Is there a relationship between humidity in the lower troposphere and seasonal precipitation in Iran? What factors play a role in these seasonal precipitations? Is there a relationship between weather and El Niño oscillations? What role does El Niño-Southern Oscillation (ENSO) play in seasonal precipitations in Iran?
This paper investigates the relationship between seasonal precipitation over Iran and low-level moisture, in terms of vertically integrated specific humidity (VISH) from the surface to 850 hPa. The VISH is calculated from ERA5 data for the domain (10N–60N, 15E–80E), and the precipitation is calculated from 50 stations across Iran, both for the period 1968–2023. Canonical correlation analysis (CCA) is applied to examine the spatial–temporal relationship between seasonal averages of moisture and precipitation during January– March (JFM), April–Jun (AMJ), and October–December (OND). VISH and precipitation are considered as the simultaneous predictor and predictand fields in the CCA, respectively. The CCA time series are correlated to global sea surface temperatures to assess the connections to large-scale, potentially predictable modes of variability. The CCA spatial patterns indicate that there is a strong relationship between low-level moisture and seasonal precipitation, with VISH over the Persian Gulf, Oman Sea, Arabian Sea, and Red Sea positively correlated with precipitation over most areas of Iran, while VISH over the Caspian Sea and Black is negatively correlated. Generally, these relationships are notably low over northwestern areas of Iran and the coastal regions of the Caspian Sea and the prediction skill of CCA remains limited over these regions. In OND, the leading CCA time series exhibits the well-known connection to the El Niño–Southern Oscillation (ENSO). However, the highest CCA skill is found for JFM precipitation, which does not exhibit an ENSO connection, and so may present an additional source of skill.
Atmospheric water vapor (AWV) is crucial in influencing the radiant properties and dynamics of the Earth's climate (Allen & Ingram, 2002; Trenberth et al., 2005). Variations in AWV impact the hydrological cycle (Trenberth et al., 2005) and the intensity of precipitation (Allen & Ingram, 2002). AWV is transferred to a region by the atmospheric circulation, and the amount of precipitation in that region is associated with the convergence of AWV transition. The highest total column water vapor (TCWV) values are observed over oceanic/marine regions during winter, while the highest TCWV values are observed over land and in particular over eastern Europe and European Russia in summer (Zveryaev et al., 2008). The relationship between the moisture flux convergent (MFC) and precipitation variabilities has been studied in different climates, including North America (Anderson et al., 2009), the Mediterranean basin (Şahin et al., 2015), Central Southwest Asia (Malik et al., 2015), China (Drumond et al., 2011; Yuan et al., 2021), the Arabian Peninsula (Athar & Ammar, 2016), Iran (Alijani, 1995; Darand & Pazhoh, 2019a), and western North America (Tan et al., 2022). The research conducted by Alijani (1995) underscores the pivotal role of the Mediterranean Sea in shaping moisture transport, which subsequently impacts precipitation trends across Iran. The Persian Gulf and Caspian Sea are crucial for supplying moisture leading to rainfall over the Zagros and Alborz Mountain ranges (Evan et al., 2003; Smith et al., 2003). Farajzadeh et al. (2007) noted that the Arabian Sea plays a vital role in providing moisture, especially at mid to high atmospheric levels, during the significant precipitation event that struck western Iran on January 7, 1996. Furthermore, the findings reveal that a substantial portion of the moisture is transported to Iran at lower atmospheric levels, emphasizing the influence of damp air flows from prevailing pressure systems in these lower levels. The presence of high-pressure systems and the considerable distance from oceanic moisture sources lead to subsidence patterns that exacerbate drought conditions in Southern Iran (Diaz & Bradley, 2004). Darand and Pazhoh (2019a) divided specific humidity (SH) of the troposphere from surface to 500 hPa, including 1000, 925, 850, 775, 750, 700, 600, and 500 hPa, over Iran into three layers including lower from surface to 850, middle from 775 to 700, and upper from 600 to 500 hPa using hierarchical cluster analysis method. They showed that the vertical integrated moisture flux convergence (VIMFC) of these layers is positively and significantly correlated with winter months (JFM) precipitation over Iran such that the relationship rate in lower layer is far more than other layers. Karimi et al. (2022) investigated the relationship between the variability of the vertical integrated moisture flux (VIMF) from 