The Global Precipitation Measurement (GPM) is an international satellite mission to provide quasi-global precipitation estimates with a high temporal resolution (30 min, daily and monthly) and spatial resolution (0.1∘) through the Integrated Multi-satellitE Retrievals (IMERG) product. The GPM mission follows the Tropical Rainfall Measuring Mission (TRMM), aiming to improve satellite-based precipitation observation capability. GPM-IMERG provides different rainfall estimates that are combined from active and passive instruments in the GPM constellation (https://gpm.nasa.gov/, last access: 25 September 2020). Further details are given by Huffman et al. (2014). The GPM data have been employed in several studies over the Mediterranean region (e.g., Retalis et al., 2018; Petracca et al., 2018; Caracciolo et al., 2018; Cinzia Marra et al., 2019; Hourngir et al., 2021). In this study, we used the IMERG version 6 GPM-L3 final precipitation product (30 min daily) to estimate the extreme precipitation characteristics for 20 years (2001–2020) over the SiP.

To investigate the synoptic–dynamical climatology associated with SiP's rainfall events, the required variables were obtained from the National Centers for Environmental Prediction and National Center for Atmospheric Research (NCEP/NCAR) (https://psl.noaa.gov/data/gridded/data.ncep.reanalysis.html, last access: 10 February 2021; Kalnay, 1996) as well as the fifth generation of the European Centre for Medium-Range Weather Forecasts (ERA5) (Hersbach et el., 2020) https://www.ecmwf.int/en/forecasts/datasets/reanalysis-datasets/era5 (last access: 12 October 2021). NCEP/NCAR and ERA5 have provided reanalysis datasets with multiple time steps at the surface and pressure levels of the atmosphere since 1948 and 1979 with a global 2.5∘×2.5∘ and 0.25∘×0.25∘ horizontal grid, respectively. In the literature, these datasets have been used over the Mediterranean region in several studies, especially with regard to the synoptic analysis of precipitation, blocking systems, and storm and cyclone tracking (e.g., Krichak et al., 2002; Trigo et al., 2004; Tolika et al., 2006; Trigo, 2006; Lois, 2009; Barkhordarian et al., 2013; Almazroui and Awad, 2016; Almazroui et al., 2014, 2017; Varlas et al., 2018; Kotsias et al., 2020). First, in this research, NCEP/NCAR data were used to study the pressure fields due to their coarser resolution, as it is believed that large-scale pressure systems such as cyclonic and anticyclonic patterns could be better represented at coarse resolution, especially at lower atmospheric levels over complex environments. Second, ERA5 data were used to quantify the moisture condition as well as the wind stream structure and profile related to the wet and dry periods at a finer resolution. The following reanalysis meteorological datasets or derived variables at multiple levels were employed: NCEP/NCAR (daily, 250 km grid) sea level pressure (SLP) (hPa), geopotential height (HGT) (m), relative vorticity (RV) (10-5 S-1), zonal (U) and meridional (V) wind components (m s-1), and vertical velocity (omega: Pa s-1), along with ERA5 (daily, 25 km grid) relative humidity RH (%) as well as U and V wind components (m s-1).

In this research, months with the lowest (or no rainfall) and highest amounts and frequencies of precipitation events are determined throughout the year in the SiP. This is important, as in follow-up SiP research the aim is to assess the regreening impacts on local hydrometeorological processes such as precipitation recycling in the SiP under a vegetated surface scenario during a naturally dry period of the year. Thus, we developed a multi-statistical approach to split the wet and dry months of the year for the period 2001–2020. This is achieved via a combination of the results obtained from three statistical measures: (i) monthly 90th percentile (Fig. 3a), (ii) frequency of occurrence of precipitation with a threshold of ≥10  mm d-1 (Fig. 3b) – after examining other thresholds of ≥5 mm d-1 and ≥20 mm d-1 (see Figs. S1 and S2), and (iii) monthly rainfall standard deviations (Fig. 3c). These methods were calculated using a set of statistical functions described in the following subsection (see Table 1). Therefore, using the approach developed in this study, wet months are determined from October to March, defined as the wet period, and dry months from April to September are defined as dry periods in SiP.

