1. Introduction
Xinjiang is located in the northwest region of China, which is in the interior of Eurasia and north of the Tibetan Plateau (Fig. 1). It is difficult for moisture originating from oceans to reach this region directly, making this a typical Asian arid–semiarid region (Chen et al. 2013). Since the mid-twentieth century, the annual precipitation in Xinjiang has exhibited an upward trend, mainly due to increasing heavy precipitation in boreal summer (Ren et al. 2005; Shi et al. 2007; Han et al. 2016; Peng and Zhou 2017). Heavy precipitation occurs every year in Xinjiang and has become enhanced in recent years (Yang et al. 2011, 2015). In addition, the heavy precipitation in Xinjiang is concentrated in the mountainous areas (Fig. 2), where it often triggers floods, landslides, and mudslides, resulting in huge losses of lives and property as well as economic losses. Therefore, it is of great significance to investigate the mechanism of heavy precipitation in Xinjiang, not only for the reasons of understanding long-term change in Xinjiang precipitation but also for disaster prevention and mitigation.
The geographical area (red line) and topography (shading; km) of Xinjiang.
Citation: Journal of Hydrometeorology 22, 8; 10.1175/JHM-D-20-0236.1
The distribution of climatological mean (1979–2018) (a) annual precipitation (mm), (b) wet season precipitation (April–September; mm), and (c) heavy precipitation in the wet season (mm) from observational CN05.1 data.
Citation: Journal of Hydrometeorology 22, 8; 10.1175/JHM-D-20-0236.1
Sufficient water vapor supply is an essential precondition of precipitation, particularly for heavy precipitation in Xinjiang (Newell et al. 1992; Dai et al. 2006; Li et al. 2008; Drumond et al. 2011; Huang et al. 2015; Guan et al. 2019). Previous qualitative analyses of the moisture sources for heavy precipitation in Xinjiang were completed using the Euler method and suggested that the Arabian Sea and the Bay of Bengal provide the main water vapor supply (Yang et al. 2011; Zhang et al. 2016, 2018). Lagrangian simulation has been widely applied to trace the precipitation moisture sources directly from related source regions and to provide more precise details about the humidity variation of air parcels in the transport process (e.g., Stohl and James 2004, 2005; Sodemann et al. 2008; Viste and Sorteberg 2013). Specifically, a few studies on the moisture source of heavy precipitation in Xinjiang were conducted using the Lagrangian method. Yao et al. (2018) examined the water vapor paths of a rainstorm event, and it is considered that there are three main paths for water vapor transport from the Arabian Sea, Baltic Sea, and other places to Xinjiang. Huang et al. (2017) used the HYSPLIT model to examine water vapor sources of heavy precipitation in Xinjiang from 1951 to 2014 and determined that most of the water vapor came from the North Atlantic, the Arctic Ocean, the Eurasian Continent, and the Indian Ocean. Zhou et al. (2019) experimented with the Flexible Particle dispersion model (FLEXPART) and concluded that water vapor sources of extreme precipitation in Xinjiang are mainly from Xinjiang and central Asia, and the moisture contribution ratios of various source areas vary with the precipitation areas.
The above studies on moisture sources of heavy precipitation events in Xinjiang are based on the precipitation itself. However, the quantity and location of heavy precipitation are also closely associated with meteorological patterns; for example, the central Asia trough and vortex have an evident influence on heavy precipitation in Xinjiang via their position and strength (Yang et al. 2011, 2015). Then, for the long-term period, does the spatial distribution of heavy precipitation in Xinjiang have the typical meteorological patterns? If so, what are the differences of moisture source distribution and transport process of heavy precipitation in Xinjiang under different meteorological patterns? Therefore, it is important to classify heavy precipitation days via the meteorological patterns in Xinjiang and understand whether the moisture source varies when heavy precipitation events are characterized by different meteorological patterns. This study attempts to answer these questions by first identifying heavy precipitation days based on daily reanalysis and observation data in the wet seasons (April–September) of 1979–2018. Then, a neural network technique, called self-organizing maps (SOM; Kohonen 1998), is used to obtain the main meteorological patterns of heavy precipitation days in Xinjiang based on the accompanying atmospheric circulations. In the SOM method, each meteorological pattern is identifiable as representing typical atmospheric characteristics of precipitation, leading to an easier understanding of the physical meaning. Additionally, SOM has been used in previous searches for moisture sources (Cavazos 1999; Skific et al. 2009). Finally, the differences of heavy precipitation distribution and associated moisture sources among SOM patterns in Xinjiang are examined through simulation using FLEXPART.
The remainder of this paper is organized as follows: the data, method, and model are introduced in section 2; section 3 provides the SOM clustering results of meteorological patterns for heavy precipitation days and presents the precipitation distribution, moisture transport process, and moisture source distribution characteristics in various SOM patterns; and the discussion and conclusion are provided in section 4.
