Quantifying the Spatial Characteristics of the Moisture Transport Affecting Precipitation Seasonality and Recycling Variability in Central Asia

As a typical semiarid and arid zone, Central Asia (CA) (Fig. 1), including Kazakhstan, Uzbekistan, Turkmenistan, Kyrgyzstan, Tajikistan, and a part of Northwest China, is highly sensitive to climate change (Giorgi, 2006; de Beurs et al., 2018). With varied geography and dense but unevenly distributed population, CA is under threat from high water stress and societal vulnerability (Yang et al., 2011; Wei and Wang, 2013; Baldwin and Vecchi, 2016). Over recent decades, millions of people in CA depending on water from snow- and glacier-melt driven rivers have suffered from water stress and water allocation conflicts due to warming climate forcing and significant land-cover and land-use changes (Klein et al., 2012). Furthermore, water availability may be a more challenging issue in the future due to climate change and expected tremendous population growth in this densely populated region (Siegfried et al., 2012). Given the vulnerability of CA to variations of precipitation and the hydrological cycle, it is therefore a meaningful and critical challenge to identify the moisture sources and to understand the relationship between water vapor transport and precipitation in CA (Yatagai and Yasunari, 1998; Baldwin and Vecchi, 2016).

Figure 1.  Map of the study region, i.e., Central Asia (CA). The study region is shown by the area with color shading by terrain elevation (units: m). In the plot, the abbreviations in normal font are the country names, i.e., TM for Turkmenistan, UZ for Uzbekistan, TJ for Tajikistan, KG for Kyrgyzstan, KZ for Kazakhstan, RU for Russia, MN for Mongolia, CN for China, IR for Iran, AF for Afghanistan, and PK for Pakistan. The abbreviations in italics represent several geographic regions, with CS for Caspian Sea, AS for Arial Sea, BL for Balkhash Lake, TB for Tarim Basin, and TP for the Tibetan Plateau.

The precipitation in CA has strong spatial heterogeneity in both climatology and long-term trend. The spatial distribution of the mean precipitation in CA, which is characterized by arid to semiarid climates, varies significantly as a result of complex terrain and a geographic location far from oceans. In CA, a relatively large amount of annual precipitation (about 900 mm) is found in mountain regions, and much less annual precipitation (below 50 mm) occurs in desert areas (Domrös and Peng, 1988; Böhner, 2006). Apart from the uneven distribution in mean precipitation, the variations in precipitation in recent decades also show opposite trends between western and eastern parts of CA under almost consistent warming rates (Shi et al., 2007; Lioubimtseva and Henebry, 2009; Wei et al., 2010; Zhai et al., 2010; Hu et al., 2016; de Beurs et al., 2018). This striking contrast in the precipitation trends inside CA suggests divergent responses to global warming between the externally transported moisture and the local evaporation in different areas of CA.

Variation in precipitation is mainly determined by the physical processes associated with the air moisture content, which consists of three sources, i.e., moisture locally residing in the air, advected moisture from external sources, and recycled moisture evaporated from the local surface (Brubaker et al., 1993; Drumond et al., 2011). For advected moisture, previous studies on the CA moisture transport (Yatagai and Yasunari, 1998; Schiemann et al., 2008; Bothe et al., 2012; Baldwin and Vecchi, 2016; Hua et al., 2017; Jiang et al., 2020) have concluded that CA precipitation is mainly influenced by three main moisture pathways, which follow midlatitude westerly, northwesterly, and southwesterly pathways, respectively. The first two pathways originate from the upstream Eurasian continent, the Mediterranean, and the North Atlantic Ocean, and transport moisture to both north and south CA; the third pathway mainly comes from the Pacific, South Asia, and the Indian Ocean, and greatly contributes to the precipitation over south CA. These pathways correspond to different circulation systems with significant seasonal variations, e.g., the southwesterly transport in relation to the monsoonal circulation (Yatagai and Yasunari, 1998).

Apart from externally sourced moisture, recycled water vapor from local evaporation can also act as an important moisture source in the presence of weak moisture advective, particularly for arid areas such as northwestern China (Wang et al., 2016). The recycling ratio, i.e., the proportion of the total precipitation contributed by local evaporation, is a useful diagnostic measure of the potential for interaction between land surface hydrology and regional climate (Eltahir and Bras, 1994), reflecting both natural and anthropogenic influences (Trenberth et al., 2003; Dominguez et al., 2006; van der Ent et al., 2010; Gimeno et al., 2012; Zhang et al., 2012). Moreover, the variation in the recycling ratio and its match relationship with the moisture transport can reflect well how the large-scale circulation and the local conditions influence the precipitation variability under the background of climate change. For the arid and semiarid area studied here, including the five countries of CA and northwestern China, the precipitation contribution from local evaporation should play an important role in the regional water budget. The variation of the recycling ratio in CA is thus a good indicator of the transition between external and local moisture sources. The coupling relationships between the recycling precipitation and the moisture contribution from different transport pathways for CA precipitation have not yet been studied in detail by previous research and are worth in-depth investigation. This is an important issue for understanding the hydrological processes in CA.

In addition to the spatial heterogeneity of CA precipitation, noticeably different or opposite precipitation trends between the eastern and western parts of CA have been found at interdecadal and long-term scales with significant seasonality (Chen et al., 2010; Jiang et al., 2020; Zhong et al., 2021). Based on the temporal and spatial characteristics of the moisture transport, this study explores how the moisture sources of CA influence the precipitation variations and how the different pathways of moisture transport lead to the spatial inconsistency of the precipitation variations inside CA.