1000 to 300 hPa and precipitation over Iran during wet and dry years. Their findings revealed that VIMF in wet years was 16.54% higher than in dry years. Furthermore, they identified the Arabian Sea as the main source of moisture transfer into the lower layers of troposphere in both dry and wet years. Following the results of Darand and Pazhoh (2019a), the vertical integrated specific humidity (VISH) from 1000 to 850 hPa was utilized to examine the VISH-Iran's seasonal precipitation. Compared with Darand and Pazhoh (2019a), the geographic domain of the VISH was extended from Iran domain to a larger domain (10N– 60N, 15E–80E). In this manner, the major sources of moisture over surrounding areas, which are associated with seasonal precipitation over Iran, are determined. The abovementioned studies relied on the VIMF to analyse the link between SH and precipitation, aiming to elucidate the mechanisms of moisture transport into Iran. These studies explored the relationships between humidity and precipitation, and these relationships have not yet to be formulated into a predictive model. Therefore, the objective of the present study is to examine and model the seasonal prediction of precipitation of Iran's precipitation using VISH, which has not been considered so far. In this manner, the VISH is considered here as the predictor field to model the relationship between moisture in the low level with seasonal precipitation over Iran using canonical correlation analysis (CCA). Linear spatial–temporal pattern between moisture and precipitation can be identified in CCA and consequently the most important sources of atmospheric moisture in the lower level at surrounding areas and adjacent seas are herein identified. The seasonal prediction skill of CCA model is evaluated using deterministic and probabilistic skills. The resting of this paper is organized as follows. Section 2 describes the data and methodology. Section 3 presents the results. Section 4 demonstrates the conclusion. 2 | DATA AND METHODOLOGY 2.1 | The study area The study area is Iran, which is located in Southwest Asia from approximately 25N to 40N in latitude and 44E to 64N (Figure 1). Based on the Köppen–Geiger climate classification, most parts of the study area are categorized as arid and semiarid climates such that 33.5% of Iran is categorized as arid and 44.6% of it as semiarid climate (Raziei, 2022). Recently, Najafi and Alizadeh (2023) studied the climate zones in Iran using principal component analysis (PCA) and identified six distinct climate zones: warm and hyper-arid in central to southeastern, cool and subhumid in the western, warm and semiarid in the southwestern, cold and temperate semiarid in the northwestern, warm and semiarid in the northeastern of Iran, and mild and humid in the southern coastal plains of the Caspian Sea. Five major air mass types (Alijani, 1995; Sabziparvar et al., 2015) affect the study area (Figure 1). The Mediterranean air mass (MedT) affects the western and southwestern regions of Iran. Mediterranean cyclones are characterized by substantial moisture content, leading to increased cloudiness and precipitation, particularly in western Iran. In the southwestern part of the study area, the Mediterranean frontal system interacts with the Sudan heat-flow system to influence weather patterns, and their combined effect often results in heavy precipitation. The southwestern region of Iran is also impacted by the continental tropical air mass (cT), commonly referred to as the Sudan air mass. In contrast, the maritime polar (mP) air mass, which carries cool, moist air from the northern and northwestern parts of Europe, affects Iran's northwestern region. Additionally, the continental polar (cP) air mass, known as the Siberian air mass, influences the northern and northeastern regions, where its interaction with the Alborz Mountain range leads to significant precipitation along the Caspian Sea coast. Finally, the maritime tropical (mT) air mass, though infrequent, can trigger sudden, intense rainfall events in the southern and southeastern parts of Iran during the dry season. Figure 2 indicates that two main mountain ranges, the Alborz and Zagros, have an important role in the Iranian atmospheric systems, resulting in the amount and distribution of precipitation in Iran (Alizadeh & Babaei, 2022; Shirvani, 2017). This figure depicts the role of topography in the distribution of annual precipitation such that low elevation and salt deserts have little-to no precipitation. The Dasht-e Kavir and Dasht-e Lut deserts are cut off from humid air masses due to their position on the leeward sides of Alborz and Zagros mountains (Figure 2) and therefore have a hyper-arid climate (Najafi & Alizadeh, 2023). The mean annual precipitation in the study area ranges from less than 50 mm in these two deserts