2. Data, method, and model
a. Data
The dataset used in the study spans from 1979 to 2018. To select heavy precipitation days, the daily observation and reanalysis precipitation data have been used. The former is taken from the CN05.1 dataset with a horizontal grid of 0.25° × 0.25° based on 2416 stations in China (Wu and Gao 2013), and the latter is obtained from the Climate Forecast System Reanalysis (CFSR; Saha et al. 2010; https://rda.ucar.edu/datasets/ds093.0/) and Climate Forecast System version 2 (CFSv2, the expansion of CFSR after December 2010; Saha et al. 2014; https://rda.ucar.edu/datasets/ds094.0/) from the National Centers for Environmental Prediction (NCEP). The 6-hourly atmospheric variables derived from the CFSR products are employed for driving FLEXPART and include geopotential height, temperature, specific humidity, relative humidity, and three-dimensional wind at 42 levels from the surface to 1 hPa on a horizontal resolution of 0.5° × 0.5°. To confirm the robustness of the results, the 6-hourly atmospheric variables from the ECMWF interim reanalysis (ERA-Interim; Dee et al. 2011; https://www.ecmwf.int/en/forecasts/dataset/ecmwf-reanalysis-interim) with a horizontal resolution of 1° × 1° grid are used for checking the results of analysis from CFSR.
b. Selection of heavy precipitation days
The heavy precipitation events in Xinjiang are determined by the percentile threshold of daily precipitation rather than a given value because CFSR precipitation is higher than that observed over Xinjiang (Hu et al. 2016). The heavy precipitation in Xinjiang occurs mainly from April to September (Fig. 2). This period is regarded as the wet season in Xinjiang, from which the heavy precipitation days are selected.
Following previous studies (e.g., Bretherton et al. 2004; Song et al. 2015), the heavy precipitation days are jointly identified in order by 1) the days with regional mean daily precipitation in Xinjiang being greater than 0.1 mm day−1 in the wet season; 2) the days with regional mean daily precipitation in Xinjiang ranking the top 25% of its values; and 3) the aforementioned two preconditions holding in both CFSR and CN05.1. A total of 893 heavy precipitation days were selected in the wet seasons from 1979 to 2018. The selection method ensures that the circulation of CFSR matches well to the observed heavy precipitation. In addition, we checked the area of precipitation for each heavy precipitation event, and found that that the main precipitation areas of the 893 days were limited in 54% (less than 50% in 888 days) of the whole Xinjiang. In other words, those heavy precipitation events are not caused by a large area of light rain in all selected days. Figure 2c shows that heavy precipitation in Xinjiang mainly occurs in the Tianshan Mountains, followed by the Junggar Basin and then the Kunlun Mountains. In terms of the climatological mean, heavy precipitation accounts for 31% of the precipitation in the wet season.
c. SOM
The SOM method is based on an artificial neural network developed by Kohonen (1998) and is essentially a clustering method; it has been applied in many meteorological studies (e.g., Sheridan and Lee 2011; Lennard and Hegerl 2015; Loikith et al. 2017; Mewes and Jacobi 2019). Compared with other clustering methods such as k-means, SOM is better able to describe the topological relationships among classifications. Specifically, unlike other clustering methods, SOM results can reflect the distance between clusters (such as the Euler distance between clusters). SOM is also unaffected by the linear assumption (Reusch et al. 2005). More details of the SOM method were supplied in Kohonen (1998).
A key parameter that must be set in advance for SOM is the number of nodes or clusters; this is usually set as a two-dimensional grid rather than a number since the SOM classification results can characterize the topological structure among the nodes. The number of nodes is achieved by balancing the considerations for interpreting the results and the representative major patterns in the dataset. A node number that is too high leads to multiple nodes with the same climatological characteristics, while one that is too low leads to different characteristics being combined into the same node, both of which are troubling and introduce deviation to the analysis. Here, the number of nodes was set to 2 × 2 (4 nodes), after experimenting with other numbers of nodes (2 × 2, 2 × 3, 3 × 3, and 3 × 4). This number of nodes was chosen because, regardless of the number of nodes permitted, four nodes covered most of the heavy precipitation days.
Similar to the approach of Lennard and Hegerl (2015) and Loikith et al. (2017), SOM employed multivariate inputs, including anomalies of daily sea level pressure (MSL) and 500-hPa (Z500) and 200-hPa (Z200) geopotential heights of heavy precipitation days over 30°–60°N, 50°–110°E. Because of this, SOM can simultaneously extract atmospheric circulation characteristics in the low, middle, and high atmosphere.
d. Model description
FLEXPART version 10.03 (Stohl et al. 2005; https://www.flexpart.en/) was used in this study and is a Lagrangian diagnostic model that converts Euler data into Lagrangian data through a series of mathematical operations. To precisely describe the atmospheric motion from 1979 to 2018, the global atmosphere was divided into five million particles by means of equal mass partition (no gaps between particles). The 6-hourly CFSR dataset was used as input data for FLEXPART. This dataset has been applied to FLEXPART to investigate water vapor sources in North China, the eastern Tibetan Plateau (Sun and Wang 2014) and the Chinese Loess Plateau (Hu et al. 2018a,b). The FLEXPART output comprises the three-dimensional position, air density, mass, temperature, and specific humidity of each parcel at 6-h intervals.