The spatiotemporal analysis and statistical measures for the satellite GPM-based daily precipitation time series were carried out for the entire SiP region. For this, a set of climate functions and indices (see Table 1 for details) was computed for the period of 2001–2020 using the Climate Data Operator (CDO) (Schulzweida, 2020) developed at the Max Planck Institute for Meteorology (https://code.mpimet.mpg.de/projects/cdo, last access: 12 July 2020).

The spatiotemporal patterns of the daily precipitation climatology (annual and biannual) over the period of 2001–2020 in the SiP were analyzed using empirical orthogonal function (EOF) analysis. According to Dawson (2016), the main aim of EOF analysis is to reduce the dimensionality of a spatial–temporal dataset by transforming it to a new basis in terms of variance. This transformation turns the input spatial–temporal dataset into a set of maps representing patterns of variance and a time series for each map that determines the contribution of that map to the original dataset at each time step. Thus, the spatial patterns are the EOFs and are considered to be basis functions in terms of variance. The associated time series are the principal components (PCs) and are the temporal coefficients of the EOFs. In this study, we used a Python-based eofs package (Dawson, 2016) to perform the EOF analysis.

Furthermore, the trends of the annual and seasonal changes in the precipitation events were also estimated for three selected sites across the SiP (see Fig. 1 for the locations) using anomaly-based analysis. The climatology mean precipitation values and spatial distributions were the two main criteria for the selection of the sites. In this way, each chosen site is representative of its surrounding area in terms of both the precipitation magnitude and spatial patterns. Thus, the selected sites in the northern, southern, and middle parts indicate the max, min, and average amounts of precipitation received across the SiP, respectively, over a 20-year time period. For this analysis, the precipitation anomalies (annual and seasonal) are calculated in three steps: (i) calculating the climatology mean of the data, (ii) subtracting the mean value from each year and season value, and (iii) drawing the trend of slopes using the least-squares method. Here, winter includes DJF months (December, January, and February) and autumn includes SON months (September, October, and November). The anomalies for spring and summer periods were found to be close to zero and are therefore excluded. It is noted that we also performed 95 % and 99 % bootstrapped confidence intervals for the mean and median values of the original dataset (seasonal and annual) for the selected sites. The results are given in Table S1.

To explore the climatology of the synoptic, dynamic, and moisture conditions at multiple levels of the atmosphere responsible for the occurrence of the (extreme) precipitation events over the SiP, the reanalysis dataset obtained from NCEP/NCAR and ERA5 was investigated. In the first place, the wet period and dry period were determined as explained earlier (see Sect. 2.3.1). In the second place, using the satellite reanalysis variables (see Sect. 2.2.), the dominant synoptic features, dynamical circulation patterns, and moisture conditions accompanied by the spatial correlations between SiP's rainfall and key meteorological variables were computed and analyzed for the wet and dry periods for the climatology period of 2001–2020.

In line with the synoptic analysis, the daily trajectories of the rainy systems precipitated over the SiP were tracked and plotted for the wet and dry periods using a manual approach developed in this study. In our approach, we merely aimed to detect and track cyclones precipitating ≥10 mm in the SiP. It is, however, challenging for an automated algorithm to detect a low system (sometimes with multiple centers, cyclones) that may or not have generated rainfall with a given threshold over a given domain. Yet, its performance is not totally error-free, in particular over heterogeneous regions with a complex atmospheric planetary boundary layer (PBL) like the Mediterranean region (e.g., Raible et al., 2008; Flaounas et al., 2014; Prantl et al., 2021). Our manual-based cyclone-tracking approach developed in this research consists of three major steps as follows. i.

First, a set of daily total precipitation patterns over the SiP was produced using GPM data separately for the wet and dry periods; by doing so, a total of 156 events (out of 7305 d) were identified, which precipitated ≥10 mm over the SiP. Accordingly, synoptic-scale daily composites of SLP, 850 hPa RV, and streamflow were produced using the reanalysis dataset for the entire study period (2001–2020, 7305 d). Here, the 850 hPa relative vorticity and streamflow were used along with SLP to better identify the lows (Flaounas et al., 2014).