to over 1000 mm in the coastal plains of the Caspian Sea (Figure 2). The mean annual precipitation in Iran is estimated to be approximately 250 mm, which is less than one-third of the global mean, with most of the precipitation occurring between October to May (Alizadeh-Choobari & Najafi, 2018; Shirvani, 2017).2.2 | Data 2.2.1 | Precipitation data The monthly observed precipitation data were obtained from the Islamic Republic of Iran Meteorological Office (IRIMO). The quality control of precipitation data is checked by IRIMO before it becomes accessible to users on the IRIMO website. The data consist of 50 synoptic stations across Iran for the period of 1968–2023 (a 56-year period). The selected stations have no missing values. In this study, the Pettitt test (Pettitt, 1979) is applied for testing the homogeneity of monthly precipitation time series data for 50 synoptic stations. The geographical location of the selected synoptic stations is shown in Figure 2. Table 1 displays the characteristics of the selected stations. Figure 3 shows the monthly precipitation percentage over the study area for the period of 1968–2023, indicating that the largest precipitation total is observed from December through March. Three seasons are considered for prediction: JFM (winter, 36.3% of annual precipitation), AMJ (spring, 16.5% of annual precipitation), and OND (autumn, 36.6% of annual precipitation).FIGURE 5 The first canonical pair of VISH and precipitation for the JFM season, with the VISH pattern in (a), the precipitation pattern in (b), the corresponding time series in (c), where the black line and dashed red line show the precipitation and VISH series, respectively. The wet (1972, 1974, 1980, 1982, 1991, 1996) and dry (2001, 2008, 2013, 2016, 2018, 2021) years are indicated in the time series plot. 2.2.2 | Reanalysis data NCEP–NCAR (National Centers for Environmental Prediction–National Center for Atmospheric Research)
and ERA are two of the most commonly utilized reanalysis products (Fu et al., 2016). ERA-Interim shows an overall better performance than NCEP–NCAR for 22 climate variables across Australia (Fu et al., 2016). Similarly, Nacar et al. (2022) showed that air temperature and precipitation in the Eastern Black Sea Basin, Turkey, from ERA5 data in comparison with ERA-Interim and NCEP-NCAR are closer to values measured from the meteorological stations. Hassan et al. (2023) reported that four meteorological variables (geopotential height, air temperature, dew point temperature, and relative humidity) from ERA5 and ERA-Interim have a strong correlation with radiosonde station data in the Eastern Mediterranean region. The total precipitable water vapor from ERA- Interim in comparison with NCEP-NCAR data is closer to observed Radiosonde data in Shiraz synoptic station, Iran (Shirvani & Nuroozi, 2019). Consequently, the ERA5 dataset is selected as a reliable reanalysis dataset in this research. Monthly tropospheric SH data for the domain (10N– 60N, 15E–80E) are provided from ERA5 data for a common period from 1968 to 2023. The used spatial resolution is 0.25 0.25. VISH from surface to 850 hPa, in three vertical pressure levels (1000, 925, and 850 hPa), is defined as the specific moisture in the lower level troposphere. Reanalysis zonal (u) and meridional (v) components of winds as well as geopotential heights are also extracted from ERA5 data. Wind and VISH units are meter per second and gram per kilogram, respectively.2.2.3 | SST data Monthly global Extended Reconstructed sea-surface temperature anomalies (ERSSTA) version 5 at a 2 2 spatial resolution over the globe (Huang et al., 2017), during 1968–2023, are obtained from the NOAA Climate Prediction Center. Warm and cold phases of Oceanic Niño Index (ONI) are generally calculated using the ERSSTA version 5 (https://www.ncei.noaa.gov/access/monitoring/ enso/sst#oni). The Pacific decadal oscillation (PDO) and Atlantic multi-decadal oscillation (AMO) are also obtained from the NOAA data (https://www.ncei.noaa. gov/access/monitoring/pdo/; https://psl.noaa.gov/data/ timeseries/AMO/). 2.3 | Methodology The following steps and subsections are implemented to investigate and model the relationship between SH and precipitation across Iran: 1- Perform homogeneity test of precipitation data 2- Extract principal components (PC) of SH and precipitation using PCA 3- Assess the linear trend of PC time series using simple linear regression (SLR) FIGURE 6 The cross-validated Sperman correlation (a), rootmean-square error (RMSE) (b), and mean difference (prediction minus observation) (c) map for JFM precipitation prediction using the constructed CCA equations. Areas where the correlation values do not have local statistical significance at the 95% level are masked out.4- Develop a CCA model based on the extracted PCs 5- Evaluate the performance CCA model using deterministic and probabilistic skill metrics A detailed explanation of these steps is provided in the following subsections. 