e. Diagnostic method of moisture source
It was necessary to determine the parcel trajectory data from the model output in order to identify the moisture source of Xinjiang precipitation, which was accomplished by using the moisture source attribution method proposed by Sodemann et al. (2008). That method fully considered the moisture contribution quantity to the target area precipitation caused by the moisture change of air particles in the transport process. For example, when the specific humidity of a particle increases during transport, all earlier contributions rate to precipitation in the target area are diluted. Conversely, when specific humidity is reduced, all earlier contribution rates are unchanged, but the contribution amount to precipitation in the target area is decreased. Therefore, the method of Sodemann et al. (2008) can directly identify the contribution amount of each moisture source to precipitation in the target area. It is noteworthy that the contribution of water vapor sources determined by this method explain simulated precipitation rather than observed precipitation directly. Simulated precipitation is defined as the specific humidity reduction of all particles in the last 6 h. Further details regarding this method can be found in Sodemann et al. (2008).
There are two main differences between the method used in this study and that of Sodemann et al. (2008). First, to better reflect the observed precipitation, only particles over the main precipitation areas are selected, instead of the particles over the entirety of Xinjiang on each heavy precipitation day. In other words, for a certain precipitation day, the main precipitation areas are those where the precipitation accounts for 80% of the total precipitation. Second, Sodemann and Zubler (2010) indicated that the moisture origins for precipitation in the target area are unlikely to be fundamentally different above and below the boundary layer, which is also true for the moisture source of Xinjiang heavy precipitation when we performed a preliminary investigation on the issue in this study. According to recent studies (Sun and Wang 2014; Hu et al. 2018a,b; Huang et al. 2018), this study does not distinguish the moisture absorption from within or outside of the atmospheric boundary layer either.
3. Results
a. SOM results
Composite circulation patterns of heavy precipitation events from the CFSR are shown in Fig. 3. When heavy precipitation events occur, negative MSL anomalies arise over the Xinjiang–Tibetan Plateau regions, and the positive centers are located on the northwest of Xinjiang. Z500 and Z200 anomalies show a zonal tripole pattern over Eurasia, where the positive–negative–positive anomaly centers are located at the ridge–trough–ridge of their climatological geopotential height. Based on the SOM patterns of anomalies in MSL, Z500, and Z200 during heavy precipitation days, there are four nodes of typical SOM patterns, including N1, N2, N3, and N4. All heavy precipitation days are assigned to the corresponding nodes. Then, the MSL, Z500, and Z200 and their anomalies of heavy precipitation days in each node are composited in Figs. 4–6. The heavy precipitation days in N1 (305 days), N2 (156 days), N3 (168 days), and N4 (264 days) account for 34%, 17%, 19%, and 30% of the total heavy precipitation days (893 days), respectively. There are apparent differences in the patterns among four nodes (Figs. 4–6). N1 has a dipole pattern of MSL anomalies with a northwest–southeast tilt over central and northern Asia, which is accompanied by the zonal pattern of Z500 and Z200, with a positive–negative–positive center over Eurasia. N4 has two small negative centers of MSL anomalies in northwest Asia and southeast Xinjiang, and a large extent of positive values over the rest of Asia; in middle-to-high levels, the positive anomalies of Z500 and Z200 are zonally located to the northeast and west of Xinjiang and a weak negative center is located in the west of Xinjiang. Patterns convert from N1 to N4 moving diagonally across the SOM. In general, the four SOM nodes can distinguish different typical characteristics of circulation patterns of heavy precipitation days in Xinjiang.
Composite sea level pressure (MSL, shading), 500- and 200-hPa geopotential heights (Z500 and Z200; shading), and their anomalies (contours) for heavy precipitation days (a total of 893 days) from April to September from the CFSR dataset.
Citation: Journal of Hydrometeorology 22, 8; 10.1175/JHM-D-20-0236.1
Composite SOM patterns of heavy precipitation days in sea level pressure (contours; hPa) and anomalies (shading; hPa) for each of the four nodes from the CFSR dataset. The number and percentage of days assigned to each node from the 893 days are shown in top right of each panel.
Citation: Journal of Hydrometeorology 22, 8; 10.1175/JHM-D-20-0236.1
As in Fig. 4, but for 500-hPa geopotential height (contours; gpm) and the anomalies (shading; gpm).
Citation: Journal of Hydrometeorology 22, 8; 10.1175/JHM-D-20-0236.1
As in Fig. 4, but for 200-hPa geopotential height (contours; gpm) and the anomalies (shading; gpm).
