The Al-Madinah region is located in the western part of Saudi Arabia between latitude 24˚28"N and longitude 39˚36"E (Figure 1). Geologically, the area lies within the Arabian Shield, which dates back to the Precambrian era ([42]). Al-Madinah’s geology has three main components: basaltic lava plateaus called Harrats; alluvial deposits of poorly sorted sand and gravel; and rocky outcrops in the west and north, seen as large massifs and small isolated outcrops ([21]; [6]). The study area is generally characterised by low rainfall, where the annual average does not exceed 122 mm/y and is primarily concentrated in the highlands, and the annual mean temperature in Al-Madinah was 27.73˚C with a maximum of 35.60˚C and a minimum of 19.89˚C ([3]). Al-Madinah ranks as the fourth largest region in Saudi Arabia in terms of population and urban development. According to data from the Saudi Central Department of Statistics and Information ([19]), the city’s population was approximately 2,137 million in 2022.

Figure 1. Study location and topographic characteristics of the study area.

The present study employed climate projections obtained from global climate models participating in the Coupled Model Intercomparison Project Phase 6 (CMIP6) through the WorldClim 2.1 dataset (Figure 2). CMIP6 projections are commonly used in Intergovernmental Panel on Climate Change (IPCC) assessments and are driven by Shared Socioeconomic Pathway (SSP) scenarios, which combine assumptions about socioeconomic development with specified levels of radiative forcing. The WorldClim CMIP6 dataset provides projections under four SSP scenarios: SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5, representing sustainability-oriented low-emissions, middle-of-the-road, regional-rivalry, and fossil-fuel-intensive high-emissions pathways, respectively ([43]; [47]). The data of CMIP6 have been downscaled and calibrated from CRU-TS-4.03 by the Climatic Research Unit, University of East Anglia, using WorldClim 2.1 for bias correction ([46]). The predictions of precipitation and temperature in two scenarios, high emissions (SSP5-8.5) and medium stabilisation (SSP2-4.5), were used. These data have been designed in monthly means for 20-year periods that could be divided into near future (2021-2040), far future that includes mid-century (2041-2060 and 2061-2080), and end-century (2081-2100). They have a spatial resolution of 2.5 minutes or about ~21 km² at the equator ([22]; [17]). The baseline period data were selected from 1970 to 2018 from historical climate data in WorldClim. The monthly values of future projections of total precipitation (mm), maximum temperature (˚C), and minimum temperature (˚C) have been derived from four global climate models (GCMs), namely BCC-CSM2-MR, CNRM-CM6-1, CanESM5, and IPSL-CM6A-LR (Table 1). These GCM models were selected because they are commonly used and the models showed acceptable performances ([30]), including studies of future climate change over the Arabian Peninsula and historical precipitation simulations over Africa and the Arabian Peninsula ([5]; [35]). Raster resolution, spatial extent, coordinate reference system, and cell alignment were verified before conducting pixel-based analysis.

Table 1. Basic information about the selected CMIP6-GCMs models.

Figure 2. Flowchart of study methodology.

The performance of the WorldClim historical climatology (1970-2018) was evaluated against the observed monthly climatology (1973-2018) using four statistical metrics: the mean absolute error (MAE), root mean square error (RMSE), Pearson correlation coefficient (r), and coefficient of determination. These metrics were selected to capture average error magnitude, sensitivity to large deviations, and consistency in seasonal patterns. The mean absolute error was calculated as:

where S_i is the WorldClim value for month i, O_i is the corresponding observed value, and n is the number of monthly observations. MAE represents the average absolute difference between the two datasets and is expressed in the original unit of the evaluated variable. It therefore provides a direct and readily interpretable measure of the typical monthly error.

