3.1. Reanalyzed versus observed precipitation and evaporation
First, we address how comparable precipitation distribution is over the AP between the reanalyses and the observations (Fig. 2). At the annual timescale, relatively high precipitation magnitudes exist along the Zagros Mountains in the north and over Asir and Sarat mountains in the southwest. These precipitation hot spots are present in all datasets. However, their strength over the southwestern AP is relatively weak in CHIRPS. All precipitation maxima are relatively stronger in ERA5, while CRU and PERSIANN tend to be wetter over Yemen and Oman. The precipitation distribution over central Saudi Arabia is comparable across all datasets, with CRU the driest among all. Overall, there are no distinct or systematic differences between reanalyzed and observed precipitation datasets. We further compare the percentage of annual precipitation that falls during the wet season across datasets (Fig. S1). The purpose of this comparison is twofold. First, we would like to ensure that there are no noticeable systematic differences in the intra-annual distribution of precipitation between the reanalyzed and observed datasets. Second, we want to highlight that November to April season, used in this study, covers most of the annual precipitation that falls over the region, consistent with earlier findings (Abdullah and Almazroui, 1998; Almazroui, 2012; Almazroui et al., 2012). Both observational and reanalyzed datasets show that over 75%, and in most cases, over 90% of the precipitation is received during the wet season used in this study. The Red Sea Coast and Asir mountains receive considerable rainfall through warm-season convection (Almazroui, 2011), reflected in relatively lower (30–60%) fractions during the rainy season. There is a discrepancy among the datasets about the wet season contribution over parts of Yemen and Oman. However, this location is in the Empty Quarter, where seasonal precipitation is very low, and minor differences can lead to more considerable fractional changes. These comparisons suggest no distinguishable systematic differences in reanalyzed and observed precipitation distributions. Therefore, this study's determined synoptic-scale moisture sources based on reanalyzed precipitation can be considered reliable.
We also analyze evaporation data over the source domain during the wet season. No gridded observed evaporation data exists; therefore, analysis is limited to the evaporation from the two reanalyses (Fig. 3). The monthly evolution of evaporation (Fig S2) shows that the mean spatial differences in evaporation throughout the wet season are much more notable than the mean temporal differences from month to month. Overall, more evaporation occurs over water bodies than inland regions, with the most vigorous land evaporation occurring near the equator.
3.2. Characteristics of primary moisture sources
The fraction of moisture falling as precipitation in a grid cell that originated as evapotranspiration from the entire evaporative source region (Fig. 1), also known as the recycling ratio, is shown in Fig. 4a as an average of two datasets (ERA5 and MERRA2). The highest recycling ratio is obtained when all the precipitation is sourced back to its origin at a grid point. In this study, the recycling ratio varies between 90 to 96% (expressed in % of the total) over most of the AP, reflecting that the evaporative source domain used in this study is sufficiently large to account for the origin of almost all precipitation falling over the AP during the wet season. The contribution from the most prominent (all) source regions is shown in Fig. 4 (Fig. S3) as an average recycling ratio based on two reanalysis datasets. Figures S4 and S5 show individual contributions from all regions separately in ERA5 and MERRA2, respectively. In the north, most of the moisture comes from the Mediterranean/Caspian Seas (up to 50%), North Atlantic, and Europe (up to 40%). The primary contributors to the precipitation over this region are extratropical weather systems originating over the Mediterranean Sea and adjacent areas (Almazroui et al. 2016). These weather systems move eastward and hence have less impact on the remaining AP region in the south. Therefore, the moisture within these systems shows more influence over (up to 40%) northern Syria and Iraq and their contribution progressively depletes southeastward. The lack of reach of extratropical weather systems over the rest of the AP diminishes contribution from the mid-latitude source regions towards the central to southern AP. In the central parts of AP, moisture advection from the African continent and South Atlantic has a sizeable contribution (up to 30% when aggregated). Potentially, a fraction of moist air near the equator over the South Atlantic and African land regions reaches the upper levels escapes towards Northern Africa and reaches the AP with the aid of the North African Jet (Figure S6).