Citation: Journal of Hydrometeorology 22, 8; 10.1175/JHM-D-20-0236.1
Previous studies have shown that the low trough and vortex over central Asia have important impacts on the occurrence of heavy precipitation in Xinjiang (Yang et al. 2011, 2015). The location of the low trough and vortex in central Asia coincides with that of the negative anomaly centers in Fig. 5. Then, it can be inferred that the evolution of the central Asian trough and vortex is closely related to those Z500 negative anomaly centers. The center position of negative Z500 and Z200 anomalies is consistent with the average position of heavy precipitation events in N1 and N4 and moves northward and southward in N2 and N3, respectively. The negative center strength is stronger for N1 and N2, weaker for N4, and close to the average for N3.
b. Comparison of precipitation between Lagrangian estimation and instrumental data
To establish the credibility of the adopted Lagrangian method, the composited maps of four nodes in observed (CN05.1) and estimated precipitation (calculated by the method of Sodemann et al. 2008) are shown (Fig. 7) to investigate the moisture sources of heavy precipitation days. Among four nodes, the Tianshan Mountains are the main concentration area for observed heavy precipitation. In addition, in N1 and N3, the heavy precipitation over the Tarim Basin and Kunlun Mountains is higher than that in N2 and N4. The above analysis shows that there is a close relationship between the spatial distribution of heavy precipitation and circulation patterns. This confirms that the SOM clustering can reflect the typical circulation related to heavy precipitation in Xinjiang.
Composite each node precipitation (mm day−1) for heavy precipitation days from (left) observational CN05.1 data and (right) Lagrangian estimation. Each node number corresponds one-to-one with the node number in Fig. 4.
Citation: Journal of Hydrometeorology 22, 8; 10.1175/JHM-D-20-0236.1
The heavy precipitation in Xinjiang simulated by Lagrangian estimation reproduces the observed characteristics of the node precipitation. This indicates similar distribution and magnitude with that of observation in large-scale areas, although there is still a small difference between them (Fig. 7). For example, there are two observational centers of heavy precipitation in Tianshan, which cannot be distinguished in the simulated results; simulated precipitation is slightly less than observed precipitation in the Kunlun Mountains and the Junggar and Tarim Basins. In addition, the overestimation of precipitation by Lagrangian estimation seen in earlier studies (Stohl and James 2005; Sodemann et al. 2008; Sodemann and Zubler 2010; Baker et al. 2015) does not appear in this study. This is because only the particles over the main precipitation area of the observed precipitation are selected, instead of all the particles over Xinjiang. Consequently, the similarity between the Lagrangian estimation and observation precipitation indicates that the former is reasonable to represent 4-node precipitation in Xinjiang and provides a reliable basis for the following analysis.
c. The proportion and distribution of moisture with backtracking time
The influence of backtracking time varies with the method of moisture source diagnosis. For the Stohl and James (2004, 2005) method, backtracking time affects the amount of simulated precipitation (Nieto and Gimeno 2019). For the improved method of Sodemann et al. (2008), which is applied in this study, backtracking time affects the proportion of moisture, rather than the amount of simulated precipitation; this can be traced to better explain precipitation in the target region. Figure 8 clearly shows that after 6 h of tracking time, the moisture contribution ratio decreases exponentially in the four nodes. Correspondingly, the cumulative contribution rates increase rapidly at first and then increase more slowly as the backtracking time grows. The percentage of the moisture reaches 90% in each node within 6 days and approximately 97% when the backtracking time is set to 10 days. That is, when the Sodemann et al. (2008) method is used and the backtracking time is increased to a certain length (here, 10 days), the proportion of moisture and the moisture contribution ratio among four nodes show little difference.
Composites of the contribution rates (bars) of moisture that can be traced with different backtracking times for each node, which are shown together with their cumulative contribution rates (curves). Pentagrams and dashed lines denote the range of one standard deviation of the contribution rates and the cumulative contribution rates, respectively.
Citation: Journal of Hydrometeorology 22, 8; 10.1175/JHM-D-20-0236.1
d. Process of moisture transport
To clearly illustrate the particle backward trajectory, trajectories with similar position and specific humidity are represented by a cluster (Moody and Galloway 1988; Dorling et al. 1992). Accordingly, a single cluster represents a large number of trajectories. Only particles at altitudes under 9000 m are used because the moisture is concentrated below 300 hPa (the mean altitude at 300 hPa is approximately 9000 m). In this study, as common practice from previous studies, the number of clusters was set to 500 (e.g., Hu et al. 2018a). Note that we have selected other numbers of clusters several times and found that the main conclusion holds despite the changing numbers.
The precipitation-related particles are mostly from the regions west of Xinjiang including central Asia, the western portion of north Asia, Europe, the northwest Atlantic, and North Africa (Fig. 9). Those particles are mainly transported along the western moisture path. Only a small number of particles are transported through the northwest and southwest paths, and even fewer particles enter Xinjiang from the east. The most obvious differences in particle trajectories among nodes is shown in N2, in which particles from Northwest Asia turn southward at the west of Balkhash Lake and converge into the westerly path into Xinjiang (highlighted with red rectangles in Fig. 9). In comparison, the particles of other nodes have no apparent steering process. In addition, the particles get wet only nearby Xinjiang, implying that the moisture source of heavy precipitation in Xinjiang is unlikely to be far away from Xinjiang.
Clustered mean trajectories of particles (number: 500) over 10 days before reaching Xinjiang for each node. Dots represent the start and end points of each trajectory. There is an obvious difference of trajectories in the red box (in the left panels).