To do this, we first try to extract the main pathways of the moisture transport to CA and estimate the quantitative contribution of moisture sources by employing a temporally backward Lagrangian water cycle model. Based on the quantitative hydrological model and vector clustering method, we identify the impact areas of different pathways of moisture transport in CA. We then determine the climatological mean and seasonal, interannual, and even longer time-scale variations of the moisture contributions due to different transport pathways, as well as the recycling processes. This paper is organized as follows. Section 2 gives a simple introduction to the data, water cycle model, and clustering method used. Detailed results are presented in section 3, including the mean and seasonal pathways of moisture transport to CA, the clusters of moisture trajectories and their impact area, and the temporal variation of different-pathway moisture transport. In section 4, we summarize the main findings of this study and present some related discussions.

As shown by Fig. 1, CA is situated far from the ocean in the central part Eurasia and is approximately defined by the domain of (35º–56ºN, 45º–100ºE), comprising Northwestern China, the five CA countries including Kazakhstan, Tajikistan, Uzbekistan, Kyrgyzstan, and Turkmenistan, and their surrounding regions. In general, CA can be divided into two parts by the mountainous area mainly composed of the Tianshan and Karakoram Mountains situated between Northwestern China and the five CA countries. These mountains isolate southeast CA from the other parts and extend on the west and east sides. The terrain of CA reaches its highest elevation of 4000–5000 m in central south CA, mainly in the Pamir region and the western part of the Tianshan Mountains, and it reaches its lowest elevation in west CA around the Caspian Sea (Badescu and Cathcart, 2011).

Meteorological data for CA were derived from the ERA-Interim (ERAI) reanalysis with 1° × 1° horizontal resolution (Dee et al., 2011), including the daily accumulated precipitation and evaporation based on the 3-h forecast fields, and precipitable water, vertically integrated water vapor, and vertically integrated eastward and northward water vapor fluxes based on 6-h analysis fields. The hydrological analysis performed in this study focuses on the period from January 1979 to December 2015. Despite the uncertainties in the physical quantities from the reanalysis data, such as precipitation and evaporation (Wang and Dickinson, 2012), introduced by numerical errors and assimilation (Trenberth et al., 2011), the data is still considered a good proxy for water cycle studies and has enriched our understanding of the hydrological cycle in different regions (Dominguez et al., 2006; van der Ent et al., 2010; Wei et al., 2012; de Vries et al., 2016), including CA (Song and Bai, 2016). In fact, reanalysis data is critical for analyses of the hydrological cycle because of the lack of routine measurements of surface fluxes, which are important parts of the atmospheric moisture cycle. In order to evaluate the precipitation characteristics described by ERAI in CA (section 3.1.), we also use precipitation data from the Global Precipitation Climatology Centre (GPCC) (Schneider et al., 2015), which has the same spatial resolution and temporal coverage as the reanalysis data.

Lagrangian approaches are suitable and realistic for identification of and establishing the relationship between moisture sources and sinks (Gimeno et al., 2012; Hu and Dominguez, 2015; Ciric et al., 2016; Hua et al., 2017). To do this, a spatially unbounded Dynamic Recycling Model (SUDRM), which was developed by Dominguez et al. (2006) and extended by Hua et al. (2017), was applied to quantitatively identify the moisture trajectory with water vapor contribution entering CA based on ERAI daily data. The SUDRM is a 2D model based on the vertically integrated moisture budget equation. For a study region $ \mathrm{\Omega } $, the moisture variation from the reanalysis abides by the equation (Hua et al., 2017)

where $ w $, $ E $, $ P $, and $\boldsymbol{Q}=\left({Q}_{x},{Q}_{y}\right)$ are the total-column moisture, evaporation, precipitation, and the vertically integrated moisture flux over region $ \mathrm{\Omega } $, respectively. These 2D variables are derived from the ERAI daily fields with horizontal resolution of 1º × 1º for 1979–2015. The term $ \varepsilon $ is the residual due to the uncertainties of the reanalysis data. As shown in Table 1, the magnitude of the residual is generally of a lower order than the main-balance terms of the water vapor equation [Eq.(1)] for CA in the four seasons. So, the term $ \epsilon $ can be omitted in most instances in this study. The term of divergence/convergence of moisture flux ($\nabla \cdot \boldsymbol{Q}$) also satisfies $\nabla \cdot \boldsymbol{Q}=\boldsymbol{V}\cdot \nabla w$, with $\boldsymbol{V}=\left(u,v\right)= $$ {\displaystyle\int }_{0}^{{p}_{0}}[\boldsymbol{v}\left(p\right)q/\mathrm{g}]\mathrm{d}p$ as the humidity-weighted wind velocity of the whole air column, where $\boldsymbol{v}\left(p\right)$ is the wind velocity at pressure level $ p $ and $ {p}_{0} $ is the surface pressure (Dominguez et al., 2006; Berrisford et al., 2011).

If we partition the precipitation into the recycling precipitation ($ {P}_{\mathrm{m}} $) from local evaporation and the advective precipitation ($ {P}_{\mathrm{a}} $) from external sources, i.e., $ P={P}_{\mathrm{m}}+{P}_{\mathrm{a}} $, the recycling ratio $ \rho ={P}_{\mathrm{m}}/P $ is the proportion of the precipitation contributed by local evaporation in the target study area (such as CA in this study). Similar to precipitation, moisture over a location is also composed of two parts, i.e., $ {w=w}_{\mathrm{m}}+{w}_{\mathrm{a}} $, where $ {w}_{\mathrm{m}} $ is the recycling moisture and $ {w}_{\mathrm{a}} $ is the advective moisture. According to Eq. (1), the equation of mass conservation for $ {w}_{\mathrm{m}} $ can be derived by substituting the two parts of $ w $ and $ P $ into it (Dominguez et al., 2006):

where the residual $\varepsilon$ in Eq. (1) has been dropped.