RMSE also measures the magnitude of the differences between WorldClim and observations, but it assigns greater weight to relatively large errors because the deviations are squared before averaging. This metric was therefore employed to determine if months with particularly large discrepancies adversely affected the agreement between the two datasets. This analysis aimed to identify whether months with particularly large discrepancies adversely affected the agreement between the two datasets. Both MAE and RMSE approach zero as agreement improves ([24]). The root mean square error was computed as:

The Pearson correlation coefficient was used to assess the strength of the linear relationship between the observed and WorldClim monthly climatology:

where O ¯ and S ¯ are the mean observed and WorldClim values, respectively. The coefficient ranges from −1 to +1 ([37]), with values approaching +1 indicating that WorldClim closely reproduces the seasonal variation of the observed climatology. Because correlation evaluates correspondence in pattern rather than absolute agreement, it was interpreted together with the error-based metrics. The coefficient of determination was calculated as:

and was used to express the proportion of the observed monthly variability associated with the linear relationship between the WorldClim and observed datasets. Values approaching 1 indicate strong correspondence in the seasonal climatic cycle. Monthly bias was calculated as the difference between the WorldClim historical value and the corresponding observed value, following the conventional model-minus-observation approach used in climate-data evaluation.

The multi-model ensemble (MME) approach was calculated as the arithmetic mean of the selected GCM outputs for each grid cell, scenario, and future period. This approach helps reduce the influence of individual model uncertainty and provides a more robust representation of projected climate conditions than reliance on a single GCM ([25]; [2]).

where X m , i is the value from model m at grid cell i, and M = 4 is the number of climate models.

The Hargreaves approach is chosen in the present study due to its requirement of fewer variables, the minimum and maximum temperature, and extraterrestrial radiation (Ra) ([9]). All these variables are available in the WorldClim dataset. The WorldClim dataset (1970-2018) has been processed by the raster package as well as the libraries’ rgdal and geosphere to extract the Eto-related maps. The code used is adapted from codes created by [41]. Hargreaves’ equation is applied as:

where ET₀ is Evapotranspiration Potential (mm/month); 0.0023 is an empirical coefficient used for unit conversion that also includes the kRs as an adjustment factor for solar radiation, and 0.408 is a factor to convert MJ m⁻² to mm. Tmax = maximum air temperature (˚C), Tmin = minimum air temperature (˚C), Ra = extraterrestrial radiation in mm/day ([18]; [9]). Reference evapotranspiration was estimated using the Hargreaves method based on monthly maximum, minimum, and mean temperature data. The calculation was applied on a pixel-by-pixel basis in the R environment to produce monthly ET₀ raster layers. The monthly ET₀ layers were then aggregated to derive the annual baseline ET₀.

The validation of the reliability of the baseline WorldClim-derived reference evapotranspiration (ET₀) was conducted using two approaches. The first approach included a comparison between the Hargreaves ET₀ derived by WorldClim and the station-based ET₀ calculated using the FAO-56 Penman-Monteith method. The standard reference method was selected as the FAO-56 Penman-Monteith method due to its comprehensive examination of the primary meteorological factors that influence evapotranspiration, including air temperature, solar radiation, wind speed, vapour pressure, and atmospheric pressure. Additionally, this method proves to be beneficial in arid, temperate, and tropical regions ([10]). The objective of this comparison was to directly assess the reliability of the WorldClim-derived ET₀ in comparison to a physically grounded station reference. The FAO-56 Penman-Monteith equation for reference evapotranspiration (ET₀) is:

where ET₀ is reference evapotranspiration (mm day⁻¹), R n is net radiation at the crop surface (MJ m⁻² day⁻¹), G is soil heat flux density (MJ m⁻² day⁻¹), T is mean air temperature at 2 m height (C), u2 is wind speed at 2 m height (m s⁻¹), es is saturation vapor pressure (kPa), ea is actual vapor pressure (kPa), e s − e a is vapor pressure deficit (kPa), Delta is the slope of the saturation vapor pressure curve (kPa C⁻¹), and gamma is the psychrometric constant (kPa C⁻¹) ([33]). For monthly calculations, daily ET₀ was converted to monthly total ET₀ as:

where N is the number of days in the month.