2.3.1 | Homogeneity test of precipitation data Ensuring the homogeneity of data is essential for obtaining trustworthy results. Pettitt test (Pettitt, 1979) is a useful nonparametric method to homogeneity test of precipitation data (Bickici Arikan & Kahya, 2019; Smadi & Zghoul, 2005). This test suggested by Pettitt to detect the change point in a series and investigates whether there is an abrupt change in a series. In this research, the Pettitt test is initially utilized to assess the homogeneity of monthly precipitation data. The Pettitt test is performed for each synoptic station. Those stations, which are homogeneous, are utilized as the input file of PCA. 2.3.2 | PCA PCA is a multivariate statistical method, which is used for dimensionality reduction while preserving as much variance as possible in a dataset. PCA creates new variables from original multivariate data such that the first new variables capture the maximum variance, the second variables capture the second most variance, and so on, while these new variables are uncorrelated. These new variables are called PCs (Jolliffe, 2002). PCA can be constructed based on the covariance, correlation, or coefficient of variation matrices (Boik & Shirvani, 2009). We use PCA based on the correlation matrix to ensure that each variable contributes equally to the analysis. Both SH and precipitation data are separately considered as the input file of PCA. The PC time series of these files are considered as the input file of CCA. However, linear trend of these series is checked before performing CCA. 2.3.3 | Linear trend of PC SLR (Wilks, 2011) is applied to check the linear trend of PC. The PC time series and time are considered as dependent and independent variables in the SLR. If the PC time series have a significant linear trend, the detrended time series using the fitted linear regression is considered as the de-trended time series and input variable of CCA. 2.3.4 | CCA CCA is a multivariate statistical method, which is used to understand and quantify the relationships between two sets of multivariate datasets. CCA identifies the optimal linear combination of two multivariate datasets (predictors and predictands) and selects pairs of patterns from spatially and temporally dependent variables to ensure that their coefficient time series are maximally correlated (Barnett & Preisendorfer, 1987). The resulting linear combination of input variables is referred to as canonical variables. Also, the correlation coefficient between pairs of canonical variables is called canonical correlation coefficient. In this study, CCA is applied to investigate the spatial–temporal relationships between VISH gridded data over (10–60N, 15–80E), as predictor filed, and Iran's precipitation station data, as predictand field. Therefore, CCA highlights the mechanism that SH controls regional precipitation variability over Iran and this is the reason why we use this multivariate method. Previous studies have not modelled seasonal precipitation predictions based on variations in humidity, which makes this study a pioneering effort in this field. It is suggested that the input file be preprocessed and orthogonalized before performing CCA (Barnston & Ropelewski, 1992). Therefore, prior to performing the CCA, pre-filtering with PCA (Jolliffe, 2002) is applied for data reduction of both predictor and predictand fields. In this study, the number of canonical variables and PCs, which is finally applied for prediction, is determined using cross-validation Spearman correlation (Wilks, 2011). The Spearman correlation coefficient (SCC) is constructed based on the ranks of the data and therefore is a robust and resistant correlation alternative to the Pearson correlation coefficient (Wilks, 2011). 2.3.5 | Evaluation of CCA model A 10-fold cross-validation method is used here to evaluate CCA model. Nine out of 10 subsamples are used for training of CCA, and one subsample is used for model validation. A bootstrap resampling method (Wilks, 2011) is applied as a significance test of cross-validation SCC. The number of bootstrap samples used is 500, and the confidence level is taken as 95%. Additionally, two deterministic skills including the cross-validated rootmean-square error (RMSE) and climatological mean difference (MD) and a probabilistic skill (relative operating characteristic [ROC]) are applied to evaluate CCA performance. The ROC is widely used for probabilistic prediction verification (Marzban, 2004; Mason, 1982; Troccoli et al., 2008). ROC scores close 1 and 0.5 indicate, respectively, perfect and random predictions, while scores above 0.5 indicate positive performance of model. In this study, cross-validation ROC (ROC) scores are computed for upper and lower 75th and 25th percentiles of the precipitation data, with the upper 75th is considered as the above-normal precipitation and lower 25th percentiles as the below-normal precipitation. The Climate Predictability Tool (CPT) version 17.7.4 is used to run the abovementioned statistical methods. We note that, because the same season is used for both predictor and predictand, the prediction skill measures the closeness of the relationship between the two fields, not the potential seasonal forecast skill. 