The dynamic recycling model assumes these two parts of moisture are well mixed (Dominguez et al., 2006), which leads to ${w}_{\mathrm{m}}/w={P}_{\mathrm{m}}/P=\rho$. Substituting $ {w}_{\mathrm{m}}=\rho w $ and $ {P}_{\mathrm{m}}=\rho P $ into Eq. (2) and combing with Eq. (1), Eq. (2) can be transformed into the recycling ratio equation (Dominguez et al., 2006):

Under Lagrangian coordinates $(\chi ,\xi ,\tau )= (x-ut,y-vt,t)$, the terms in Eq. (3) can be transformed into

Combining with Eqs. (4–6), the recycling ratio equation Eq. (3) in Lagrangian coordinates has the form of

For the Lagrangian moisture trajectory $\boldsymbol{s}\left(\tau \right)=\boldsymbol{s}\left(t\right)$, Eq. (7) describes how the recycling ratio changes along the trajectory. Further integration of Eq. (7) along the 2D Lagrangian trajectory $\boldsymbol{s}\left(t\right)$ yields the solution (Dominguez et al., 2006):

Therefore, the SUDRM first generates the moisture trajectory for the target location by the moisture tracking method (Vries and Döös, 2001) and then calculates $ \rho $ along the moisture trajectory by approximating Eq. (8). With the recycling ratio $ \rho $ of any location on trajectory $ \boldsymbol{s} $, the moisture contribution over the trajectory segment $ \left[\boldsymbol{s}\left(t\right),\boldsymbol{s}\left(t+\Delta t\right)\right] $ can be expressed by $ {\left[\rho \left(t+\Delta t\right)-\rho \left(t\right)\right]}_{\boldsymbol{s}}w $. For any given location, the moisture contribution can be further projected via summing the moisture contribution of all the trajectory segments passing this location. The resultant map of the moisture contribution gives the distribution of the moisture sources. Similarly, through simply summing the trajectory number over any given place, we can get the trajectory count, i.e., the concentration of the moisture trajectory that reflects the features of the moisture transport pathways (Hua et al., 2017). In this study, the trajectory count and moisture contribution were calculated by the SUDRM to show the features of moisture transport to CA.

Different from the dynamic model for regional recycling processes, the SUDRM traces moisture sources without spatial restriction, i.e., the moisture contribution from anywhere on earth can be traced. For daily precipitation in CA, this study uses a 15-day backward moisture tracking method to detect the moisture sources and contributions. According to the work of Sodemann and Stohl (2009), the 15-day backward moisture trajectory can generally attribute about 90% of the moisture over a sink region, which can capture the characteristics of the moisture transport well. More detailed descriptions of this model and its application can be found in the previous work of Hua et al. (2017) and Zhong et al. (2019).

Cluster analysis is also employed to separate the major moisture transport pathways for CA precipitation in different seasons. The self-organizing map (SOM) clustering technique is an effective tool to exact and classify the information from input data based on an unsupervised neural network (Kohonen, 1982, 1998). Compared to the traditional linear method, the advantage of SOM is that it can map nonlinear high-dimensional structure in data onto a two-dimensional linear space. At the same time, the topological relationships between the input data can be preserved (Cavazos et al., 2002; Liu et al., 2006). In previous work, SOM has been applied to the analysis of meteorological and climatological variables in order to extract their primary structures and features (Risien et al., 2004). The SOM toolbox is presented on the website http://www.cis.hut.fi/projects/somtoolbox/, and a detailed description of how SOM is employed to extract the primary moisture transport pathways from trajectories data can be found in the work of Zhong et al. (2019).

In this work, there are 1346 grid cells in CA at 1° × 1° horizontal resolution, which means more than 18 million trajectories are identified for the daily precipitation during the 37 years studied. This huge number of trajectories makes it difficult for the SOM clustering procedure to process the data in a single clustering analysis. Therefore, we first project the daily trajectories (1346 grid cells per day) onto a 3 × 4 map (i.e., 12 groups) by the SOM training procedure for the 37 years studied. Twelve categories are found to be enough to capture the main features of the daily moisture transport to CA. The aim of this step is to reduce the dimension of the daily moisture transport. Other map sizes, such as 4 × 4 and 3 × 3, were also examined, and only tiny differences were found in the final seasonal clusters. After projecting the original 1346 daily moisture trajectories into 12 groups, the 12 daily cluster-mean trajectories are further classified into 4 groups by projecting them onto a 2 × 2 SOM map for each season. Taking the summer season (June–July–August) as an example, there are 40 804 cluster-mean trajectories in total for the 37 study years, i.e., 12 mean trajectories per day × 92 days per summer × 37 summers. Then, the SOM clustering procedure is applied to the 40 804 trajectories to obtain the final four (2 × 2) groups of moisture pathways. The sensitivity of the seasonal clustering to the map size was also evaluated via projecting the seasonal trajectories onto a 3 × 2 and a 4 × 2 map, i.e., six and eight groups. The results of the SOM training procedure producing more categories of transport pathways show more detailed transport features. However, for climatological research, four groups (2 × 2) are enough to represent the main features of the moisture transport pathways to CA.