The second approach compared the Hargreaves ET₀ from WorldClim with the station-based Hargreaves ET₀. This comparison aimed to provide methodologically consistent validation, as the spatial ET₀ maps generated in this study also used the Hargreaves method. Using the same ET₀ equation for both WorldClim and station data enables a more precise distinction between uncertainties from the input climate data and those arising from differences among ET₀ methodologies.

In addition, inter-model consistency of projected reference evapotranspiration (ET₀) was evaluated for the four selected CMIP6 GCMs—BCC-CSM2-MR, CNRM-CM6-1, CanESM5, and IPSL-CM6A-LR—prior to calculating the multi-model ensemble mean. Pairwise Pearson correlation coefficients of monthly ET₀ projections were calculated separately for each scenario and 20-year future period to assess agreement in the monthly ET₀ pattern among the models. Inter-model spread in projected ET₀ magnitude was quantified using the coefficient of variation (CV), calculated as the standard deviation of model projections divided by their ensemble mean. Lower CV values indicate stronger inter-model agreement, whereas higher values indicate greater projection uncertainty. This assessment provides an indication of the consistency and uncertainty of individual-model projections before their use in the multi-model ensemble mean.

Calculating ET₀ rates enables better monitoring of drought conditions and water resources in Al-Madinah by integrating evapotranspiration estimates with precipitation data. Consequently, the standardized precipitation evapotranspiration index (SPEI) is considered a beneficial index to show the impacts on the hydrological system and water resources by linking these ET₀-based estimates with the temporal pattern of drought events and by providing a good understanding of the water surplus or deficit cases during the period investigated ([15]; [38]). The SPEI is formulated based on the monthly climatic water balance and is usually represented by calculating the difference between precipitation and evapotranspiration ([1]). The monthly climatic water balance over a specified time period (D_i) is calculated:

where P_i stands for the amount of precipitation, and PET_i represents Potential Evapotranspiration for the same month “I”. The negative value between quantities of rain and ET₀ often indicates an existing rainfall deficit (RD), whereas surplus (RS) occurs when rainfall is more than evapotranspiration ([16]). The SPEI index included the aggregation of D_i values in different time scales, such as 1-month or 6-month periods. This aggregation is then converted into standardizing values using the log-logistic distribution that represents how many standard deviations a D_i value is from the mean of the log-logistic distribution. This provides a consistent method for comparing and interpreting drought or wet conditions across different periods and regions. The mathematical representation of the probability density function for the log-logistic distributed variable is as follows:

where α, β, and γ are scale, shape, and origin parameters respectively ([45]; [38]). The Log-logistic distribution adopted for standardizing the D series for all time scales is given by:

f(x) value is then transformed to a normal variable by means of the following approximation:

where P= 1 − f(x), and W is given by

where C₀, C₁, C₂, d₁, d₂, and d₃ are constants equal to 2.515517, 0.802853, 0.010328, 1.432788, 0.189269, and 0.001308, respectively ([45]). Moreover, the drought and moisture conditions would be identified via the SPEI values as displayed in Table 2. These values range from positive, meaning wet conditions, to negative values that exhibit drought events, while the SPEI value indicates the event’s intensity ([32]). Therefore, drought years were identified with the threshold values of −1 ([11]; [29]).

Table 2. The degree of drought and moisture in SPEI.

The SPEI includes multiple timescales that exemplify different types of droughts. Short timescales are more suitable for detecting meteorological drought and agricultural drought, which are (1 - 3 months averages) and (3 - 6 months averages). The long-time scales often used for hydrological drought and water resources average (12 to 24 months averages) ([36]; [1]; [15]). In order to produce the SPEI index for the study area, the values of historical ET₀ (1970-2018) and future (2021-2100) have been extracted from the maps generated based on the Worldclim database, utilising the climate station’s locations in Al-Madinah, which are 40430, 40439, M001, M002, and M004.