In comparison, moisture transport from the Western Indian Ocean dominates the southern half of the AP (30 to 65%). It is important to note that the geographical footprint of moisture from the Western Indian Ocean, which also includes the Arabian and Red Seas in our study, is the largest in the southern AP, with maxima over the southwestern tip of Yemen (~ 65%), and the smallest in the northern AP, with minima over the Zagros Mountains along the border of Turkey in the north (~ 5%). A relative maximum in evaporation over the Red Sea (Fig. 3) combined with the North Africa Jet allows a Red Sea trough to enhance precipitation over northern and central Saudi Arabia. At the same time, southeasterly winds advect moisture from the Arabian Sea and Gulf of Aden over the southern Arabian Peninsula. Interestingly, local moisture recycling does not significantly affect the precipitation distribution over the AP, as its maximum contribution remains less than ~ 10–15%. Through the western two-thirds of the target region, local recycling contribution is under 5%, indicating its less importance than advection from outside sources. The AP is overall arid, and very little moisture is available for evaporation (Fig. 3). As the ground is primarily dry, rains penetrate the soil, leaving little contact between the water and the atmosphere for evaporation. Therefore, local recycling contributes very little to atmospheric moisture over the region. Overall, the relative contribution of individual evaporative sources is comparable between ERA5 and MERRA2 (Figs. S4, S5), reflecting minor sensitivity to the choice of reanalysis to determine moisture source regions. Due to this close correspondence between the two datasets, we present the rest of the analyzed results based on their average.
The total recycled precipitation is the multiplicative product of the recycling ratio and the total precipitation. When the total recycled precipitation is averaged over the study region, there is considerable sub-seasonal variability in moisture contribution from individual source regions (Fig. 5). Note that a sub-region can show high recycling ratio values over the parts of the target region (AP). However, its contribution to the area-averaged recycled precipitation over the target region may still be low if higher magnitudes of recycling ratio are over the areas that receive low seasonal precipitation magnitudes. For instance, the southern AP receives less overall precipitation; therefore, the contributions from source regions to the northern AP are more heavily weighted than the southern AP. The contribution from the Mediterranean/Caspian Seas is the highest until the beginning of March, when it gradually starts diminishing. Note that the northern part of the AP, where the Mediterranean/Caspian Seas have the most significant contribution, receives relatively less rainfall towards the end of the rainy season (Fig. S7). The sub-seasonal precipitation variability over the northern AP region coincides with the annual cycle of evaporation from the Mediterranean and adjacent seas that exhibits gradual decreases in the wet season with a yearly minimum in April–May (Romanou et al., 2010), explaining the decline in the moisture contribution from the Mediterranean/Caspian Sea. On the other hand, the moisture contribution from the Western Indian Ocean remains relatively steady until March before strengthening to reach its maxima towards the end of the rainy season, which is consistent with relatively large amounts of rainfall in March and April over the southern half of the AP (Fig. S7). The late-season strengthening of moisture advection from the Indian Ocean is supported by the gradual shift in the cross-equatorial flow over the ocean east of Africa (Fig. S6). The flow progressively shifts from southward in November to February to zonal in March and northwestward in April. The same dynamical changes support the sub-seasonal variability in the contribution from Eastern Africa, which is in sync with that from the Western Indian Ocean. Note that, over this period, the intertropical convergence zone gradually moves southward over southwest Africa (away from the AP) and back over Eastern Africa (Ashfaq et al., 2021). Some syncing is also noticed between moisture from the Mediterranean/Caspian Seas and the North Atlantic, which is understandable as similar dynamical characteristics influence the movements of storm tracks and moisture from the Mediterranean and beyond (Batibeniz et al., 2020). However, it is interesting to note that moisture contribution from Europe gradually increases over the wet season and reaches its maximum at a time when moisture supply from North Atlantic, Mediterranean/Caspian Seas depletes. Such an increase in moisture from a mid-latitude terrestrial source region is explicable given the annual cycle of surface temperatures that continue to rise during the AP wet season. Moreover, storm tracks continue to move north during the latter half of the wet season. The relative maximum in European contribution indicates that the storms that begin further north have more access to precipitation originating from evaporation over land in Europe.
