain.

Crop evapotranspiration was calculated using the following equation:

where K c is the crop coefficient corresponding to the crop type and growth stage. Crop coefficients for irrigated lowland rice have been determined from field measurements under semi-arid Sahelian conditions [29].

Net irrigation water requirements were calculated as follows:

Net irrigation requirements were calculated at the daily time step. When effective precipitation exceeded crop evapotranspiration, the resulting negative values were set to zero before aggregation over the crop cycle, and excess effective precipitation was not carried over to subsequent days. Following the FAO-56 crop-coefficient approach, these requirements represent the water needed to meet crop evapotranspiration and do not include land preparation, initial ponding, deep percolation or drainage specific to lowland rice. These excluded components depend primarily on soil hydraulic properties, field conditions and irrigation management rather than on the climate product considered; since the study compares two climate products, restricting the requirement to its evapotranspiration-driven component avoids introducing variability unrelated to the comparison. On the heavy clay soils that dominate the delta rice areas, measured percolation is moreover modest (1.5 - 2 mm·day⁻¹; Raes et al. [30]). This choice is consistent with the regional assessments of Diop et al. [20] and Djaman et al. [31], and the rice requirements represent an evapotranspiration-based lower bound of the total flooded-field requirement. Gross irrigation water requirements were then calculated using the following equation:

where E i is irrigation efficiency. Irrigation efficiency varies among schemes with the method, canal condition and management, and measured project efficiencies in the valley range from 48% to 61% [30]. In the absence of scheme-specific measured values across the study area, a single reference value of 0.70 was adopted, the default recommended by FAO for well-managed gravity irrigation [27] and consistent with the value used for the Senegal River Delta by Diop et al. [20]. It was applied uniformly to ensure comparability across crops, seasons and products. Because gross requirements scale inversely with efficiency, adopting the lower measured efficiencies would raise the reported values by about 15% to 46%; as a common factor applied identically to both products, the efficiency affects only the absolute level of the requirements and not the comparison between AgERA5 and NASA POWER. The cropping seasons, sowing windows and crop parameters used in the study are presented in Table 2 and Tables 3.

Table 2.Cropping seasons, crops, cycle lengths and sowing windows considered in the study.

Table 3.Parameters of the main crops considered in the study.

L represents the duration of the growth stage. The subscripts ini, dev, mid and late refer to the initial, development, mid-season and late-season stages, respectively. K_c is the crop coefficient. Sources: Doorenbos and Pruitt [29], Allen et al. [10] and FAO [4].

The performance of AgERA5 and NASA POWER was evaluated using eight ground stations divided into two complementary groups. Five ANACIM synoptic stations located in open environments were used for validation: Saint-Louis, Louga, Matam, Kédougou and Tambacounda. Matam was retained but analysed separately because wind speed was estimated rather than measured. Three stations located within irrigated perimeters, namely Fanaye, Ndiaye, and Diama, were used to assess the spatial representativeness of the products.

Comparisons were conducted at the monthly scale, which was the only temporal scale common to all records. At the ANACIM stations, ET₀ was calculated directly at the monthly scale using the FAO-56 Penman-Monteith method and accounting for monthly soil heat flux [10]. At Fanaye, Ndiaye and Diama, ET₀ was calculated at the daily scale from the observed meteorological variables and then aggregated into monthly means. The eight retained stations differ only in the time step of the FAO-56 computation, monthly at the ANACIM synoptic stations and daily aggregated to monthly at the three perimeter stations a difference of minor magnitude, the Penman-Monteith equation yielding similar estimates at the monthly and daily steps [10]; all reference values are computed with measured Penman-Monteith. For each climate product, daily ET₀ calculated using the FAO-56 Penman-Monteith method was extracted from the grid cell nearest to each station and then aggregated into monthly means. Estimates from the products and stations were therefore compared at the same temporal scale. For each station, the evaluation period corresponds to the common period between the available observations and the climate products. These periods are presented in Table 1.

Relative humidity was directly available in both products and was not derived, and solar radiation was converted to MJ·m⁻²·day⁻¹ where required. The station records were screened with physical-range checks on each variable, and values outside plausible bounds were discarded; the Ndiaye series excludes 2019-2020 because of a humidity-sensor drift, and days with fewer than 20 hourly values were removed at Diama. The number of monthly values retained for each station and season is reported in Table 4. Monthly means were computed from the quality-checked daily values available in each month, and station-season combinations with fewer than five monthly values were not evaluated.