The accuracy assessment indicates that WorldClim temperature data showed strong agreement with the observed station records for all temperature variables (Table 3 and Figure 3). The correlation coefficients were notably high, ranging from 0.998 to 0.999, indicating that WorldClim effectively captured the seasonal variation of maximum, minimum, and mean temperature. However, the error statistics reveal some differences in accuracy among the variables. The minimum temperature had the best performance, with the lowest RMSE and MAE values, with RMSE = 0.955˚C and MAE = 0.733˚C. In contrast, maximum temperature showed the largest error, with RMSE = 2.010˚C and MAE = 1.939˚C. The mean temperature showed intermediate error values, reflecting the combined influence of both minimum and maximum temperature estimates. The bias results showed that WorldClim generally underestimated observed temperatures, as indicated by negative bias across all variables. This underestimation was more pronounced for maximum temperature, while minimum temperature showed the smallest bias. The monthly bias pattern further confirms that the magnitude of underestimation varied throughout the year, with larger differences occurring during some cooler months (Figure 3). The precipitation dataset showed strong agreement between WorldClim and observed records, with R = 0.937 and R² = 0.878 (Table 3 and Figure 3), indicating that WorldClim captured most monthly precipitation variability. The bias value was slightly negative (−0.332 mm), indicating a small overall underestimation by WorldClim. Monthly bias patterns showed that WorldClim overestimated precipitation mainly in March and April, while it underestimated precipitation in most other months, as shown in Figure 3. Overall, the precipitation data showed acceptable performance, but with lower reliability than the temperature data.

Table 3. Accuracy assessment of WorldClim temperature and precipitation data against observed station records.

The study area is characterised as a dry region, experiencing intense summer heating and limited rainfall that is concentrated in certain seasons. Temperatures, as shown in Figure 4, rise gradually from winter to summer, with maximum (Tmax), minimum (Tmin), and mean (Tmean) temperatures peaking during July and August before declining towards the winter months. This pattern illustrates the presence of hot summers and relatively cooler winters. Conversely, precipitation exhibits a highly seasonal and irregular distribution, with the majority of rainfall occurring in March to April and November to December, while there is very low precipitation during the summer months, particularly from June to September (Figure 4). These climatic characteristics provide an important basis for interpreting the temporal and spatial distribution of ET₀. Therefore, WorldClim-based ET₀ estimation allows the baseline spatial distribution and future changes of ET₀ to be assessed consistently, which also supports the later interpretation of SPEI-based drought conditions.

Figure 3. Taylor diagram evaluating the agreement between the WorldClim historical climatology and observed monthly climate data. The diagram summarizes Pearson’s correlation coefficient, normalized standard deviation, and normalized centered root mean square (RMSD) difference. Variables positioned closer to the observational reference point exhibit stronger agreement with the observed climatology.

Figure 4. Monthly comparison between observed and WorldClim climate variables in Al-Madinah.

The validation statistics are summarised in Table 4and Figure 5. WorldClim-derived ET₀ showed a strong correspondence with station-based FAO-56 Penman-Monteith ET₀, with R = 0.960 and R² = 0.921, indicating that the seasonal pattern of monthly evapotranspiration was well captured. Nevertheless, the relatively high RMSE (62.271 mm month⁻¹), MAE (59.124 mm month⁻¹), and negative bias (−59.124 mm month⁻¹) indicate underestimation of ET₀ magnitude relative to the FAO-56 reference. In contrast, a stronger and more consistent performance was obtained when WorldClim-derived ET₀ was evaluated against station-based Hargreaves estimates, for which the correlation increased to R = 0.975 (R² = 0.951), while RMSE and MAE decreased substantially to 11.911 and 10.089 mm month⁻¹, respectively. The small positive bias of 3.208 mm month⁻¹ indicated that the WorldClim-derived estimates were closely aligned with the station-based Hargreaves values, with only slight overestimation. Overall, these findings indicated that the WorldClim-derived product can adequately represent local evapotranspiration dynamics. The particularly close correspondence with station-based Hargreaves-ET₀ further supports the reliability of WorldClim climatic data for estimating ET₀ at the local scale, while the larger differences relative to FAO-56 Penman-Monteith are likely associated primarily with methodological differences rather than with an inability of WorldClim data to reproduce local ET₀ variability.