Citation: Journal of Hydrometeorology 22, 8; 10.1175/JHM-D-20-0236.1
To show the process of particle aggregation more clearly, we define the particle aggregation degree as the number of particles in each 0.5° × 0.5° grid divided by the total number of particles in a node at every moment. Figure 10 shows that precipitation-related particles gather in central Asia and Xinjiang 120 h before the heavy precipitation day. Those particles further gather around Xinjiang at 24 h before the heavy precipitation day. Meanwhile, there are apparent differences among the nodes in the particle aggregation state. For example, at −72 h, the particles in N1 and N2 are concentrated in the large area from Xinjiang to the Caspian Sea, while the particles in N3 and N4 are concentrated in southern Xinjiang and central Asia nearing Xinjiang. At −24 h, the aggregation degrees of particles in N1 and N3 over southern Xinjiang are higher than those in N2 and N4. This is especially true in the region to the south of Balkhash Lake, where the maximum particle aggregation exists in N2 and the minimum appears in N3. The aggregation degree of particles in northern Xinjiang at −24 h increases in order from N1 to N4.
Aggregation degree of particles (10−4) at different backtracking times for each node (e.g., “−120h” means the 120th hour before precipitation). The particle aggregation degree is defined as the number of particles in each 0.5° × 0.5° grid divided by the total number of particles in a node. Note that the color bar is logarithmic, following Bohlinger et al. (2017) and Huang et al. (2018).
Citation: Journal of Hydrometeorology 22, 8; 10.1175/JHM-D-20-0236.1
The difference in the earlier particle aggregation process among nodes cannot be simply explained by the previous atmospheric circulation in Figs. 5 and 6, which are classified by atmospheric circulation only on the heavy precipitation days in Xinjiang. However, previous research (Yang et al. 2015; Yang and Zhang 2017) points out that the precipitation in Xinjiang is related to the location and strength of the central Asia low pressure system (vortex and trough). It is clear that these four SOM patterns have a good correspondence with the anomalies of the central Asia low pressure system in Figs. 5 and 6. In detail, the contrast between N1 and N4 reflects the strength difference of the low pressure system. The position differences of low pressure system are visible between N2 and N3 (Figs. 5 and 6). When the location of the low pressure system is more northward in N2, more precipitable particles are concentrated in northern Xinjiang and northern central Asia (Fig. 10). On the contrary, particles are more concentrated in southern Xinjiang in N3. Correspondingly, the heavy precipitation in Xinjiang also moves in the meridional direction. Generally, the central Asia low pressure system affects the location of heavy precipitation and associated moisture source in Xinjiang by modulating the location of precipitable particle aggregation.
e. Moisture source patterns of Xinjiang precipitation
According to the method of Sodemann et al. (2008), the distribution of moisture sources in each node of heavy precipitation in Xinjiang is displayed in Fig. 11. The results show that the main moisture (green contour) of heavy precipitation in Xinjiang comes from Xinjiang and central Asia. The large value centers of the moisture sources vary with the nodes. In N1, the main moisture source extends westward to the Aral Sea, with the large value centers in the northwest portion of the Tarim Basin and to the northwest of the Tianshan Mountains. In N2, the main moisture source extends westward to the Caspian Sea, with the maximum in the northwest of the Tianshan Mountains. Similar to N1, the moisture source center of N3 is located in the northwest portion of Tarim Basin, with a smaller westward extension range. The large value centers of N4 are similar to those of N2, except that the westward extension range of N4 is smaller.
Composites of the moisture uptake (mm day−1) of heavy precipitation days for each node. The 80th percentiles of the mass contributed by moisture sources are shown in contours. Note that the color bar is logarithmic, following Bohlinger et al. (2017) and Huang et al. (2018).
Citation: Journal of Hydrometeorology 22, 8; 10.1175/JHM-D-20-0236.1
The distribution of moisture sources is closely connected to the location of particle aggregation before precipitation in the four nodes (Figs. 10 and 11). Specifically, the moisture source centers are related to the state of particle aggregation at 24 h before precipitation. The range of main moisture sources and the particle aggregation position are well matched from −120 to −72 h. For heavy precipitation in Xinjiang, the above results indicate that, during short-term (within 24 h before precipitation) accumulation, the location of the precipitation particles is more likely to be the main source of moisture.
To quantify the contribution rate of moisture from specific regions to heavy precipitation, Xinjiang and central Asia, as the most important source regions, are further divided into four subregions, namely, northern Xinjiang, southern Xinjiang, north-central Asia, and south-central Asia (Fig. 12a), and are divided according to the moisture source distribution characteristics. Based on the moisture proportion contributed by each subregion, the most important moisture source subregion is southern Xinjiang, followed by north-central Asia, northern Xinjiang, and south-central Asia; these moisture source subregions contribute 29%–37%, 19%–27%, 14%–19%, and 13%–16% of the total moisture in Xinjiang heavy precipitation, respectively (Fig. 12b). The moisture contribution rate of each source area differs among the four nodes. The most conspicuous differences between N2 and N3 are exhibited in southern Xinjiang and north-central Asia. As mentioned above, these differences can be reasonably explained by the particle aggregation degree. In N2, the aggregation degree of north-central Asia is the highest at −24 h, corresponding to the largest contribution rates over north-central Asia. In contrast, that of southern Xinjiang is the lowest of the other three nodes and leads to the smallest contribution rates over southern Xinjiang. The particle aggregation degree can also explain the different moisture contribution rates of the remaining nodes for the same subregion.