Performance was evaluated separately for each season using the root mean square error, normalised root mean square error, percent bias and Pearson correlation coefficient.

where P i and O i represent the ET₀ estimated from the climate product and calculated from station observations at time i , respectively. n is the number of observations and O ¯ is the mean of the station-derived values.

Percent bias was used to quantify systematic overestimation or underestimation:

A positive value indicates overestimation, while a negative value indicates underestimation.

The Pearson correlation coefficient measures agreement in temporal variability between the estimates and observations. It was calculated as follows:

where P ¯ is the mean of the values estimated from the climate product.

AgERA5 estimates were mapped on their native 0.1˚ grid, which represents spatial variations within the sub-UGP. Given its resolution of 0.5˚ × 0.625˚, NASA POWER provides only a limited number of grid cells across the study area. Its estimates were therefore calculated as area-weighted zonal means for each sub-UGP. For each unit, the zonal mean was computed as a cosine-latitude-weighted average of the grid cells assigned to that unit by the management-unit mask, each cell being assigned to the unit containing its centre; basin-scale means were computed in the same way over the five units.

The two products were compared at the common spatial scale of the sub-UGP [32]-[35]. AgERA5 estimates were therefore aggregated to the same scale as NASA POWER estimates. This scale corresponds to the units used for irrigation planning and management, and errors associated with the spatial aggregation of climatic variables generally remain limited [36].

For each sub-UGP, the difference was calculated by subtracting the NASA POWER estimate from the AgERA5 estimate. Positive values therefore indicate higher requirements with AgERA5, while negative values indicate higher requirements with NASA POWER. This approach allows the two products to be compared and the spatial variability of irrigation water requirements to be analysed across sub-UGP and cropping seasons.

A global sensitivity analysis was conducted using the variance-based Sobol method [37]. This method partitions the total variance of the output into the individual effects of the input variables and their interactions. It is widely used in agrohydrological modelling to identify the variables that contribute most to output variability [38]-[40]. It has been applied in particular to analyse the climatic factors controlling crop yield and crop water use [41][42].

For a model with k input variables, the total output variance V ( Y ) is expressed as follows:

where V i represents the individual contribution of variable i , V i j represents the interaction between variables i and j , and V 1 , 2 , ⋯ , k represents higher-order interactions.

The first-order index of variable i was calculated as follows:

The S i index measures the direct contribution of variable i to the output variance. The total-order index accounts for this direct contribution and all interactions involving variable i . It is expressed as follows:

where V i represents the contribution to variance of all terms that do not involve variable i . A small difference between S i and S T i indicates limited interactions, whereas a large difference indicates a stronger contribution from interactions [39].

Sampling was performed using the Saltelli scheme [43], with a base sample size of 512 and the calculation of second-order indices. Three climatic variables were analysed: mean air temperature, relative humidity, wind speed and solar radiation. The daily temperature range was held at its median value. With three input variables, the sampling design required 5120 model evaluations.

For each sample, ET₀ was recalculated using the FAO-56 Penman-Monteith method. The climatic variables were sampled independently between their 5th and 95th percentiles. These limits were determined separately for each climate product, sub-UGP and cropping season. The resulting indices therefore represent ET₀ sensitivity within the range of variation specific to each climatic context.

The performance of AgERA5 and NASA POWER in estimating monthly ET₀ varied according to season and station type. PBIAS, NRMSE and the correlation coefficient were calculated for each station and cropping season. Detailed results are presented in Table 4.

At the three synoptic stations with measured wind speed, AgERA5 generally provided the estimates closest to the observations. Its PBIAS ranged from −5% to +13% during the hot dry season, from +4% to +14% during the cool dry season and from −9% to +3% during the rainy season. NRMSE ranged from 7% to 17% across all seasons. NASA POWER showed larger deviations, particularly during the hot dry season, with PBIAS ranging from +4% to +29% and NRMSE from 13% to 31%. Its PBIAS decreased during the cool dry and rainy seasons, ranging from +4% to +17% and from −4% to +18%, respectively. Correlation coefficients varied substantially according to station and season, with neither product showing a consistent advantage. The results for Matam were considered separately because wind speed was estimated rather than measured. At this station, PBIAS ranged from +10% to +25% for AgERA5 and from +4% to +20% for NASA POWER, depending on the season. Correlations were high during the hot dry and rainy seasons but decreased markedly during the cool dry season.