The correlation analysis demonstrated strong agreement among the four selected CMIP6 GCMs in reproducing the seasonal pattern of projected Hargreaves ET₀. Under SSP2-4.5, the pairwise correlations remained consistently high across all future periods, indicating that the models produced highly similar monthly ET₀ patterns under the intermediate-forcing scenario. Under SSP5-8.5, the correlations were also generally high, but a wider range was observed during the late-century

Table 4. Statistical validation of WorldClim-derived Hargreaves ET₀ against station-based FAO-56 Penman-Monteith and station-based Hargreaves ET₀.

Figure 5. Linear relationships between WorldClim-derived Hargreaves ET₀ and station-based ET₀ references: (a) station-based FAO-56 Penman-Monteith ET₀ and (b) station-based Hargreaves ET₀. The dashed line represents the fitted regression line, while the solid line represents the 1:1 agreement line.

periods. Specifically, pairwise correlations ranged from 0.986 to 0.999 in 2021-2040, 0.963 to 0.998 in 2041-2060, 0.935 to 0.998 in 2061-2080, and 0.894 to 0.997 in 2081-2100. The lowest correlation values occurred mainly between CanESM5 and IPSL-CM6A-LR, particularly during 2081-2100, suggesting increased inter-model divergence under SSP5-8.5 toward the end of the century (Figure A1 and Figure A2). The coefficient of variation showed that the inter-model spread in projected ET₀ magnitude was generally low. Under SSP2-4.5, annual CV values ranged from 1.976% to 2.378%, reflecting close agreement among the four GCMs throughout the future periods. Under SSP5-8.5, CV values were higher, rising from 2.115% in 2021-2040 to 4.411% in 2081-2100, indicating greater projection spread under stronger climate forcing (Table A1). Nevertheless, the overall CV values remained relatively low, confirming that the multi-model ensemble mean can provide a good representative estimate of future ET₀ projections.

The baseline annual ET₀ map (Figure 6) showed clear spatial variability across the study area. Higher ET₀ values were mainly concentrated in warmer and lower-elevation areas, such as Wadi Al-Hamd and Al-Harrat in the middle, eastern, and southeast parts, while lower ET₀ values were observed in cooler or higher-elevation zones in the north and northwestern areas. This spatial pattern indicates that baseline ET₀ is strongly influenced by temperature gradients and topographic variation. The peak values were recorded during the summer months (June, July, and August), whereas the lowest rates occurred in winter, specifically in December and January, with November also noted, as illustrated in Figure 7. Moreover, the annual evapotranspiration rates of Al-Madinah show substantial amounts of increase across the years, ranging from 1493 to 2151 mm/year. There are similarities between the historical trend results of ET₀ in Al-Madinah in this study compared to those concluded by [12], who investigated the ET₀ rate by applying the Food and Agriculture Organization (FAO) Penman-Monteith (PM) method in 29 meteorological stations distributed all over Saudi Arabia for the period 1980 to 2008. The findings revealed the highest amount of ET₀ in the summer and the lowest in the winter in the country’s entirety. The annual evapotranspiration rate of Al-Madinah was high (2151 mm/year), a value also reported

Figure 6. Spatial distribution of baseline annual reference evapotranspiration (1970-2018) in Al-Madinah.

Figure 7. The spatial distribution of monthly averaged ET₀ in Al-Madinah (1970-2018).

by [27], and was slightly higher (ranging from 2059 to 2405 mm/year during 1992-2014). These values are often expected in arid and semi-arid regions such as Saudi Arabia. For example, the ET₀ amount of the central regions (Riyadh, Al-Qassim, Hail provinces) was estimated to be above 2100 mm/year from 1950 to 2013 ([28]). These regions mostly experience marginally hotter temperatures than Al-Madinah.