3 | RESULTS The results of the Pettitt test revealed that five (Ardebil, Neyshabour, Gorgan, Khoy and Khash) out of 50 stations are nonhomogeneous at 95% significance level. These stations are excluded from further analysis. Figure 4a shows the first PC scores time series for JFM precipitation. The fitted linear trend of this series is also plotted in this Figure. The equation of the fitted linear trend is expressed as follows: PCscoreð Þ¼ t 0:03t þ1960,t ¼ 1968,1969,…,2021, where the estimated coefficient, 0.03, with t statistic and p-value of 3.2 and 0.002, indicating a significant decreasing trend for the first PC scores time series at 95% significance level. The notable decrease in trend suggests that the risk of drought during the JFM season has increased in Iran. Figure 4b shows the de-trended time series, using this fitted linear regression, and indicates no trend. This detrended time series is considered as the input series of CCA. The reminder PC score time series for JFM precipitation and all PC score time series for AMJ and OND have no significant trend. These results are like previous study (Shirvani & Landman, 2022) that reported a significant (non-significant) decreasing trend in the time series of JFM (OND) standardized precipitation index (SPI) over Iran for the period 1968–2017. The highest average cross-validated Spearman correlation between observed and predicted JFM precipitation is produced by the first pair of canonical variables. The association between the spatial pattern of VISH and JFM precipitation over Iran is shown in Figure 5a,b, respectively. The associated time series is shown in Figure 5c. The canonical correlation value between this first pair of canonical variables is 0.88, which is statistically significant at 95% significance level.The first pair of spatial patterns (Figure 5a,b) indicates a positive association between the spatial pattern of the VISH over the Persian Gulf, Oman Sea, Arabian Sea, and southern parts of Red Sea and precipitation for JFM over most areas of Iran. However, the VISH over northern latitudes (38N–60N) has a negative association with JFM precipitation over most areas. Figure 5c indicates that above normal JFM precipitation in seasons such as 1972, 1974, 1980, 1982, 1991, and 1996 go with high VISH over the Persian Gulf, Oman Sea, Arabian Sea, and southern parts of Red Sea (Figure 5a,c). Moreover, belownormal JFM precipitation such as in 2001, 2008, 2013, 2016, 2018, and 2021 go with low VISH over those seas. The cross-validated SCC between observed and predicted JFM precipitation is statistically significant at the 95% significance level for 35 out of 45 synoptic stations (Figure 6a and Table 2). At the 95% significance level, significant cross-validated SCC values between observed and predicted JFM precipitation are observed over most areas of Iran (Figure 6a). However, precipitation predictions are not significant in the northwestern regions, as well as in some parts of northern areas of Iran. The spatial patterns of cross-validated RMSE and MD between observations and predictions, using the constructed CCA, are calculated and plotted in Figure 6b,c, respectively. The prediction error for north areas, where the amount of precipitation is high, is greater than in other areas (Figure 5a). Figure 6c indicates that the precipitation is overestimated for the JFM season in most regions, while it is underestimated in the north and northwestern regions. The spatial patterns of cross-validated ROC scores for precipitation categories are presented in Figure 7. This figure indicates that for both upper and lower 75th and 25th percentiles of the JFM season, the constructed CCA model produces higher ROC scores in the south, east, and west regions of Iran. However, in the northwestern and northern regions of Iran, the model produces lower ROC scores. Overall, the maps of JFM precipitation prediction skill demonstrate substantial performance in most areas, except for the northwestern areas and some parts of the north. These results indicate that the VISH has a strong association with JFM precipitation across most areas of Iran. Figure 8a,b illustrate the first pair of canonical variables, identified as the optimal pair, between VISH and AMJ precipitation over Iran. This indicates a positive