The geography of CA varies, as shown in Fig. 1, and high precipitation is mainly concentrated in and around the mountainous region in all the four seasons (Fig. 2). Compared to the CA precipitation from GPCC (Figs. 2e–h), ERAI (Figs. 2a–d) tends to overestimate precipitation in summer and spring. In fact, similar precipitation overestimation by ERAI is also reported for most of the Northern Hemisphere (Dee et al., 2011). In spite of this limitation, ERAI has good consistency with GPCC in the seasonal mean and spatial distribution of the precipitation in CA. Therefore, the ERAI data is acceptable for use in further hydrological analysis.

Figure 2.  The seasonal-mean precipitation (mm d⁻¹) of ERAI (a–d) and GPCC (e–h) in CA during 1979–2015 for spring (a and e), summer (b and f), autumn (c and g), and winter (d and h). The vectors in plots (a–d) are the seasonal-mean total-column moisture flux derived from ERAI.

As shown by the seasonal-mean moisture flux superimposed on Figs. 2a–d, southwest and north CA are generally dominated by moisture flux from the western boundary of CA in all four seasons. Weak moisture influx via the southern boundary is found to flow into southeast CA. In summer, a noticeable anticyclonic flux covers southwest CA, leading to southerly moisture transport dominating in the southwest corner of CA. The features of moisture flux mentioned above suggest that the westerly and southerly related moisture flux together determine the mean and seasonal CA moisture transport. Furthermore, the main balance of the moisture budget Eq. (1) of CA also changes with season. As shown by Table 1, the terms of moisture flux divergence ($ \nabla \cdot \boldsymbol{Q} $), evaporation (E), and precipitation (P) primarily determine the CA moisture budget in winter and summer. The mean $ \nabla \cdot \boldsymbol{Q} $ and (E−P) are negative/positive in winter/summer, suggesting CA is a moisture sink/source in winter/summer. However, in spring and autumn, CA has much weaker moisture convergence/divergence, which leads to the main balance being between evaporation and precipitation.

Figure 3 further demonstrates the distributions of the maximum and minimum precipitation months in CA. Both datasets show consistent seasonal distributions of CA precipitation, with the exception of a narrow band between 45°–50°N (Figs. 3a, b). As shown by Fig. 3, the spatial distributions of the CA maximum and minimum precipitation are quite different, which reflects different seasonal variability. The ERAI and GPCC datasets both show (Figs. 3a, b) that southwest CA experiences its peak precipitation during the cold season (November to March), particularly in March; however, the rest of CA tends to experience its rainy season during the warm season (June to August), mostly in July. This kind of west–east contrast of seasonality is also found in the distribution of the minimum precipitation month (Figs. 3c, d). Southwest CA experiences its dry season during summer to early autumn (July to September), but in late autumn to early spring (November to March), the north and east parts of CA undergo their dry seasons. The orographic effect due to the Tian Shan and Karakoram Mountains is an important factor for the aforementioned differences in the seasonal precipitation (Baldwin and Vecchi, 2016). This out-of-phase relationship of the seasonal precipitation also suggests the moisture transport pathways to the different areas of CA are dominated by totally different circulation systems, which provides motivation to examine the pathways of the moisture transport to CA via the SUDRM in order to explore the source–receptor relationship by solving the recycling ratio in all four seasons.

Figure 3.  Precipitation seasonality manifested in spatial distribution of the maximum and minimum precipitation months in CA. The spatial distributions of the maximum (a, b) and minimum (c, d) precipitation months in CA are derived from ERAI (a, c) and GPCC (b, d) during 1979–2015. The shading represents the month in which maximum or minimum precipitation occurs.

As mentioned in the previous section, the SUDRM is used to quantitatively identify the moisture transport to CA. We first present the distribution of the long-term annual mean of moisture trajectory count and moisture contribution in Fig. 4. As shown in Fig. 4a, the moisture trajectories of CA mainly come from the upstream Eurasian continent, with partial origination from the Mediterranean and the Arabian Peninsula. Some long-distance pathways from places such as the North Atlantic Ocean, the Arctic, and the Indian Ocean can also be discerned from the map of trajectory count and moisture flux. The distributions of transport pathway and moisture flux (Fig. 4a) suggest CA is mainly influenced by the westerlies and the Indian Ocean monsoon. The distribution of the moisture sources (Fig. 4b) displays a similar pattern as the transport pathway (Fig. 4a) but with more local characteristics. In general, the greatest evaporation contribution comes from nearby sources (Keys et al., 2012; Hua et al., 2017). This is true for CA, which has prominent nearby evaporation sources that are vital for its water vapor supply. Apart from the external moisture sources over Eurasia and the Atlantic and Indian Oceans, strong local contribution from the recycling processes inside CA can also be found. It should be noted that the evaporation over the Tibetan Plateau and the mountains to its northwest (including Tian Shan Mountain) strongly contribute to CA precipitation.

Figure 4.  Distributions of the annual-mean trajectory count and moisture contribution for the CA precipitation during 1979–2015 based on the Lagrangian back moisture tracking: (a) trajectory count (d^–1); (b) moisture contribution (mm d^–1). The vectors in (a) and (b) are the annual-mean total-column moisture flux from ERAI.

Figure 5 further demonstrates the seasonal variations of the moisture transport and moisture sources. In summer, the anomalous transport with respect to the yearly mean (Fig. 5a) features a northeast–southwest seesaw pattern, with enhanced local and external moisture contribution over the north and southeast areas and weakened contribution from the southwest external sources (Fig. 5e). Nearly opposite patterns of anomalous transport pathway and moisture contribution are found in winter (Figs. 5c, g). Between summer and winter, transitional transport patterns appear, featuring a northeast–southwest tripolar patten in autumn (Figs. 5b, f) and a weak northeast–southwest dipole in spring (Figs. 5d, h). This seasonal transition also reflects the seasonal variability of the large-scale circulation dominating Eurasia and the surrounding oceans, particularly the variations of the westerly and southwesterly winds.