At the stations located within irrigated perimeters, both products systematically overestimated observed ET₀. However, the magnitude of this overestimation varied according to station and season. During the cool dry season, PBIAS ranged from +26% to +76% for AgERA5 and from +8% to +54% for NASA POWER, with the highest values observed at Fanaye. During the hot dry season, PBIAS ranged from +19% to +40% for AgERA5 and from +17% to +41% for NASA POWER. The deviations generally decreased during the rainy season, ranging from +7% to +32% for AgERA5 and from +3% to +36% for NASA POWER.

Table 4.Performance of AgERA5 and NASA POWER in estimating monthly ET₀ by station, climate product and cropping season. Seasonal performance is reported using PBIAS, NRMSE and the correlation coefficient.

Correlations marked with an asterisk are not significant at p < 0.05.

These results reveal a clear contrast between the synoptic stations located in open environments and those installed within irrigated perimeters. Both products performed satisfactorily at most synoptic stations but showed greater overestimation of ET₀ within irrigated perimeters, particularly during the dry seasons.

Gross irrigation water requirements varied with season, crop cycle length and sowing date, as shown in Figure 2 and Figure 3 and Table 5. During the hot dry season, mean basin-scale requirements for rice increased from approximately 15,800 to 19,600 m³·ha⁻¹ with AgERA5 and from 17,200 to 20,900 m³·ha⁻¹ with NASA POWER between the 115-day and 150-day cycles. For the 115-day cycle, the AgERA5 estimate was close to the reference dose of 16,000 m³·ha⁻¹, whereas the NASA POWER estimate exceeded this value. For the 125-day and 150-day cycles, estimates from both products exceeded the reference dose.

During the hot dry season, irrigation requirements generally decreased with delayed sowing. Correlation coefficients ranged from −0.74 to −0.98. The difference between the least and most water-demanding sowing dates ranged from 905 to 3672 m³·ha⁻¹, equivalent to 5.7% to 17.6% of the mean requirement. This difference increased with cycle length and was consistently larger with NASA POWER. Maximum requirements occurred for sowings conducted at the beginning of February or during its first three weeks, whereas minimum requirements occurred for sowings at the end of March.

During the rainy season, mean requirements ranged from approximately 10,000 to 13,800 m³·ha⁻¹ with AgERA5 and from 8300 to 11,900 m³·ha⁻¹ with NASA POWER. At basin scale, they remained below the reference dose of 12,000 m³·ha⁻¹, except for the 150-day cycle with AgERA5. Early-season sowings showed the lowest requirements. Delayed sowing increased irrigation requirements, with correlation coefficients ranging from +0.92 to +0.96. The amplitude of variation ranged from 1352 to 2360 m³·ha⁻¹, equivalent to 11.1% to 21.1% of the mean requirement. Minimum values occurred during the first half of July and maximum values at the end of August.

Figure 2.Variation in gross irrigation water requirements for rice according to sowing date during the hot dry and rainy seasons, at basin scale and across the five sub-UGP. Solid lines represent AgERA5 and dashed lines represent NASA POWER.

During the cool dry season, mean requirements were approximately 11,000 m³·ha⁻¹ for onion and ranged from 12,900 to 13,100 m³·ha⁻¹ for tomato. They exceeded the reference dose of 9000 m³·ha⁻¹ with both products. Early-season sowings showed the lowest requirements. Delayed sowing led to an almost linear increase in irrigation requirements between 1 October and 30 November, with correlation coefficients equal to or greater than 0.999. The amplitude of variation ranged from 1411 to 2833 m³·ha⁻¹, equivalent to 12.8% to 21.8% of the mean requirement.

Figure 3.Variation in gross irrigation water requirements for onion and tomato according to sowing date during the cool dry season, at basin scale and across the five sub-UGP. Solid lines represent AgERA5 and dashed lines represent NASA POWER. The shaded area indicates the sowing window, and the horizontal line indicates the SAED reference dose.