The spatial precipitation patterns in terms of the climatology average, the rainiest month, and the wettest day for the period of 2001–2020 in SiP are illustrated in Fig. 2a–c, respectively. The climatology map of precipitation markedly demonstrates that the northeastern and southwestern parts of the SiP receive the highest (between 100 and 150 mm yr-1) and the lowest (between 20 and 30 mm yr-1) amounts of annual rainfall, respectively (Fig. 2a). This implies that precipitation is unevenly distributed over the SiP. However, most parts of the region do not receive precipitation as high as 40 mm yr-1, except for the northern areas close to the Mediterranean Sea. With respect to the occurrence of precipitation extremes, we discovered that the rainiest month (out of 240 months) was March 2020 (Fig. 2b) with a wide range of rainfall values from 15 to 30 mm per month in the south and from 50 to 70 mm per month in the north. Interestingly, the wettest day (out of 7305 d) also occurred in the same month and year, which is 12 March 2020 (Fig. 2c); thus, it is not surprising to see an analogous spatial pattern when compared to the rainiest month but with less magnitude.

Additionally, we also identified the 12 rainiest months out of 240 months (see Fig. S3) and the 12 wettest days out of 7305 d (see Fig. S4). It was found that 9 out of 12 extreme month and day cases occurred in the winter season (January, February, and March) with the highest frequency of occurrence in January (five cases), while only 3 out of 12 cases took place in autumn (October and December). Further, we plotted monthly precipitation climatologies (2001–2020) together with ranks of 12 months (out of 240) with the highest amount of rainfall received in the SiP (Fig. S5). The most extreme precipitation event occurred in March 2020 over the past 2 decades, followed by February 2019 and January 2013. The severest storm was recorded during 11–13 March, and the peak rainfall hours (>30 mm) occurred in the afternoon of 12 March 2020, as shown by the onset and termination of the most powerful rainy system in hourly intervals of the subplot in Fig. S5c. It may be worth mentioning that the exceptional storm event of 11–13 March 2020 over the SiP is comprehensively investigated via a data analysis and simulation experiment approach in follow-up research. Overall, in almost all the precipitation cases either in climatologies or extremes, a similar spatial precipitation pattern was captured, meaning that the maxima were recorded in the north and the minima in the south of the SiP.

As shown in Fig. 2d–f, the dryness and wetness conditions across the SiP were also explored by computing the simple daily intensity index (SDII), number of consecutive dry days (CDDs), and number of wet days (RR1). It can be seen that the highest SDII is observed in the northeast with an intensity of ≥6 mm d-1. Interestingly, the lowest SDII is not seen in the south (even though the minimum precipitation magnitude and frequency are located there – see Fig. 2a), but in central parts of the SiP with ≤3 mm d-1 (Fig. 2d). CDD is remarkable in the south with ≥500 out of 7305 d, indicating that these areas receive less than 1 mm of rainfall for a long period; however, it gradually decreases northward with ≤300 d (Fig. 2e). Unlike CDD, it is not surprising to observe that RR1 is the lowest in the south (≤100 d) and innermost parts (≤200 d), but it rapidly increases towards the northeast of the region (≥350 d), as shown in Fig. 2f. These results are in good agreement with the abovementioned findings over the SiP.

i.

Percentile approach (Fig. 3a). The monthly 90th percentile of precipitation reveals that percentiles ≥10 mm per month are merely observed from October to March (wet period), while for the period from April to September (dry period) very low or no rainfall is measured, suggesting a prolonged naturally dry period in the SiP. Further, temporally, the winter months receive higher values (of extreme rainfall with 90th percentile) when compared to the autumn months (with <25 mm per month) during the wet period. Spatially, percentile maxima >50 mm per month are only seen in SiP's northeast across the year.