(a) Moisture source subregions: northern Xinjiang (NX), southern Xinjiang (SX), north-central Asia (NCA), and south-central Asia (SCA). (b) Composite of the relative contribution rates (%) of each subregion for each node.
Citation: Journal of Hydrometeorology 22, 8; 10.1175/JHM-D-20-0236.1
4. Conclusions and discussion
A total of 893 heavy precipitation days in Xinjiang are selected based on CFSR and CN05.1 data from the wet seasons (April–September) of 1979–2018. The SOM is used to classify the selected heavy precipitation days through accompanying atmospheric circulation. Moisture source regions and moisture contributions to the heavy precipitation in Xinjiang are then systematically investigated using the Lagrangian model, FLEXPART, which simulates the motion of air parcels in the global atmosphere. The effects of different meteorological patterns on the spatial distribution of heavy precipitation and associated moisture sources and moisture transport processes in Xinjiang are explored. The results are as follows:
The atmospheric circulations of heavy precipitation events in Xinjiang during the wet season are divided into four SOM nodes, namely, N1, N2, N3, and N4. Composite analyses of the atmospheric conditions of four SOM nodes display a good corresponding relationship with the strength and position of negative anomaly centers of middle-to-high geopotential heights, which reflect the anomalous evolution of the central Asia vortex and trough.
The distributions of heavy precipitation vary with SOM nodes. When the center of a negative anomaly in middle-to-high geopotential heights is strong (N1) or southward (N3), the heavy precipitation is concentrated in the Tianshan Mountains and the Kunlun Mountains. When the negative anomaly center is weak (N4) or northward (N2), the precipitation is dispersed in areas around the Tianshan Mountains.
Four regions contribute most of the heavy precipitation moisture in Xinjiang, in descending order: southern Xinjiang (29%–37%), north-central Asia (19%–27%), northern Xinjiang (14%–19%), and south-central Asia (13%–16%). Additionally, the contribution rate of each source regions varies among nodes. Furthermore, the changes in contribution rates of moisture sources for heavy precipitation are well explained by the aggregation processes of precipitation-related particles before heavy precipitation days.
Finally, we would like to mention that this study mainly focuses on the direct moisture source of heavy precipitation in Xinjiang. It is worth further investigating how the indirect moisture source is distributed and how much water vapor is delivered from distant source areas by indirect means, as similar research has been carried out for eastern China (Fremme and Sodemann 2019).
Acknowledgments
We acknowledge Andreas Stohl and other developers for providing the FLEXPART. We also thank the two anonymous reviewers for their helpful comments. This research was supported by the National Natural Science Foundation of China (41991284) and the Second Tibetan Plateau Scientific Expedition and Research Program (2019QZKK0101).
REFERENCES
Baker, A., H. Sodemann, J. Baldini, S. Breitenbach, K. Johnson, J. Van Hunen, and P. Zhang, 2015: Seasonality of westerly moisture transport in the East Asian summer monsoon and its implications for interpreting precipitation δ18O. J. Geophys. Res. Atmos., 120, 5850–5862, https://doi.org/10.1002/2014JD022919.
Bohlinger, P., A. Sorteberg, and H. Sodemann, 2017: Synoptic conditions and moisture sources actuating extreme precipitation in Nepal. J. Geophys. Res. Atmos., 122, 12 653–12 671, https://doi.org/10.1002/2017JD027543.
Bretherton, C. S., M. E. Peters, and L. E. Back, 2004: Relationships between water vapor path and precipitation over the tropical oceans. J. Climate, 17, 1517–1528, https://doi.org/10.1175/1520-0442(2004)017<1517:RBWVPA>2.0.CO;2.
Cavazos, T., 1999: Using self-organizing maps to investigate extreme climate events: An application to wintertime precipitation in the Balkans. J. Climate, 13, 1718–1732, https://doi.org/10.1175/1520-0442(2000)013<1718:USOMTI>2.0.CO;2.
Chen, X., F. Jiang, Y. Wang, Y. Li, and R. Hu, 2013: Characteristics of the eco-geographical pattern in arid land of Central Asia (in Chinese). Ganhanqu Yanjiu, 30, 385–390.
Dai, X., W. Li, and Z. Ma, 2006: Changes of Xinjiang’s water vapor sources in the past ten years (in Chinese). Prog. Nat. Sci., 16, 1651–1656, https://doi.org/10.3321/j.issn:1002-008X.2006.12.018.
Dee, D. P., and Coauthors, 2011: The ERA-Interim reanalysis: Configuration and performance of the date assimilation system. Quart. J. Roy. Meteor. Soc., 137, 553–597, https://doi.org/10.1002/qj.828.