association between VISH in the southern latitudes (10N–32N) and AMJ precipitation across most areas of Iran. High coefficients in the spatial pattern of VISH are observed over the Persian Gulf, Oman Sea, Arabian Sea, Aden Gulf, and Red Sea (Figure 8a). Conversely, the lowest coefficients for the spatial pattern of AMJ precipitation are observed in the southeast of Iran (Figure 8b). There is a strong and significant canonical correlation between the first pair of canonical variables for the AMJ season. Figure 8c presents the first pair of canonical variables for this season over the period from 1968 to 2022. Above-normal AMJ precipitation, observed in years such as 1968, 1969, 1972, 1976, 1992, and 1995, corresponds with high VISH values over the aforementioned seas (Figure 8a,c). Conversely, below-normal AMJ precipitation in years such as 1985, 1989, 2001, 2008, and 2015 is associated with low VISH values over these same seas. There are statistically significant cross-validated Spearman correlations between observed and predicted AMJ precipitation at the 95% significance level for 38 out of 45 stations (Figure 9a and Table 2). Notably, the correlation coefficient reaches a high of 0.73 at the Tehran station. However, in southeastern Iran, the correlations are weak and not statistically significant. During AMJ season, the RMSE in the northern regions exceeds that of other areas (illustrated in Figure 9b). Additionally, the MD is low, indicating an overestimation of precipitation across most regions during the AMJ season (as shown in Figure 9c). The ROC cores reveal that the prediction skill for above-normal precipitation is superior to that for belownormal conditions during this season (depicted in Figure 10a,b). The prediction skill maps for AMJ demonstrate significant predictive ability across most of the study area, with the exception of the southeastern regions, which is consistent with the ROC findings. During OND season, the highest average cross-validated Spearman correlation is achieved using a pair of canonical variables. The first pair, illustrated in Figure 11a,b, indicates a positive correlation between VISH over the Persian Gulf, Oman Sea, Aden Gulf, Red Sea, and northern parts of the Indian Ocean and OND precipitation across Iran. The corresponding time series are displayed in Figure 11c. A significant canonical correlation of 0.75 between VISH and precipitation is observed at the 95% confidence level. Instances of above-normal OND precipitation, such as those recorded in 1977, 1982, 1997, 2000, 2015, and 2018, are associated with high VISH over the aforementioned seas (as shown in Figure 11a,c). Conversely, years of below-normal OND precipitation, including 1973, 1974, 1990, 1996, 1998, and 2010, are linked to reduced VISH over the aforementioned seas. The SCC for observed versus predicted OND precipitation is statistically significant at 33 out of 45 synoptic stations (Figure 12a and Table 2), with the highest correlation of 0.68 observed at the Ahvaz station. A significant crossvalidated SCC is observed between the observations and predictions for the OND season across eastern, western, and southern regions of Iran Figure 12a–c present the spatial distributions of cross-validated RMSE and MD for OND season as derived from CCA, respectively. The RMSE is high in the northern and northwestern areas compared with other regions (Figure 12b). Additionally, Figure 12c reveals that precipitation during the OND season is frequently overestimated across most regions. The cross-validated ROC scores suggest that the established CCA model achieves higher ROC values in the majority of regions, although the coastal areas of the Caspian Sea exhibit lower ROC scores relative to elsewhere (Figure 13a,b). Furthermore, the atmospheric moisture in the middle layer (775–700) and upper layers (600–500 hPa) as well as TCWV have been analysed as predictor fields in both PCA and CCA to forecast seasonal precipitation in the study area (specific figures not displayed). The highest predictive skill is observed from the low-level data (surface to 850 hPa) in relation to seasonal precipitation. Moreover, the lowest and insignificant skill is for upper layer and the CCA pattern of TCWV is similar to the lowlevel pattern. Findings from Darand and Pazhoh (2019a) also highlight that the connection between JFM precipitation and VIMFC in the lower atmospheric layer is notably stronger than that in other layers. To develop a seasonal precipitation prediction model using the VISH (as a single predictor) is effective for operational forecasts when contrasted with VIMFC, which relies on the wind component that its variability is high and difficult to predict. Nevertheless, to comprehensively explain and understand the physical processes involved in the transport of atmospheric moisture into the