Table 5.Variation in gross irrigation water requirements according to sowing date, season, crop and climate product at basin scale.

Across all sub-UGP, crops and seasons, irrigation water requirements ranged from 7158 to 22,269 m³·ha⁻¹. At basin scale, NASA POWER provided higher estimates during the hot dry season, whereas AgERA5 produced higher values during the rainy season. Depending on the crop and season, the effect of sowing date was comparable to or greater than the difference between the two products. The amplitude associated with sowing date was consistently larger with NASA POWER.

Irrigation water requirements computed with AgERA5 are represented on its native 0.1˚ grid, whereas NASA POWER estimates are aggregated across the five sub-UGP. The spatial distributions obtained for the different cropping seasons are presented in Figures 4-6.

During the hot dry season, irrigation requirements increase with crop cycle length in all sub-UGP. With AgERA5, they range from 15,295 - 17,148 m³·ha⁻¹ for the 115-day cycle to 18,682 - 21,549 m³·ha⁻¹ for the 150-day cycle. With NASA POWER, the corresponding ranges increase from 15,196 - 18,054 to 18,764 - 22,269 m³·ha⁻¹. For both products, the requirements of the 125-day cycle lie between those of the other two cycles.

Figure 4.Spatial distribution of gross irrigation water requirements for rice during the hot dry season for the 115-day, 125-day and 150-day cycles. The first column presents AgERA5 estimates on the native 0.1˚ grid. The second presents NASA POWER estimates aggregated by sub-UGP. The third presents the difference between AgERA5 and NASA POWER.

The highest requirements are mainly observed in the Middle Senegal Valley and Thiangol-Louguéré. The lowest values occur in Upper Ferlo with AgERA5 and in the Senegal Delta with NASA POWER. NASA POWER provides higher estimates than AgERA5 in all sub-UGP except the Delta. The mean difference, calculated as AgERA5 minus NASA POWER, ranges from +244 m³·ha⁻¹ in the Delta to −2125 m³·ha⁻¹ in Upper Ferlo. The location of the maximum depends on the cycle length and climate product. For the 115-day and 125-day cycles, AgERA5 places the maximum in the Middle Senegal Valley, whereas NASA POWER places it in Thiangol-Louguéré. For the 150-day cycle, both products place the maximum in the Middle Senegal Valley.

During the rainy season, irrigation requirements are markedly lower but also increase with crop cycle length. With AgERA5, they range from 8989 - 11,616 m³·ha⁻¹ for the 115-day cycle to 12,649 - 15,583 m³ m·ha⁻¹ for the 150-day cycle. With NASA POWER, the corresponding ranges increase from 7158 - 9876 to 10,811 - 13,547 m³·ha⁻¹. AgERA5 provides higher estimates than NASA POWER in all sub-UGP, with mean differences ranging from 1523 to 2044 m³·ha⁻¹. Despite this systematic difference, both products produce the same spatial ranking for all three cycles. The highest requirements occur in the Middle Senegal Valley and the lowest in Upper Ferlo.

Figure 5.Spatial distribution of gross irrigation water requirements for rice during the rainy season for the 115-day, 125-day and 150-day cycles. The first column presents AgERA5 estimates on the native 0.1˚ grid. The second presents NASA POWER estimates aggregated by sub-UGP. The third presents the difference between AgERA5 and NASA POWER.

During the cool dry season, differences among the sub-UGP are less pronounced. Onion requirements range from 10,543 to 11,456 m³·ha⁻¹ with AgERA5 and from 10,388 to 11,351 m³·ha⁻¹ with NASA POWER. Tomato requirements range from 12,386 to 13,409 m³·ha⁻¹ and from 12,214 to 13,573 m³·ha⁻¹, respectively.

The two products provide similar estimates of the overall level of irrigation requirements, but their spatial distributions differ. AgERA5 produces higher values in the Delta and the Middle Senegal Valley, whereas NASA POWER provides higher estimates in Lower and Upper Ferlo. With AgERA5, the maximum occurs in the Middle Senegal Valley and the minimum in Lower Ferlo. With NASA POWER, the maximum occurs in Upper Ferlo and the minimum in the Delta.