To investigate the patterns of precipitation variabilities in time and space in the SiP, EOF analysis was performed on the monthly precipitation dataset on the annual scale (Fig. 4). The first two leading EOFs account for 60 % and 11 % of the variance. The EOF1 spatial pattern is entirely in negative mode SiP-wide, indicating a below-average rainfall condition (drier trend), especially in the northern SiP (Fig. 4a). Correspondingly, the PC1 time series indicates a dominant negative temporal variability of EOF1 for the entire period (Fig. 4b). Conversely, it is seen that EOF2 values are mostly in positive mode, showing an above-average rainfall condition (wetter trend) in most parts, in particular in the southern SiP (Fig. 4c), and the positive temporal variability of the EOF2 is mostly seen in recent years, as shown in the PC2 time series (Fig. 4d). However, the northern SiP remains in negative mode, suggesting a severely decreasing trend in the annual precipitation rate when it is combined with EOF1, with 60 % of variance explained.

Besides annual analysis, seasonal spatiotemporal variabilities of the EOF patterns were also calculated separately for the wet period (Fig. S6) and dry period (Fig. S7). We found that annual EOFs and PCs strongly resemble the seasonal EOF spatial patterns and PC temporal variabilities in the SiP's wet period. This implies that both wet period patterns and annual patterns capture a decreasing trend in the north and an insignificant increasing trend in the south of the SiP. It is also noted that grid-based spatiotemporal variations obtained by the EOF analysis are in good agreement with the site-scale anomaly-based temporal changes in the annual and seasonal precipitation trends observed at selected sites across the SiP region (see Fig. S8 for details).

Figure 5 represents the precipitation regime climatology concerning the ratios and standard deviation estimates on a monthly basis over the SiP. High (>20 %) ratios of monthly precipitation over annual precipitation are estimated in the winter months of January and February, mostly found in the middle to north of the SiP. March indicates some patches of high ratios in the south and northwest also, as shown in Fig. 5a. However, the period of April to September (colored in black in the legend) receives less than 20 % of the annual precipitation. This implies that the spring and summer months experience longer dry weather periods than the winter season. Considering the autumn months, the areas with a 20 % ratio of annual precipitation remain largely out of the SiP domain, except for a few mini-patches. Therefore, winter is the rainiest season throughout the SiP. The monthly SD estimates (Fig. 5h) also follow a pattern similar to the ratios across the year. This means that temporally winter (summer) months hold the highest (lowest) variation values, and spatially the northern (southern) SiP possesses the highest (lowest) values with a max value of 17.0 mm per month estimated in March in the northeast of the SiP. It is also noted that the full ratios of monthly to annual precipitation for individual months of the year are illustrated in Fig. S9, and the full grid-based SD estimates for the entire SiP on a monthly basis are represented in Fig. 3c, which could provide further details on SiP's precipitation regime climatology.

Furthermore, to compare the precipitation monthly ratios across the SiP, bar charts for the given sites covering the whole SiP were plotted (Fig. 5b–g). The highest and lowest ratios are found in winter and summer months, respectively. However, with a closer look, it becomes clear that the chosen sites do vary in terms of magnitude and trends in the monthly precipitation ratios. For instance, at most sites the highest monthly ratio is observed in February (>18 %), except for the sites located in the SiP southwest (which is January with >19 % – Fig. 5d) and southeast (which is March with >17 % – Fig. 5g). Likewise, an inconsistent seasonal trend is also remarkable for the autumn months, meaning that the northern sites indicate a positive trend from the late summer to the end of autumn (Fig. 5a, b, e). The southern sites, however, represent a contrasting pattern with respect to the monthly rainfall regime (Fig. 5d, f, g).