In addition, the temperature will potentially rise steadily under both SSPs, with an average of about one degree Celsius per 20-year period in the near future (2021-2040) and the mid-century (2041-2060 and 2061-2080), with a more noticeable increase of around two degrees Celsius by the end of the century (2081-2100), in most months. These results agree with those obtained by [11], who concluded that the warming trend on the Arabian Peninsula is predicted to be larger under SSP5-8.5 (4˚C increases) than the SSP2-4.5 scenario (2˚C increases), an almost similar finding was also reported by [43], who estimated that the temperature is likely to rise gradually under SSP5-8.5 to around 1.3 - 1.6 C and 2.1 - 3.7 C by 2025-2044 and 2045-2064 periods, respectively, to reach a high point in 2065-2084 by increasing by 3.2 - 3.8 C.

The evapotranspiration rate increases in response to rising temperatures. This study’s projections of future evapotranspiration closely align with temperature trends. Evapotranspiration is projected to increase steadily across all periods under both SSP2-4.5 and SSP5-8.5 scenarios, relative to the baseline period of 1970-2018. These increases will likely intensify later in the century, particularly from 2081 to 2100 and under the more extreme scenarios. For example, under SSP5-8.5, the increase is stronger. Mean monthly ET₀ rises from 166.07 mm/month during 2021-2040 to 182.47 mm/month during 2081-2100. The corresponding annual ET₀ increases from 1992.83 mm/year to 2189.68 mm/year, representing an increase of 8.90% to 19.66% relative to the historical baseline. This indicates a substantial intensification of atmospheric evaporative demand, especially under the late-century high-forcing scenario.

The most substantial increases are expected in summer months, reaching 18.93% by 2081-2100, while the winter months show smaller gains (Figure 8). This condition would also be similar to the winter across the Kingdom ([31]). A modest increase of up to 5% (9.7 mm) is anticipated in spring (April and May) under SSP5-8.5 during 2081-2100. This result is consistent with [31] estimations, which reported that the ET₀ rate of the Al-Madinah region is projected to increase slightly in the springtime, around 2% by 2080, more than other areas located in the southeast and southwest parts of Saudi Arabia. This study’s projected trend in evapotranspiration agrees with the estimates of [31], which concluded that evapotranspiration would be higher in Saudi Arabia by the end of the 2080 period.

The spatial distribution of projected ET₀ shows a consistent pattern across all periods and scenarios (Figure 9). Under SSP2-4.5, the increase appears gradual from 2021-2040 to 2081-2100, with moderate expansion of higher ET₀ classes. Under SSP5-8.5, the increase is more pronounced, especially during 2061-2080

Figure 8. Projections of average evapotranspiration (ET₀)/mm, in Al-Madinah under SSP2-4.5 and SSP5-8.5 scenarios from 2021 to 2100.

Figure 9. Spatial distribution of projected annual reference evapotranspiration (ET₀; mm/year) in Al-Madinah under SSP2-4.5 and SSP5-8.5 scenarios from 2021 to 2100.

and 2081-2100, when high ET₀ values become more widespread across the study area. This suggests that atmospheric water demand is expected to intensify more strongly under the high-emission scenario.

Projected precipitation, Hargreaves ET₀, and climatic water balance were analysed under SSP2-4.5 and SSP5-8.5 for four future periods: 2021-2040, 2041-2060, 2061-2080, and 2081-2100. The results showed that annual precipitation is projected to increase moderately relative to the historical WorldClim baseline of 75.665 mm/year, with increases ranging from 10.97% to 12.80% (83.97 to 85.36 mm) under SSP2-4.5 and from 12.27% to 15.45% (84.95 to 87.36 m) under SSP5-8.5 (Figure 10). However, the projected increase in ET₀ is substantially larger than the increase in precipitation. Under SSP2-4.5, ET₀ increases from 1991 mm/year during 2021-2040 to 2072 mm/year during 2081-2100. Under SSP5-8.5, ET₀ increases more strongly, from 1992.8 mm/year to 2189 mm/year over the same period. As a result, the climatic water balance, calculated as precipitation minus ET₀, becomes increasingly negative under both scenarios, as shown in Figure 11. This indicates that projected precipitation gains are not sufficient to compensate for the increase in evaporative demand.