Figure 5.  Seasonal transition of the moisture trajectory count and moisture contribution for CA precipitation: (a–d) the distributions of the anomalous trajectory count (d^–1) in summer, autumn, winter, and spring with respect to the annual mean; (e–h) the same as (a–d) but for the moisture contribution (mm d^–1).

Figure 6 demonstrates the main groups of the moisture trajectories to CA (Figs. 6a–d) with the corresponding moisture contribution (Figs. 6e–h) based on the 2 × 2 SOM clustering. The four groups generally represent the transport pathways dominated by the northwesterly (SOM1), westerly (SOM2), northerly (SOM3) and southerly (SOM4) winds. The first two pathways are mainly related to the upstream zonal (westerly) wind, but the last two seem to be more associated with meridional (northerly and southerly) wind. These four moisture pathways are identified well in all four seasons and, on average, explain 27.5%, 14.6%, 14.7%, and 43.2% of the total moisture trajectories for SOM1–4, respectively. The southerly pathway accounts for most trajectories and also contributes the most water vapor (48.4%) (Fig. 6h); the Tibetan Plateau seems to be the most important source region for the southerly moisture pathway. For this transport pathway, the southern belt of the trajectory count is also discernable over the Sahel, the Arabian Sea, the Bay of Bengal, and the south edge of the Tibetan Plateau (Figs. 6d, h). The moisture contribution belt along the east coast of Africa indicates the effect of the Somali cross-equatorial flow (Joseph and Sijikumar, 2004). The inland seas, i.e., the Mediterranean, the Caspian Sea, the Black Sea, and the Persian Gulf, also supply considerable water vapor for CA via the southerly pathway. The northwesterly pathway is the second most important transport pathway, contributing 26.4% of the total moisture transported to CA (Fig. 6e). The moisture carried by the northwesterly pathway comes mostly from northeastern Europe and local sources inside CA, particularly over the mountainous area, where the high contribution suggests the effect of the windward slope on precipitation. The other two pathways (Figs. 6f, g), i.e., the westerly and northerly pathways, mainly influence north CA, transporting moisture from terrestrial sources over Europe as well as sources over the North Atlantic (Fig. 6f) and the Barents Sea in the Arctic (Fig. 6g).

Figure 6.  The clusters of the moisture transport pathways for the annual-mean CA precipitation based on the method of the self-organized map (SOM): (a–d) the distributions of the mean trajectory count (d^–1) of the SOM groups; (e–h) the same as (a–d) but for the moisture contribution (mm d^–1). The percentage printed in the bottom left corner of each plot is the explained ratio of each SOM group to the total trajectory count or total moisture contribution to CA. The group SOM1, SOM2, SOM3, and SOM4 approximately correspond to the moisture transport via the northwesterly, westerly, northerly, and southerly transport pathway, respectively.

From the annual-mean transport pathways shown in Fig. 6, the four moisture pathways seem to control different areas in CA. Therefore, we further explore the impact of a specific transport pathway/SOM group on CA by standardizing the trajectory count of this transport pathway over all the grid points in CA. The area composed of the grid points with a trajectory count higher than 0.5 standard deviations is regarded as the impact area of the studied transport way. Taking the northwesterly pathway in summer as an example, we first calculate the trajectory count of the northwesterly moisture trajectories over each grid points of CA during 1979–2015. The resultant map of the northwesterly trajectory count can reflect the degree of the influence of northwesterly transport over different areas in CA. Then, we standardize the northwesterly trajectory counts over all the CA grid points and define those with trajectory count exceeding 0.5 standard deviations as the area under the control of northwesterly moisture transport. Applying this method to different seasons, we can thus obtain the control areas of the four moisture channels in the four seasons (Fig. 7).

Figure 7.  Impact areas of the main transport pathways to CA in summer (a), autumn (b), winter (c), and spring (d). For each season, SOM clustering is employed to extract the four moisture transport pathways based on the daily data. The dominated area of a specific transport pathway is defined by the CA grid points having the number of the daily trajectories belonging to the corresponding SOM cluster be over 0.5 standard deviations compared to all grid points in CA. The overlapping areas are jointly impacted by multiple moisture transport pathways. The blank areas are the regions without significant influence by any of the four transport pathways.

As seen from the impact areas of the four moisture channels shown by Fig. 7, large seasonal variations exist mainly in the west and northeast parts of CA. In summer (Fig. 7a), southwest CA is mainly under the control of northerly transport, and at the same time, a northwesterly transport belt dominates the mountainous area in central south and northeast CA. Surrounded by northwesterly transport on the west and north sides, southerly transport mainly dominates southeast CA. In addition to this, north CA is a multichannel-impacted area that is simultaneously influenced by westerly and northerly transport in most seasons.

Compared with the transport pathway dominating southeast CA, the dominant moisture transport pathways of southwest and north CA are more diverse and more variable with season, as shown by Figs. 7b–d. From summer to the following spring, the influence of northerly transport retreats from southwest to north CA (Figs. 7a–d) and then extends zonally to cover almost all of north CA in spring (Fig. 7d). At the same time, the control area of northwesterly transport tends to shift westward (Fig. 7b) and southwestward (Fig. 7c) in the cold season. In contrast, southerly transport consistently dominates in southeast CA throughout the whole year, with only a slightly westward extension in autumn and spring (Figs. 7b, d). The persistent influences of southerly transport on southeast CA and northwesterly transport on southwest CA together form a west–east contrast of transport pathway in south CA in winter and spring, which fundamentally determines the basic features of precipitation for southwest and southeast CA (Fig. 3). In addition, westerly and northerly moisture transport jointly dominate most areas of north CA in all four seasons, extending eastward from autumn to the following spring (Fig. 7). As shown by the analysis above, the seasonal variations of the transport pathways to CA mainly concentrate in the areas west and north of the mountainous area, which corresponds well to the seasonality of precipitation shown in Fig. 3. The effects of the topography seem to isolate southeast CA from the water supply conveyed by the westerly, northwesterly, and northerly transport pathways.