region, VIMFC remains essential and indispensable. Figure 14a–c present the Pearson correlation maps illustrating the relationship between the time series of the first canonical VISH CCA model and global ERSSTA for the three seasons. Figure 14c reveals that the OND CCA mode is correlated with the ENSO SST pattern, consistent with the well-known ENSO influence on this region. The JFM CCA mode, however, is not linearly correlated with SSTA over the eastern and central parts of the Pacific Ocean associated with ENSO, as depicted in Figure 14a. Instead, the JFM mode displays statistically significant correlations with SSTs over parts of the western tropical and subtropical Pacific (WTSTP), the subtropical and mid-latitude area of the North Atlantic (STMLNA), some parts of the northern Indian Ocean (NIO), and the eastern Mediterranean Sea and Black Sea (EMSBS). For further analysis, the area-average of SSTs in these four areas are computed to check which of these areas are more important and highly correlated with the first canonical VISH CCA time series. The Pearson correlation coefficients between this canonical time series and the SSTs over these areas indicate that the SSTs over WTSTP and EMSBS are significantly correlated with VISH CCA in the JFM season. When the SSTs over these four areas are simultaneously considered as the predictors in the stepwise regression for predicting VISH CCA, only the SSTs over the eastern Mediterranean Sea and Black Sea are significant and remained in the regression equation. This connection suggests a potential source of predictability in addition to ENSO for the region,although whether the JFM SST pattern in Figure 14a represents a potentially predictable mode of coupled or oceanic variability, or whether it is the imprint of intrinsically unpredictable atmosphere-only internal variability remains to be established. Moreover, the Pearson correlation coefficients between the VISH CCA time series with the PDO and AMO were not statistically significant during JFM season. Figure 15 presents the composite maps of the first VISH CCA patterns derived from the CCA and vector wind anomalies at low level for JFM, AMJ, and OND seasons. Figure 15a indicates that the VISH CCA pattern over the Persian Gulf, Oman Sea, Arabian Sea, and southern parts of Red Sea as well as a low-level anticyclone centred over the Arabian Sea is connected with JFM precipitation. Moreover, the VISH and SSTs over the Mediterranean Sea are positively correlated with Iranian precipitation during the JFM season and also the wind flows from the west to east over this sea. During the AMJ season, the wind flow is south easterly over the Mediterranean Sea and there is no predominant wind direction over Saudi Arabia, and therefore, there is a decreased moisture transfer from Saudi Arabia to Iran during this season (Figure 15b). The OND patterns over the south waters are similar to JFM season but the VISH south parts of the Mediterranean Sea are connected with OND precipitation (Figure 15c). The previous works also reported that the Arabian anticyclone over the Arabian Sea and the eastward movement of the Mediterranean trough towards Iran are the appropriate models of moisture flux over Iran (Farajzadeh et al., 2007; Karimi et al., 2022).4 | CONCLUSION The purpose of this study is to identify the links between seasonal precipitation over Iran and the important sources of atmospheric moisture in the low level (surface to 850 hPa) over an area between 10N and 60N and from 15E to 80E, using CCA and PCA methods. The spatial patterns identified by CCA indicate that there is a strong relationship between low-level moisture and seasonal precipitation such that VISH over the Persian Gulf, Oman Sea, Arabian Sea, and Red Sea is positively correlated with seasonal precipitation over most areas of Iran. Furthermore, the spatial CCA patterns have indicated that there is a negative association between the VISH over the Caspian Sea and east of the Black and Arabian Seas with seasonal precipitation over most areas of Iran. Darand and Pazhoh (2019a) also reported that Iran's atmospheric moisture sources at lower levels are southern water bodies that is supplied by Saudi Arabia dynamic high-pressure system and Sudan's low-pressure system. The primary source of moisture for the prevalent and widespread precipitation associated with low-level originates from Oman, the Arabian Sea, and the Persian Gulf and this moisture are directed towards Iran through high-pressure clockwise flows over the Oman Sea and low-pressure clockwise flows over Saudi Arabia. However, the moisture contributing to precipitation along the southern coasts of the Caspian Sea originates from the Caspian Sea itself and this moisture is transported to these areas by the cold air associated with the European high-pressure system (Darand and