Figure 6 represents the climatology of the synoptic patterns and cyclogenesis at the surface and 500 hPa atmospheric levels during the wet and dry periods over the Mediterranean basin including the SiP (marked by a red box). In the wet period at the surface level (Fig. 6a), two major sources of strong cyclonic activities (cyclogenesis) are observed over the Mediterranean western part (at the lee of the Alps over Gulf of Genoa) and eastern part (at the lee of Taurus Mountains over Cyprus) – see Fig. 12 for the locations. These areas are found by the closed SLP contours along with strong positive vorticity in the western and eastern parts of the Mediterranean Sea, respectively. The Cyprus low is allegedly responsible for the occurrence of the majority of rainfall events over the eastern Mediterranean including the SiP. For instance, about 80 % of the rainfall in the cold period of Israel is associated with Cyprus cyclone systems, as pointed out by Saaroni et al. (2010). In the wet period, the Red Sea trough, as a lower-level system, is another significant synoptic system that influences the eastern Mediterranean region, but mostly in the autumn (Ziv et al., 2021). As shown in Fig. 6a, this trough is developed as a result of the coexistence of the eastern African cyclone, namely the Sudan low and Saudi Arabian anticyclone. Its high impact on the eastern Mediterranean area depends on the position of the Red Sea trough axis: that is, the eastern position, as pointed out by, e.g., Saaroni et al. (1998) and Tsvieli and Zangvil (2005). However, the impact of the Red Sea trough on SiP's precipitation is limited compared to the northeastern parts of the Mediterranean basin, mostly due to the geographical location of the SiP. In line with lower levels, the pressure pattern at 500 hPa shows a synoptic-scale trough (of the persistent low center) with high positive vorticity, providing a suitable condition for the occurrence of rainfall events over the Mediterranean region extending towards the Middle Eastern areas (Fig. 6b).

In contrast to the wet period, the surface-level pattern of the dry period differs strongly over the region (Fig. 6c). In the dry period, hardly any cyclones are produced in the western Mediterranean as it is dominated by high-pressure systems extending from the North Atlantic Ocean and north of Africa. Limited low-pressure systems, however, are typically developed over the eastern Mediterranean. This is due to the formation of a trough extending from the Persian Gulf (which develops as the result of the topographic impact of the Zagros Mountains in western Iran) via the Taurus Mountains in southern Turkey into the eastern Mediterranean basin (see Fig. 12 for the locations). The SiP region located in the southeastern Mediterranean basin, as shown in Fig. 6c, is highly influenced by the ridge of the North African so-called Azores anticyclone rather than the Persian trough that mostly impacts the northeastern Mediterranean. Thus, at the midlevel of 500 hPa geopotential height, the eastern Mediterranean is mostly subjected to persistent air subsidence, and only a limited trough is formed with relatively high positive vorticity over the eastern Mediterranean (Fig. 6d). This prevents rainfall to a large extent over the region during the dry period. Therefore, the SiP receives much less precipitation in terms of magnitude and frequency compared to that received over the northeastern parts (such as Israel) of the Mediterranean basin. These results are in good agreement with the findings reported by Alpert et al. (1990) and Saaroni and Ziv (2000).

Besides the synoptic pressure systems described above, the vertical velocity motions (omega) could further reveal discrepancies between the wet and dry periods from a dynamical perspective. An increase in synoptic precipitation events over the wet period is inevitably attributed to the existence and duration of strong rising parcels of air and upward vertical streams over the SiP and in the nearby regions. The omega cross-section along the longitude (Fig. 7a) represents a maximum core with a negative value of -0.03 Pa s-1 at 800–700 hPa levels (above 36∘ E) extending up to 250 hPa. It also indicates that, unlike the western parts, the eastern parts of the SiP experience a relatively strong rising condition at multiple levels of the atmosphere during the wet period. A similar pattern analogous to the longitude cross-section is also observed along the latitude (Fig. 7b). This means that the maximum core of vertical velocity with the value of -0.006 Pa s-1 is seen towards the northeast of the Sinai (below 35∘ N), in particular at higher levels. However, when it comes to the dry period, a much weaker negative omega is observed, mostly limited to lower levels of the atmosphere along the longitude (Fig. 7c), and it is allegedly positive (sinking), in particular in the southern parts of the Sinai along the latitude (Fig. 7d). In such circumstances, the rising of air is strictly restricted. This (Fig. 7) therefore further clarifies, among others, why the northeastern parts of the SiP receive a higher (intense) amount of precipitation compared to the rest of the SiP: that is, partially due to the stronger vertical velocity motions in both the dry and especially wet periods.