Closely connected with the seasonality of the impact areas of the transport pathways, the trajectory count and moisture contribution of different transport pathways also show significant seasonal variability. As shown by Fig. 8a, the southerly pathway explains the greatest trajectory count, which peaks (~50%) in the cold season (autumn and winter). The northwesterly pathway, as the second largest contributor of trajectory count, reaches its highest (~40%) moisture contribution in spring. The other two pathways, i.e., the westerly and northerly pathways, both explain about 10%–20% of the trajectory count; the northerly trajectories tend to concentrate in summer and autumn, and the westerly trajectories are more likely to appear in spring and summer. Similar to trajectory count, the southerly transport pathway contributes the most water vapor to CA (over 50%), but in summer instead of the cold season (Fig. 8b). Even in winter, when the southerly pathway reaches its minimum contribution, about 45% of CA moisture can still be attributed to southerly transport. The northwesterly pathway still acts as the second largest contributor of CA moisture, particularly in winter and spring. The westerly pathway reaches its maximum moisture contribution (~21%) in winter. The northerly pathway contributes the least amount of water vapor (Fig. 8b), with a peak (11%–12%) in summer and autumn. In winter and spring, the moisture amount carried by the southerly pathway sharply decreases to 5%–6% (Fig. 8b).

Figure 8.  Seasonal and interannual-to-interdecadal variations of the moisture transport, moisture contribution, and regional recycling ratio of CA. The variations of the explained ratios of the four SOM transport pathways, i.e., northwesterly, westerly, northerly, and southerly, are shown by seasonal (left column) and annual (right column) means. The plots in the left column are the seasonal variability of the trajectory count (a), moisture contribution (b), and regional recycling ratio (c) for the four SOM transport pathways. Similarly, the plots (d–f) in the right column are the annual time series of the three variables in (a–c). The black filled line in (f) is the total regional recycling ratio.

The analysis above shows similar results to previous studies (Yatagai and Yasunari, 1998; Jiang et al., 2020), i.e., CA moisture transport is strongly regulated by midlatitude westerly systems and monsoon circulation systems. The middle-to high-latitude large-scale circulation anomalies, such as the Europe blocking and the subtropical westerly jet, can induce westerly or northwesterly moisture flux to CA. For simplicity, the following analysis incorporates the westerly and southwesterly pathways as the westerly-related pathway to reflect the influence from the circulation systems superimposing on the westerlies. Similarly, the south Asian monsoon carries water vapor to CA via southwesterly or southerly pathway. So, the southerly-related pathway (SOM4) can be approximately regarded as the effect of the south Asian monsoon. From Fig. 6, the westerly-related and southerly-related pathways totally explain 85.3% trajectory count and 90.3% moisture contribution. To a large extent, the relationship between these two kinds of transport pathways determines the CA precipitation variation.

As shown by the correlations between the four SOMs of the moisture transport pathways in Table 2, the trajectory count of the northwesterly/westerly pathway shows significant negative correlation of –0.61/–0.53 with the southerly pathway. The seesaw-like relationship between westerly-related and southerly-related pathways is also reflected in the moisture contribution, as shown by the significantly negative correlation of –0.53/–0.59 between northwesterly/westerly pathway and southerly pathway (Table 2 and Fig. 8e). In winter and spring, the two westerly-related pathways in total contribute more moisture than the southerly pathway (Fig. 8b) and cause the rainy season in southwest CA (Figs. 3a, b). It is therefore concluded that the seasonality of the moisture contribution to CA is mainly caused by the balance between westerly-related transport and southerly-related transport; the former dominates in the cold season, but the latter mainly determines the warm season, because the westerly-related pathway is stronger in midlatitudes in winter and the southwesterly monsoon becomes active in the warm season. On a longer time scale, the trajectory count of the southerly pathway has a lower-than-average period from the early 1980s to mid-1990s but an above-average period from the mid-1990s to late 2000s (Fig. 8d). Similar but opposite interdecadal changes are also found in the trajectory counts of the two westerly-related pathways. However, the moisture contribution of the main transport pathways (Fig. 8e) does not show significant interdecadal variability but does show more interannual features. This discrepancy suggests that the low-frequency variation of the moisture transport to CA is mainly determined by the circulation and not by the moisture supply over the source regions. For the northerly pathway, trajectory count is also significantly negatively correlated (r = –0.47) with the two westerly-related transport pathways (Table 2) but not significantly correlated with moisture contribution. The weak northerly–westerly/northwesterly correlation of the moisture contribution is mainly caused by the insignificant correlation in the cold season, the northerly–northwesterly correlation of –0.04 in spring, and the northerly–westerly correlation of –0.09 in winter. This seasonally varied correlation in moisture contribution reflects the seasonal change of the evaporative condition over the moisture source regions. We further perform the correlation analysis between the trajectory count and moisture contribution of the northerly pathway. It is found that the correlation between trajectory count and moisture contribution reaches a maximum of 0.72 (99% confidence level) in summer and a minimum of 0.33 (95% confidence level) in winter. That is to say, the large number of moisture trajectories from the northern high latitudes does not necessarily correspond to the large amount of moisture transported from the northern area to CA in winter. This suggests a weakened role of wind in moisture transport, or the enhanced impact from the evaporation in the northern area. As shown by Fig. 8b, the northerly pathway explains the year-round lowest moisture contribution in winter. This could be caused by weakened evaporation over the northern high-latitude source regions, such as the West Siberian plain and the Arctic. The low temperature and the snow and sea-ice cover can inhibit evaporation over the above moisture source regions. But in the warm season, the enhanced moisture supply over the high-latitude areas is related to the snow and sea-ice melting therein. Therefore, the seasonality of northerly moisture transport is strongly regulated by the seasonal change of the evaporative condition over the moisture source regions, which is strongly influenced by the local hydrological process.