3.3. Thermodynamic pathways of ENSO/NAO influence over the AP
ENSO exhibits a positive association with precipitation variability over the AP, meaning that more rains are expected during the wet season when the warm phase of ENSO is active and vice versa (Abid et al. 2016; Atif et al. 2020). While earlier studies suggest above normal moisture advection from the Indian Ocean during El Niño (Abid et al. 2020; Abid et al. 2018; De Vries et al. 2016), a more specific understanding of anomalies in moisture from oceanic and terrestrial moisture source regions is yet to be determined. Our analyses suggest that ENSO enhances the moisture influx over the AP from almost every source region during the positive phase of ENSO (Figs. 6, S8). In particular, a statistically significant increase in moisture advection from the Indian Ocean and the African continent occurs. Note that Walker circulation disruption during El Niño causes anomalous convection (subsidence) in the Western (Eastern) Indian Ocean (Abid et al. 2020). As a result, an anomalous westward flow emerges over the Indian Ocean in the lower atmosphere, reflected in the regressed 850 mb winds map (Fig. 6a). These conditions favor enhanced moisture advection from the Western and Eastern Indian Ocean and northern parts of Eastern Africa.
Moreover, wetter than normal conditions are well known over equatorial and Eastern Africa during the warm phase of ENSO (Mbigi and Xiao, 2021), meaning that more moisture is available to be transported to the AP from those areas. Additionally, with weakened easterlies and more divergence in equatorial Africa during ENSO (Spinks and Lin 2014), less moisture already available near the equator is advected towards the tropical Atlantic Oceans. Thus, a more significant percentage of precipitation advected out of equatorial Africa becomes available for transportation to other regions. Excessive moisture transport is also witnessed from Europe and the Mediterranean/Caspian Seas. It is well known that the warm phase of ENSO favors the southward shift in the extratropical cyclones, causing enhanced moisture advection from the mid-latitude source regions over the northern AP. However, most of these increases are not statistically significant. Local recycling also exhibits a noticeable rise in Iraq and Saudi Arabia along the Arabian Gulf (Fig. S8).
Substantial sub-seasonal variability in ENSO teleconnection and associated moisture anomalies from respective source regions exists over the AP (Figs. S10-S12). Overall, ENSO teleconnection becomes insignificant and reverses over Iraq and adjacent Saudi Arabia region in Jan–Feb creating a reversal in moisture anomalies from primary source regions (Europe, North Atlantic, and Mediterranean/Caspian Seas). The circulation anomalies in the lower atmosphere (850 mb) exhibit anomalous dry air entrainment from northern Africa during Jan–Feb in contrast to the rest of the season, where moisture advection from the abovementioned source regions is favored (Figs. S10-S12). At the individual level, the increase in contribution from Eastern Africa and the Western Indian Ocean is statistically more robust during Nov–Feb, while that from South Atlantic only becomes significant during Mar–Apr. Note that near the equatorial belt over east Africa, ENSO teleconnections lead to onshore wind anomalies from the Indian Ocean, and winds over land change direction in Mar–Apr, limiting the moisture advection from these regions to the AP (Figure S12).
As previously noted, NAO’s influence on the AP climate is not as robust as it is in the case of ENSO. A positive phase of NAO causes the westerlies over the Mediterranean and Northern Africa to weaken, decreasing the amount of moisture that advects from Europe and the North Atlantic into the target region. Spatially, the moisture change from the NAO teleconnection exhibits a dipolar pattern over the AP, with a negative association in the northern two-thirds and a positive association in the southern one-third. However, this negative influence over that region is mostly statistically insignificant (Fig. 7a) due to a lack of robustness in moisture anomalies from the primary source regions (Fig. 7b-7d). More specifically, the negative association in the northern two-thirds is caused by reduced moisture advection from Europe and North Atlantic and potentially a westward movement in Mediterranean storm tracks. While the moisture advection from the Mediterranean/Caspian Seas also decreases significantly over parts of central Saudi Arabia, its reduction is insignificant over Syria. In contrast, it shows a mixed response over Iraq. The positive association over the parts of southern AP is due to an enhanced moist advection from Western African, Indian Ocean, and Eastern Africa, and more local recycling (Fig. S9). The effects of NAO teleconnection over the AP also exhibits sub-seasonal variability (Figs. S13-S15). For instance, the statistically significant negative NAO association over the central AP only exists during Nov–Dec, when moisture supply from most prominent sources decreases. Similarly, an increase in moisture advection from the Indian Ocean and parts of Africa over the southern AP is only significant during Mar–Apr.