Dorling, S., T. Davies, and C. Pierce, 1992: Cluster analysis: A technique for estimating the synoptic meteorological controls on air and precipitation chemistry-method and applications. Atmos. Environ., 26A, 2575–2581, https://doi.org/10.1016/0960-1686(92)90110-7.
Drumond, A., R. Nieto, and L. Gimeno, 2011: Sources of moisture for China and their variations during drier and wetter conditions in 2000–2004: A Lagrangian approach. Climate Res., 50, 215–225, https://doi.org/10.3354/cr01043.
Fremme, A., and H. Sodemann, 2019: The role of land and ocean evaporation on the variability of precipitation in the Yangtze River Valley. Hydrol. Earth Syst. Sci., 23, 2525–2540, https://doi.org/10.5194/hess-23-2525-2019.
Guan, X., L. Yang, Y. Zhang, and J. Li, 2019: Spatial distribution, temporal variation, and transport characteristics of atmospheric water vapor over Central Asia and the arid region of China. Global Planet. Change, 172, 159–178, https://doi.org/10.1016/j.gloplacha.2018.06.007.
Han, X., H. Xue, C. Zhao, and D. Lu, 2016: The roles of convective and stratiform precipitation in the observed precipitation trends in Northwest China during 1961–2000. Atmos. Res., 169, 139–146, https://doi.org/10.1016/j.atmosres.2015.10.001.
Hu, Q., D. Jiang, X. Lang, and B. Xu, 2018a: Moisture sources of the Chinese Loess Plateau during 1979–2009. Palaeogeogr. Palaeoclimatol. Palaeoecol., 509, 156–163, https://doi.org/10.1016/j.palaeo.2016.12.030.
Hu, Q., D. Jiang, and X. Lang, 2018b: Sources of moisture for different intensities of summer rainfall over the Chinese Loess Plateau during 1979–2009. Int. J. Climatol., 38, e1280–e1287, https://doi.org/10.1002/joc.5416.
Hu, Z., Q. Hu, C. Zhang, X. Chen, and Q. Li, 2016: Evaluation of reanalysis, spatially interpolated and satellite remotely sensed precipitation data set in central Asia. J. Geophys. Res. Atmos., 121, 5648–5663, https://doi.org/10.1002/2016JD024781.
Huang, W., S. Feng, J. Chen, and F. Chen, 2015: Physical mechanisms of summer precipitation variations in the Tarim Basin in northwestern China. J. Climate, 28, 3579–3591, https://doi.org/10.1175/JCLI-D-14-00395.1.
Huang, W., S. Chang, C. Xie, and Z. Zhang, 2017: Moisture sources of extreme summer precipitation events in North Xinjiang and their relationship with atmospheric circulation. Adv. Climate Change Res., 8, 12–17, https://doi.org/10.1016/j.accre.2017.02.001.
Huang, W., T. Qiu, Z. Yang, D. Lin, J. S. Wright, B. Wang, and X. He, 2018: On the formation mechanism for wintertime extreme precipitation events over the southeastern Tibetan Plateau. J. Geophys. Res. Atmos., 123, 12 692–12 714, https://doi.org/10.1029/2018JD028921.
Mewes, D., and C. Jacobi, 2019: Heat transport pathways into the Arctic and their connections to surface air temperatures. Atmos. Chem. Phys., 19, 3927–3937, https://doi.org/10.5194/acp-19-3927-2019.
Kohonen, T., 1998: The self-organizing map. Neurocomputing, 21, 1–6, https://doi.org/10.1016/S0925-2312(98)00030-7.
Lennard, C., and G. Hegerl, 2015: Relating changes in synoptic circulation to the surface rainfall response using self-organising maps. Climate Dyn., 44, 861–879, https://doi.org/10.1007/s00382-014-2169-6.
Li, W., K. Wang, S. Fu, and H. Jiang, 2008: The interrelationship between regional westerly index and the water vapor budget in Northwest China (in Chinese). Bingchuan Dongtu, 30, 28–34.
Loikith, P. C., B. R. Lintner, and A. Sweeney, 2017: Characterizing large-scale meteorological patterns and associated temperature and precipitation extremes over the Northwestern United States using self-organizing maps. J. Climate, 30, 2829–2847, https://doi.org/10.1175/JCLI-D-16-0670.1.
Moody, J. L., and J. N. Galloway, 1988: Quantifying the relationship between atmospheric transport and the chemical composition of precipitation on Bermuda. Tellus, 40B, 463–479, https://doi.org/10.3402/tellusb.v40i5.16014.
Newell, R. E., N. E. Newell, Y. Zhu, and C. Scott, 1992: Tropospheric rivers? A pilot study. Geophys. Res. Lett., 19, 2401–2404, https://doi.org/10.1029/92GL02916.
Nieto, R., and L. Gimeno, 2019: A database of optimal integration times for Lagrangian studies of atmospheric moisture sources and sinks. Sci. Data, 6, 59, https://doi.org/10.1038/s41597-019-0068-8.