The SUDRM used in this work can also quantitatively estimate the precipitation recycling, which is the contribution of evapotranspiration to local precipitation (Brubaker et al., 1993). In terms of the annual-mean recycling ratio, the moisture evaporated from CA contributes to roughly 24% of the CA precipitation, as shown by the black-shaded series in Fig. 8f. But this ratio has strong seasonal variation ranging from about 8% in winter to 41% in summer (Fig. 8c). That is to say, the winter water supply in CA is almost completely dominated by advective moisture (Fig. 8c), but the recycling processes associated with the local processes become comparable to the external moisture transport in summer. Between these two seasons, spring has the next largest recycling ratio (31%) compared to summer and is followed by autumn (about 18%).

As seen from Figs. 8c and f, the local evaporation contributing to CA rainfall mainly comes from the area dominated by the northwesterly and southerly transport pathways in summer and spring. As seen from the impact areas of the transport pathways shown in Fig. 7, southeast CA is dominated by the southerly pathway in summer (Fig. 7a), and the western and eastern parts of south CA are jointly dominated by the northwesterly and southerly pathways in spring (Fig. 7d). We can thus make two conclusions about CA recycled precipitation: one is that local evaporation over south CA, particularly the southeastern area, contributes the most recycled precipitation in CA; the other is that changes in the recycling processes in CA are mainly coupled with the variations in northwesterly and southerly moisture transport.

This coupling can also be found in the annual-mean recycling ratios shown by Fig. 8f. There exists strong interdecadal variability of recycling ratio over both the northwesterly and southerly dominating areas, which shapes the variation of the total recycling ratio (the black shaded in Fig. 8f) manifesting as an interdecadal transition around 2005. After this year, the recycling ratio of CA underwent an about 10-year (2005–16) lower-than-average period. But before 2005, the recycled ratio had remained above average for about 20 years, from the early 1980s to early 2000s. This interdecadal variability of the recycling ratio is well coupled with the local evaporation over the areas impacted by northwesterly (r = 0.66 over 99% confidence level) and southerly (r = 0.47 over 99% confidence level) transport, as shown by the rightmost column in Table 2. This feature prompts further exploration on the moisture transport feature associated with the interdecadal variability of the recycling ratio in CA.

Figure 9 shows the regressions of trajectory count and moisture contribution on the annual-mean total recycling ratio shown in Fig. 8f. Positive/negative regression coefficients in Fig. 9 mean enhanced/weakened trajectory count or moisture contribution corresponding to increased/decreased recycling ratio. As shown by Figs. 9a and b, the increased recycling ratio in CA corresponds to a meridional tripolar pattern characterized by enhanced trajectory count and moisture contribution over the middle latitudes, extending from the midlatitude Atlantic to CA, and prohibited trajectory count and moisture contribution in the high and low latitudes, including north Eurasia and the Arctic (the Kara sea and the Laptev Sea) on the north side and the India Peninsula, the Arabian Peninsula, and the north Indian Ocean on the south side. Since 2005, more than a decade with anomalously low recycling ratio has been coupled with the contemporary weakened moisture transport from the midlatitude Atlantic and terrestrial sources over Europe and the reinforced moisture contribution from the north and south sides of CA. The regression patterns shown in Figs. 9a and b are very similar to the patterns of differences between the moisture transport before and after 2005 (figures not shown). That is to say, the variation of the moisture transport to CA mainly reflects the interdecadal variability of the recycling ratio shown by Fig. 8f.

Figure 9.  The regression coefficients of the trajectory count (left column) and moisture contribution (right column) on the annual-mean total regional recycling ratio shown in Fig. 8f. The regression maps of the total trajectory count and moisture contribution are shown in (a) and (b), respectively. Similar regressions are also shown for the SOM groups of northwesterly (c, d), westerly (e, f), northerly (g, h), and southerly (i, j). The stippled area in each plot is that passing the 95% significant test.

Furthermore, the trajectory count and moisture contribution of the four transport pathways are separately regressed on the variation of the annual-mean recycling ratio to show the relationships between the different-pathway moisture transport and the recycling precipitation. It is found that the northwesterly and southerly pathways are the two most closely connected transport modes with the recycling ratio change in CA (Figs. 9c, d and Figs. 9i, j). This is also shown by the significant correlations of these two pathways with the CA recycling ratio in Table 2. From Figs. 9d and j, the externally sourced moisture contribution from the northwesterly/southerly pathway shows in-phase/out-of-phase variability with the variability of the local moisture contribution in CA. According to these relationships, the lower-than-average recycling ratio since 2005 (Fig. 8f) should be accompanied by anomalously low moisture contribution over Europe and high moisture contribution over the Indian subcontinent and the Indian Ocean, of which the former could be attributable to northwesterly transport (Fig. 9d) and the latter attributable to southerly transport (Fig. 9j). At the same time, corresponding to the low recycling ratio in CA, the westerly transport pathway tends to carry less water vapor to CA from the midlatitude North Atlantic (Fig. 9f) and there is increased moisture contribution transported from north Eurasia and the Arctic via the northerly pathway (Fig. 9h). In short, the lower-than-average recycling ratio of CA in recent decades is coupled with weak westerly-related (westerly and northwesterly) moisture transport and enhanced meridional (northerly and southerly) moisture transport.