Al-Madinah could face drought events and water deficiency challenges caused by the increased ET₀ rate with reduced rainfall, negatively affecting the hydrological and water resources. The analysis of historical data for meteorological droughts by the SPEI index showed Al-Madinah was characterised by wetter conditions in the first half of the 20th century (1970-2000) than in the last half, which experienced more drought cases (Figure 12). These results are in agreement with those obtained by The National Centre of Meteorology in Saudi Arabia [34], which applied the Precipitation Index (SPI) to monitor drought situations (1985 to 2019 in 12-month); the findings have confirmed that there were moderate

Figure 10. Projected annual precipitation under SSP2-4.5 and SSP5-8.5 for the periods 2021-2040, 2041-2060, 2061-2080, and 2081-2100. The figure shows moderate temporal and scenario-based variability in precipitation.

Figure 11. Projected climatic water balance, calculated as precipitation minus ET₀, under SSP2-4.5 and SSP5-8.5. Negative values indicate increasing atmospheric moisture deficit.

Figure 12. The SPEI in the Al-Madinah region from 1970 to 2018 at 12-month (above) and 24-month (bottom). *Red color represented drought situations with negative values, and blue color symbolized wet conditions with positive values.

and severe wet conditions covering most parts of Saudi Arabia in the period 1985 to 1997, with fluctuations and frequent dry conditions from 1999 until 2018. [7] also confirmed that the KSA experienced extreme drought events from (January) 2000 to (July) 2013 using the SPI index, whereas [4] concluded that applying the SPI index from 1978 to 2017 indicated that the drought occurred typically from June to September in Saudi Arabia, as well as October despite it being a transition month, October experienced drought conditions for around 29 years out of 40 years used in the study.

The SPEI projections (Figure 13) showed a clear temporal shift in hydroclimatic conditions under both emission scenarios. Conditions shifted from comparatively moist in the early to mid-21st century to progressively arid after mid-century. Drought intensification became most pronounced in the late-century period, especially after 2080. Compared with SSP2-4.5, SSP5-8.5 showed a stronger and more persistent drought signal, consistent with the anticipated impact of elevated greenhouse gas forcing on evapotranspiration demand and water deficits. Overall, the findings indicate that drought risk is expected to rise significantly by the end of the century, particularly under the high-emission SSP5-8.5 scenario. These findings agree with the [44] report about extreme climate indices for the Arab region, pointing to the dry periods increasing by the end of the century, particularly in the western and northern parts of the Arabian Peninsula. Moreover, according to [31], the more vulnerable areas to desertification by 2080 are expected to be in the regions in the northwest of Saudi Arabia, where Al-Madinah is located.

The use of WorldClim data provided a consistent and accessible basis for estimating reference evapotranspiration (ET₀) across the study area, particularly

Figure 13. The SPEI in the Al-Madinah region under SSP2-4.5 and SSP5-8.5 scenarios.

where long-term and spatially distributed station observations are limited. However, because WorldClim is a gridded climate dataset, the derived ET₀ values should be interpreted as spatial estimates rather than direct field measurements. Although this study validated WorldClim-derived Hargreaves ET₀ against station-based FAO-56 Penman-Monteith and station-based Hargreaves ET₀, uncertainty may remain due to spatial interpolation, local topographic effects, and differences between gridded and point-based observations. Future work could further improve the assessment by incorporating additional meteorological stations, observed evapotranspiration records where available, and other gridded or remotely sensed climate datasets. This would help refine the spatial patterns of ET₀ and increase confidence in future drought and evapotranspiration projections.