3.4. Wet season precipitation trends
We use the Mann-Kendall test to calculate trends in total precipitation and moisture contribution from respective evaporative sources and test their significance at a 95% confidence level. Generally, the wet season precipitation over the AP exhibits negative trends (Fig. S16), with evaporation trends in reanalysis showing similar trends (Fig. S20). Unfortunately, the spatial trends are inconsistent across observational datasets and between observations and reanalyses over much of the AP. The lack of correspondence among observations is because the density of ground observations is insufficient over the AP region. Some gridded observations rely on ground observations, while others blend remotely sensed data with ground observations. Nonetheless, more research is needed to understand the sources of uncertainty in gridded observations and their connection with the precipitation trends over the AP.
Despite the disagreements among datasets, the negative precipitation trends over the central and southwestern AP region are relatively robust across observations and reanalyses. Recent studies based on station observations have consistently shown the central AP as a region of decreasing precipitation leading to more severe drought (Alsaarann and Alghmadi 2021; Syed et al. 2022). Therefore, we limit our focus to these regions, where we assume the drying trend is robust, to understand the potential causes of these trends. Since moisture source analyses are based on the reanalyzed data, the trends in moisture advection from individual moisture should help explain the precipitation trends in the reanalyzed precipitation over the AP. The examination of trends in moisture contribution from source regions (Fig. 8) suggests that negative precipitation trends over the central and southwestern AP (Fig. S16) are driven mainly by reduced moisture advection from Western Africa, South Atlantic, Western Indian Ocean, and Eastern Africa (ordered from most to least significant). When trends are analyzed over the terrestrial parts of the evaporative source region, both ERA5 and MERRA2 display statistically significant strong negative trends over the Congo basin and parts of central Africa (Figs. S16, S20) – the area which is partitioned as Western Africa in our subdivision of moisture sources (Fig. 1). This region experiences a dry season between Nov to March (Ashfaq et al., 2021), which is why moisture contribution from Western Africa gradually increases during the wet season over the AP (Fig. 5; mustard line). Studies have shown that dry season length has recently increased over Equatorial Africa (Froidurot and Diedhiou, 2017), leading to less overall moisture available for evaporation. With less terrestrial evaporation, the northward moisture fluxes from equatorial Africa should reduce. Indeed, reanalyses-based evaporation trends (Fig. S20) show a decrease in these areas, supporting evidence that these may be related to reduced precipitation in equatorial Africa. However, it is essential to note that these patterns are not seen in all observational datasets, so there is still significant uncertainty in this finding (Fig. S16). As previously mentioned, the moisture contribution from the Eastern African region also exhibits a decline over the southwestern AP, which is present in both reanalyses (Figs. S18, S19). This reduction in moisture supply from Eastern Africa is most likely due to a drying trend over the Ethiopian highlands, which is present in both datasets, leading to an overall decrease in precipitation in the circulations over central Africa. Most gridded observations also show a negative trend over the Ethiopian highlands (Fig. S16). Despite the noticeable differences between the reanalysis and observations in the spatial distribution of precipitation trends over the evaporative source region, the dependence of AP precipitation on the terrestrial evaporative sources demonstrates how precipitation trends over one source region can impact precipitation distribution over a distant sink region and highlight the complexity and intricacy of thermodynamic drivers influencing the precipitation variability over an area. Although we are seeing an increase in evaporation over the ocean sources (Fig S20), this does not translate to advection into the AP, indicating further that much of the variability in the AP precipitation with a warming atmosphere is due to the changes in evaporation over terrestrial sources. Overall, these findings suggest that changes in evaporation over a warming ocean are less impactful in determining prevailing AP precipitation trends than the changes in evaporation over terrestrial source regions.