Figure 6.Spatial distribution of gross irrigation water requirements for onion and tomato during the cool dry season. The first column presents AgERA5 estimates on the native 0.1˚ grid. The second presents NASA POWER estimates aggregated by sub-UGP. The third presents the difference between AgERA5 and NASA POWER.

Overall, the intensity of spatial variability depends on the season. The range among the sub-UGP represents approximately 48% of the mean requirement during the rainy season, compared with 17% during the hot dry season and 10% during the cool dry season. The mean absolute difference between AgERA5 and NASA POWER is approximately 997 m³·ha⁻¹, equivalent to 7% of the mean requirement. Both products therefore reproduce the same spatial contrasts during the rainy season, whereas the ranking of the sub-UGP is less consistent during the dry seasons. The mechanisms underlying these variations and the effects of spatial resolution are examined in the Discussion.

The sensitivity analysis quantified the contribution of each climatic variable to ET₀ variability and assessed the importance of interactions according to season and climate product. The results are presented in Figure 7.

Figure 7.Sobol sensitivity indices of reference evapotranspiration to wind speed (u₂), relative humidity (RH), mean air temperature (Tmean) and incoming solar radiation (Rs) by season and climate product. Panels (a) and (b) present the first-order indices (Sᵢ), and panels (c) and (d) the total-order indices (ST). Panels (a) and (c) correspond to AgERA5, and panels (b) and (d) to NASA POWER. Values are averaged across the five sub-UGP.

The direct effects of the three climatic variables jointly explain between 94% and 98% of the total variability in ET₀. Interactions among these variables therefore account for only the remaining 2% to 6%. Their limited contribution is also confirmed by the similarity between the first-order and total-order indices, which differ by no more than 0.06. During both dry seasons, wind speed is the main factor controlling ET₀ variability, with a mean index of 0.42. During the hot dry season, it is followed by relative humidity with an index of 0.29, mean air temperature with 0.22 and solar radiation with 0.04. A similar hierarchy is observed during the cool dry season, with indices of 0.35 for relative humidity, 0.15 for mean air temperature and 0.04 for solar radiation.

The hierarchy changes during the rainy season. Relative humidity becomes the main factor, with a mean index of 0.37, followed by solar radiation with 0.30. The contribution of wind speed decreases to 0.23, whereas that of mean air temperature remains low at 0.07. At the annual scale, relative humidity remains the main factor controlling ET₀ variability, with an index of 0.47. It is followed by wind speed with 0.26, mean air temperature with 0.14 and solar radiation with 0.08.

The two products show the same ranking of factors during the dry seasons and at the annual scale. However, a difference appears during the rainy season. With NASA POWER, relative humidity dominates solar radiation, with respective indices of 0.40 and 0.27. With AgERA5, the two variables make equal contributions, with an index of 0.33 each. Both products therefore show that the importance of wind speed decreases during the rainy season. During this season, relative humidity becomes the main factor with NASA POWER, whereas relative humidity and solar radiation contribute equally with AgERA5.

At the synoptic stations located in open environments, AgERA5 and NASA POWER reproduce monthly ET₀ satisfactorily overall. AgERA5 provides the estimates closest to the observations, with small seasonal biases and NRMSE values between 10% and 13%. The larger deviations observed with NASA POWER during the hot dry season may be related to its coarser spatial resolution and to the representation of the atmospheric variables used in the Penman-Monteith equation. Nevertheless, both products show their smallest biases and normalised errors during the rainy season.

The lowest, and in several cases non-significant, correlation coefficients are obtained in the cool dry season, and at some stations in the rainy season. This does not necessarily indicate poorer accuracy. The limited month-to-month variability of ET₀ in these seasons reduces the ability of the correlation coefficient to capture temporal variations, so that a weak or non-significant correlation may partly reflect the limited temporal variability of ET₀ and the available sample size and should not, on its own, be interpreted as evidence of poor product performance. PBIAS and NRMSE are therefore more appropriate for evaluating performance in these seasons.