Pazhoh, 2019b; Darand et al., 2019). Based on the CCA, the precipitation prediction skill over northwestern areas and the coastal regions of the Caspian Sea are notably low, particularly during the JFM and OND seasons. The Spearman correlations of these regions does not show statistical significance, indicating that the constructed CCA patterns are not relevant to the relationship between moisture levels in the low level and seasonal precipitation in these regions. Building on previous studies (Masoodian, 2011; Mirmousavi et al., 2020), it is recommended to focus on precipitation in the southern coasts of the Caspian Sea as the target variable for analysis, rather than considering the entire region of Iran. Mirmousavi et al. (2020) highlighted the significant role of the Black Sea high-pressure system in the low level in transporting moisture to the southern coasts of the Caspian Sea, where the prediction skill of CCA remains limited. The correlations between the CCA modes and SSTA indicate that the regional precipitation–moisture variability is linked to large-scale climate variability. In OND, the VISH over the Persian Gulf, Oman Sea, Adan Gulf, and southern parts of the Red Sea is correlated with seasonal precipitation over Iran and both above-normal VISH over these seas and warm SSTs over the central Pacific and cold SSTs over the western Pacific, in an El Niño pattern. This is consistent with previous researches, which has shown that the La Niña conditions and warm SST over the western Pacific enhance the probability of severe droughts over the Middle East (Barlow et al., 2016; Barlow & Hoell, 2015) and central south Asia (Barlow et al., 2002; Hoell et al., 2017). Although the well-known ENSO relationship is clear in the OND season, the closest relationship is found for JFM precipitation and is not correlated to ENSO. This JFM relationship is potentially a source of forecast skill beyond ENSO for the region and merits further attention. To compare these CCA patterns with circulation patterns, the composite maps of moisture in the low level and vector wind are plotted in Figures 16–18. It is observed that the moisture anomalies over the Oman Sea, Persian Gulf, Arabian Sea (as the main source of supplied moisture for Iran), and most areas of Iran are above normal during wet season, and this is composited with inflow wind direction into Iran (Figures 16a, 17a, and 18a). On the other hand, the moisture anomalies over abovementioned seas are below normal during dry season and are composited with outflow wind direction into Iran (Figures 16b, 17b and 18b). These low-level patterns are connected with upper circulation patterns. Figure 19a (Figure 19b) shows that the below- (above) normal geopotential heights at 250 hPa over the Mediterranean Sea, Iraq, and Iran are associated with wet (dry) condition during JFM season. The below-normal geopotential heights at 250 hPa over the whole area-centred over Mediterranean Sea, Iraq, and Iran are associated with wet condition during AMJ season (Figure 20a). However, the association between dry AMJ season and above-normal geopotential heights at 250 hPa over Mediterranean Sea, Iraq, and Iran is weak (Figure 20b). The below- (above) normal geopotential heights at 250 hPa— centred over Iran and China—are associated with wet (dry) condition during OND season (Figure 21). These upper layer patterns are consistent with mid- and lowlayer patterns. The relationship between the 500-hPa and climate of Iran revealed that the regional troughs and ridges close to the country significantly influence winter precipitation and temperature (Alijani, 1995). Also, there is a negative association between the lowlevel geopotential height fields over Red Sea, Saudi Arabia, and Persian Gulf and monthly precipitation in Iran during January, February, March, April, and November (Shirvani et al., 2019). Therefore, the upper circulation patterns as well as the amount of moisture and vector wind direction over the abovementioned water bodies play key roles in dry and wet conditions in Iran. The simultaneous relationship between the JFM CCA mode and SSTs in the Mediterranean Sea and western Pacific Sea suggests the possibility of seasonal prediction skill beyond that associated with the well-known ENSO teleconnection. Establishing the predictability of this large-scale mode and how it translates to the predictability of Iranian precipitation is an important next step. How well this mode is represented in climate models and how it is projected to change is also an important question for this water stressed region. There is a substantial range in the ability of current climate models to reproduce the ENSO teleconnection to the region (Barlow et al., 2021); the apparent importance of this additional mode suggests that it needs to be evaluated in climate models as well.