Figure 8 illustrates the climatology of moisture conditions and wind patterns separately for the wet and dry periods in the SiP (red box) and in the nearby areas. Overall, a remarkable difference is observed with regard to the moisture availability during the wet and dry periods in the region, especially over the SiP. During the wet period, the prevailing westerlies at 850 hPa (which is typically considered the condensation level) over the Mediterranean Sea along with the presence of an anticyclonic circulation pattern over the north of Africa transferred abundant moisture (on average 50 %–70 %) to the eastern parts of the Mediterranean basin including the SiP (Fig. 8a). Also, the vertical cross-sections of moisture content and wind profile at pressure levels indicate that the majority of the moisture needed for condensation is found at lower levels (950–850 hPa) over the region (Fig. 8b). The abovementioned moisture and wind patterns, however, largely differ (RH reduced on average 25 %–45 %) during the dry period at the 850 hPa level (Fig. 8c) and pressure levels (Fig. 8d). This could be the result of the displacement of northern Africa's high-pressure center towards the higher latitudes (from 25 to 30∘ N), which resulted in the development of northwesterly streams over the region. Thus, unlike the wet period, less moisture is transferred into the SiP during the dry period.

In this section, daily-scale relationships of SiP's precipitation associated with the regional atmospheric variations responsible for the occurrence of wet and dry periods are explored. Figure 9a and b show the spatial correlation patterns between SiP's rainfall and regional sea level pressure (SLP) during the wet period and the dry period, respectively. A negative correlation (r=-0.1) is seen over the SiP. This indicates that there is a strong association between higher rainfall events (magnitude and frequency) and lower surface pressure fields over the eastern Mediterranean including the SiP in the wet period (Fig. 9a). In contrast, a positive correlation (r=0.25) is found between the rainfall and SLP over SiP (Fig. 9b), highlighting the dominance of high-pressure fields over the region that restrict rising of the air during the dry period. The spatial patterns at the midlevel of 700 hPa also represent a negative correlation (r=-0.24) between SiP's rainfall and geopotential height (HGT) during the wet period (Fig. 9c).

A similar spatial pattern with a higher correlation coefficient (r=-0.3) is also observed in the dry period. However, a significant decrease in the region's rainfall could be justified by the predominance of subtropical high-pressure centers and an increase in HGT during the dry period; thus, a meaningful relationship is formed between the two (Fig. 9d). The potential vorticity (PV) at the low level of 1000 hPa correlates positively with the rainfall in both wet and dry periods, indicating a cyclonic circulation in the lower atmosphere over the SiP region. However, positive PV (r=0.12) dominated the eastern Mediterranean including the SiP during the wet period (Fig. 9e), whereas its impact remarkably diminished over the region in the dry period (Fig. 9f), resulting in a decrease in precipitation in the eastern Mediterranean basin.

A coupling correlation pattern, as shown in Fig. 10, is observed concerning the precipitation and the meridional wind (V wind) at the 925 hPa level over the SiP during the wet period (Fig. 10a). This indicates that SiP's precipitation is positively correlated (r=0.12) with the southerlies found across the Middle East with a core in Mesopotamia (see Fig. 12 for the locations) but negatively correlated (r=-0.15) with the northerlies found over the central–eastern Mediterranean and north of Africa. This provides a suitable condition for moisture transport from the Red Sea (by the southerlies) and the Mediterranean Sea (by the northerlies) into the study area. In contrast, the region is dominated by southerly winds during the dry period (Fig. 10b), which limits the role of the Mediterranean in feeding the region with abundant moisture; thus, rain events are largely reduced. Interestingly, likewise, for the V wind, a similar coupling pattern is also observed between precipitation and zonal wind (U wind) at the 925 hPa level over the area during the wet period (Fig. 10c). In such circumstances, SiP's rainfall positively correlates (r=0.15) with westerlies over the eastern Mediterranean basin. However, in the dry period (Fig. 10d), SiP's precipitation is largely associated with the negative predominant westerlies over Mesopotamia and the north of Saudi Arabia. Finally, SiP's wet period precipitation correlates negatively (r=-0.18) with the omega in the lower atmosphere (at 850 hPa, Fig. 10e) over the eastern Mediterranean basin, indicating a strong vertical velocity. The relationships of SiP's rainfall and vertical velocity are largely weakened (r=-0.08) during the dry period (Fig. 10f), thus limiting the rising of air to a large extent.