Deviations are markedly larger at stations located within irrigated perimeters, particularly during the dry seasons. This contrast with the synoptic stations is consistent with an oasis effect. Irrigation locally maintains cooler and more humid conditions than those in the surrounding areas. Stations located within irrigated perimeters may therefore record lower evaporative demand than that represented by the climate-product grid cell, which reflects regional conditions [44][45]. The reduction in deviations during the rainy season supports this interpretation because the contrast between irrigated areas and their surroundings becomes less pronounced. However, the oasis effect cannot explain all the observed deviations. Differences between point measurements and gridded estimates may also arise from spatial resolution, the distance between a station and the centre of the corresponding grid cell, and uncertainties in wind speed [31], relative humidity and solar radiation. These variables directly influence ET₀ estimated using the Penman-Monteith method [10]. Previous studies have also shown that the performance of gridded climate products may decline in semi-arid environments and during periods of high evaporative demand [16][17][19][46].

The performance obtained at the synoptic stations indicates that AgERA5 and NASA POWER can complement observational networks in regional analyses of ET₀. The finer resolution of AgERA5 allows spatial variations within the sub-UGP to be represented in greater detail, although finer resolution alone does not guarantee higher accuracy at the local scale. For operational irrigation scheduling at the local or field scale, both products should be evaluated against observations representative of the intended scale of application. Bias correction or local calibration may be required where substantial deviations occur.

Irrigation water requirements depend on the season, the cycle length, the sowing date and the location. The highest values are observed during the hot dry season, particularly for the 150-day rice cycle, because of the high evaporative demand and the small contribution of effective precipitation. The increase in requirements with cycle length is consistent with the FAO-56 approach, since cumulative requirements depend on crop evapotranspiration over the whole cycle [10]. During this season, requirements generally decrease as sowing is delayed. This decrease does not, however, always correspond to an achievable saving, because late sowings extend the cycle towards the rainy season and may encroach on the following campaign. When only the cycles completed before the start of that campaign are retained, the minimum requirements increase by 690 to 1660 m³·ha⁻¹ for the 115- and 125-day cycles; for the 150-day cycle, only sowing on 1 February satisfies this constraint. In the rainy season the relationship is reversed: early-season sowings show the lowest requirements, whereas delayed sowing shifts an increasing part of the cycle beyond the wettest period. Early sowings therefore make better use of rainfall and may also limit the exposure of the sensitive phases of rice to the low temperatures of the end of the season, which affect spikelet fertility and yield; in the Senegal River valley, an appropriate choice of sowing date has been shown to avoid severe cold sterility [45].

For onion and tomato grown during the cool dry season, requirements increase almost linearly with delayed sowing. Sowings carried out early in the season benefit from a lower evaporative demand, whereas late sowings shift an increasing part of the cycle towards the hot dry season. The estimates exceed both reference values available for the valley: the planning dose used by SAED is 9000 m³·ha⁻¹, whereas the theoretical requirements retained by OMVS reach 7030 m³·ha⁻¹ for onion and 7200 m³·ha⁻¹ for tomato [47]. This comparison must nevertheless remain cautious, as these values may rest on different definitions of the requirement and of irrigation efficiency. The project efficiencies measured in thirteen schemes of the delta and the valley, ranging from 48 to 61% [30], are all below the value of 0.70 adopted here; this assumption therefore tends to lower rather than raise the computed gross requirements. The observed differences are thus not sufficient to demonstrate that the applied doses are insufficient. Any revision would require a comparison with the volumes actually delivered, the irrigation practices and the yields obtained.

The relative spatial variability of requirements is more marked during the rainy season than during the dry seasons. This contrast is consistent with the Sahelian rainfall regime, characterised by a strong spatial heterogeneity of rainfall totals [48]. This heterogeneity, which propagates directly to effective precipitation, also tends to increase: the share of annual totals contributed by extreme events rose from 17% over 1970-1990 to 21% over 2001-2010, while the number of rainy days declined [49]. During the dry seasons, the contribution of rainfall becomes marginal and requirements depend mainly on evaporative demand, whose variations between sub-UGP are less pronounced. In the rainy season, the spatial range represents about 48% of the mean requirement, against a mean absolute difference of about 7% between AgERA5 and NASA POWER. The spatial signal then dominates the inter-product disagreement and both products give the same ranking, with higher requirements in the Middle Senegal Valley than in Upper Ferlo. In the hot dry and cool dry seasons, the spatial range falls to 17% and 10% respectively, and the differences between products become large enough to change the location of the maximum requirements.