This study focuses on the seasonal and interdecadal variability of the precipitation and moisture transport of Central Asia (CA), which includes Northwest China and five countries, i.e., Kazakhstan, Uzbekistan, Turkmenistan, Kyrgyzstan, and Tajikistan. By using the Lagrangian moisture tracking method, we try to figure out the quantitative roles of the main transport pathways that are responsible for the precipitation seasonality and the interdecadal transition of the recycling ratio based on reanalysis data and observation data for 1979–2016.

The seasonality of the CA precipitation shows remarkable spatial differences characterized by the different rainy seasons in different areas of CA. Cold-season precipitation contributes most of the yearly precipitation over southwest CA, but warm-season precipitation dominates the yearly precipitation over southeast and north CA. This spatial difference of the seasonal variability is accompanied by the inter-seasonal transition of the moisture transport. By applying the self-organizing map (SOM) clustering method, four pathways undertaking the moisture transport to CA are identified: the northwesterly, westerly, northerly, and southerly transport pathways explain 28%, 15%, 15%, and 43% of the moisture trajectories to CA, respectively. These four transport pathways dominate different areas of CA and have strong seasonal variability, which leads to diverse precipitation seasonality in north, southeast, and southwest CA.

The spatial distribution of the precipitation seasonality of CA can be explained well by the variations of the impact areas of the main pathways of the moisture transport. The northwesterly and westerly pathways that span from the middle to high-latitude Atlantic and northwest Eurasia to CA mainly dominate southwest CA. Of these two westerly-related branches of moisture transport, the westerly pathway tends to be dominant for summer and autumn precipitation in southwest CA, but the northwesterly pathway plays a more dominant role in winter and spring. For precipitation over north CA, the transport pathways exerting important impacts are more diverse. The northerly pathway shows impact on north CA in all four seasons. The water vapor over west Siberia and the Arctic can be transported to CA via the northerly pathway year-round. The two branches of the westerly-related (northwesterly and westerly) pathways also greatly contribute to the seasonal variation of precipitation in north CA. Different from the two CA subregions mentioned above, southeast CA is isolated from westerly and northerly-related moisture transport by the mountainous areas located between Northwestern China and the five CA countries. Southeast CA is thus dominated by southerly moisture transport from the Arabian Peninsula, the Indian Peninsula, and the Indian Ocean year-round. To a great extent, the remarkable spatial difference of the precipitation seasonality in CA, particularly the opposite seasonal variability of precipitation between southeast and southwest CA, could be attributed to the seasonality of the westerlies and the Indian monsoon, as shown by the moisture contributions via the westerly-related and southerly-related pathways, which are negatively correlated and account for about 42% and 48% of the moisture over CA, respectively.

The advected moisture through the transport pathways mentioned above is also coupled with the contribution due to the local evaporation associated with the recycling processes. Like the precipitation variability, the regional recycling ratio of CA also shows strong seasonal variation, with the highest level in summer and the lowest level in winter. This seasonal variability of recycling ratio is mainly coupled with the moisture transport via the northwesterly and southerly pathways. Moreover, this coupling relationship projects an interdecadal transition around 2005 on the annual-mean total recycling ratio over the whole CA area. This interdecadal variation of recycling ratio corresponds to a tripolar meridional moisture contribution pattern, in which the prohibited moisture contribution over the middle latitudes (Atlantic, Europe, and CA itself) is in between the enhanced contributions over the northern sources (west Siberia and the Kara Sea) and the southern sources (India and the Indian Ocean). It should also be noted that, aside from the influences from northwesterly and southwesterly transport, northerly transport plays an important role as well in coupling the high-latitude moisture variation with the change of the local recycling ratio in CA.

The Indian summer monsoon has been in a period of recovery from a period of decline since the early 2000s with modulation by the interdecadal Pacific Oscillation (IPO) (Huang et al., 2020). This interdecadal transition of the monsoon generally coincides with the transition of the regional recycling ratio of CA shown in this study. After the early 2000s, the coupling between the decreased recycling ratio and the increased moisture contribution from the southerly transport pathway well reflects the enhanced Indian summer monsoon. So, the modulation of the IPO and other related internal climate variability (Zhong et al., 2021) on CA hydrological processes is a natural extension of this study. In addition, the enhancement of the moisture contribution of the northerly pathway is also found to be associated with the decrease of the recycling ratio in CA since the early 2000s. This corresponds to the recent amplified warming over the Arctic (Serreze et al., 2009; Cohen et al., 2014). Considering that melting of snow and ice cover could enhance evaporation in the high-latitude area, it is reasonable to presume that Arctic warming should have an impact on the northerly moisture flux to CA and, therefore, the recycling processes in CA. It is also meaningful to quantitatively study the moisture pathways and moisture contributions in the future climate and their implications on forced and internal variability of the climate system by employing state-of-the-art climate models. The issues mentioned above deserve further exploration in order to identify the physical connections between the CA hydrological variations and both tropical and polar climate change.

Acknowledgements. This research was jointly supported by the Strategic Priority Research Program of Chinese Academy of Sciences under Grant No. XDA20020201 and the National Natural Science Foundation of China (NSFC) under Grant Nos. 41975099, U2006210, and 41475072.