Spatial resolution also contributes to the representation of these contrasts. AgERA5 resolves variations within the sub-UGP, whereas NASA POWER provides a more aggregated representation. The aggregation of climatic variables can influence the outputs of agronomic models, particularly where precipitation is spatially heterogeneous [32][34][35]. These results show that irrigation planning must take season, crop, cycle length, sowing date and sub-UGP jointly into account. The gradient common to both products during the rainy season can support the distribution of volumes between management units. During the dry seasons, the spatial ranking depends more on the climate product and should not be used on its own to prioritise allocations; a single dose per season and per crop, with a margin covering the difference between products, is then a more defensible basis than a spatial differentiation that the data do not support.

The Sobol analysis identifies two distinct seasonal regimes. During the dry seasons, wind speed is the main factor influencing ET₀ variability, followed by relative humidity and mean air temperature. This hierarchy is consistent with the important role of the aerodynamic term of the Penman-Monteith equation in dry and windy environments [10][18][50]-[52]. It also agrees with the global synthesis of McVicar et al. [53], which shows that near-surface wind speed largely governs the aerodynamic component of evaporative demand and thereby contributes to variations in reference evapotranspiration.

The hierarchy changes during the rainy season. Relative humidity becomes the main factor, while the contribution of solar radiation increases. With AgERA5, these two variables have comparable importance. This change may be associated with the more humid conditions of the monsoon. Variations in relative humidity directly affect the vapour pressure deficit and may therefore exert a greater influence on ET₀. At the same time, variability in cloud cover increases the contribution of incoming solar radiation. This result agrees with sensitivity analyses that identify relative humidity as one of the main variables influencing ET₀ [54]. At the annual scale, ET₀ is most sensitive to relative humidity, followed by wind speed, mean air temperature and solar radiation. The two products therefore show a comparable hierarchy during the dry seasons and at the annual scale. Their main difference occurs during the rainy season. NASA POWER assigns a greater contribution to relative humidity, whereas AgERA5 assigns comparable importance to relative humidity and solar radiation.

The similarity between the first-order and total-order indices indicates that interactions among the variables contribute little to the calculated variability of ET₀. This interpretation should nevertheless remain cautious because the classical Sobol analysis assumes that the input variables are independent [55], whereas climatic variables are naturally correlated. The results therefore identify the variables to which ET₀ is most sensitive but do not account for the correlations among these variables under actual climatic conditions.

This study provides an integrated assessment of ET₀ and irrigation water requirements across the Senegalese part of the Senegal River Basin. Three findings are particularly relevant to regional planning. Irrigation requirements depend strongly on season and crop cycle length, their variation within a given season is sensitive to sowing date, and their spatial distribution depends on the climate product when contrasts among the sub-UGP are small. The agreement between AgERA5 and NASA POWER on the seasonal hierarchy and the main spatial gradient observed during the rainy season supports the robustness of these findings. Their divergence in some sub-UGP during the dry seasons provides an indication of the sensitivity of the estimates to the choice of climate product.

These results can support cropping-season planning and the allocation of water volumes at the sub-UGP scale. They show that irrigation doses should be adjusted according to season, crop and cycle length. They also highlight the potential benefits of early-season sowing during the rainy and cool dry seasons, provided that agronomic constraints are respected. AgERA5 provides a detailed representation of spatial variations, whereas NASA POWER provides estimates that can be interpreted directly at the management-unit scale. The joint use of both products therefore helps identify robust findings and the situations in which the choice of climate data has a greater influence on the estimated water volumes.

The operational scope of these estimates remains linked to their regional scale. Irrigation efficiency and crop coefficients were applied consistently throughout the study area to ensure comparability across crops, seasons and sub-UGP. Applying the results to daily irrigation scheduling at the field scale would require local data on delivered water volumes, irrigation-system performance, cropping practices and cultivated varieties. Strengthening the observation network would also allow the products to be evaluated in sub-UGP that are currently sparsely monitored. Such data would extend the proposed regional framework towards operational recommendations adapted to local conditions.