Peng, D., and T. Zhou, 2017: Why was the arid and semiarid Northwest China getting wetter in the recent decades? J. Geophys. Res. Atmos., 122, 9060–9075, https://doi.org/10.1002/2016JD026424.
Ren, G., and Coauthors, 2005: Climate changes of China’s mainland over the past half century (in Chinese). Acta Meteor. Sin., 63, 942–956.
Reusch, D. B., R. B. Alley, and B. C. Hewitson, 2005: Relative performance of self-organizing maps and principal component analysis in pattern extraction from synthetic climatological data. Polar Geogr., 29, 188–212, https://doi.org/10.1080/789610199.
Saha, S., and Coauthors, 2010: The NCEP Climate Forecast System Reanalysis. Bull. Amer. Meteor. Soc., 91, 1015–1058, https://doi.org/10.1175/2010BAMS3001.1.
Saha, S., and Coauthors, 2014: The NCEP Climate Forecast System version 2. J. Climate, 27, 2185–2208, https://doi.org/10.1175/JCLI-D-12-00823.1.
Sheridan, S. C., and C. C. Lee, 2011: The self-organizing map in synoptic climatological research. Prog. Phys. Geogr., 35, 109–119, https://doi.org/10.1177/0309133310397582.
Shi, Y., Y. Shen, E. Kang, D. Li, Y. Ding, G. Zhang, and R. Hu, 2007: Recent and future climate change in northwest China. Climatic Change, 80, 379–393, https://doi.org/10.1007/s10584-006-9121-7.
Skific, N., J. A. Francis, and J. J. Cassano, 2009: Attribution of projected changes in atmospheric moisture transport in the Arctic: A self-organizing map perspective. J. Climate, 22, 4135–4153, https://doi.org/10.1175/2009JCLI2645.1.
Sodemann, H., and E. Zubler, 2010: Seasonal and inter-annual variability of the moisture sources for Alpine precipitation during 1995–2002. Int. J. Climatol., 30, 947–961, https://doi.org/10.1002/joc.1932.
Sodemann, H., C. Schwierz, and H. Wernli, 2008: Interannual variability of Greenland winter precipitation sources: Lagrangian moisture diagnostic and North Atlantic Oscillation influence. J. Geophys. Res. Atmos., 113, D03107, https://doi.org/10.1029/2007JD008503.
Song, S., L. Li, X. Chen, and J. Bai, 2015: The dominant role of heavy precipitation in precipitation change despite opposite trends in west and east of northern China. Int. J. Climatol., 35, 4329–4336, https://doi.org/10.1002/joc.4290.
Stohl, A., and P. James, 2004: A Lagrangian analysis of the atmospheric branch of the global water cycle. Part I: Method description, validation, and demonstration for the August 2002 flooding in Central Europe. J. Hydrometeor., 5, 656–678, https://doi.org/10.1175/1525-7541(2004)005<0656:ALAOTA>2.0.CO;2.
Stohl, A., and P. James, 2005: A Lagrangian analysis of the atmospheric branch of the global water cycle. Part II: Moisture transports between Earth’s ocean basins and river catchments. J. Hydrometeor., 6, 961–984, https://doi.org/10.1175/JHM470.1.
Stohl, A., C. Forster, A. Frank, P. Seibert, and G. Wotawa, 2005: Technical note: The Lagrangian particle dispersion model FLEXPART version 6.2. Atmos. Chem. Phys., 5, 2461–2474, https://doi.org/10.5194/acp-5-2461-2005.
Sun, B., and H. Wang, 2014: Moisture sources of semiarid grassland in China using the Lagrangian particle model FLEXPART. J. Climate, 27, 2457–2474, https://doi.org/10.1175/JCLI-D-13-00517.1.
Viste, E., and A. Sorteberg, 2013: Moisture transport into the Ethiopian Highlands. Int. J. Climatol., 33, 249–263, https://doi.org/10.1002/joc.3409.
Wu, J., and X. Gao, 2013: A gridded daily observation dataset over China region and comparison with the other datasets (in Chinese). Chin. J. Geophys., 56, 1102–1111, https://doi.org/10.6038/cjg20130406.
Yang, L., and Y. Zhang, 2017: Summary of current research on Central Asian vortex. Adv. Climate Change Res., 8, 3–11, https://doi.org/10.1016/j.accre.2017.03.002.
Yang, L., X. Li, and G. Zhang, 2011: Some advances and problems in the study of heavy rain in Xinjiang (in Chinese). Climatic Environ. Res., 16, 188–198, https://doi.org/10.3878/j.issn.1006-9585.2011.02.08.
Yang, L., Y. Zhang, and H. Qin, 2015: Some advances and problems of Middle-Asia vortex (in Chinese). Desert Oasis Meteor., 9, 1–8, https://doi.org/10.3969/j.issn.1002-0799.2015.05.001.
Yao, J., Q. Yang, W. Mao, and X. Han, 2018: Analysis of a summer rainstorm water vapor paths in Tianshan Mountains (Xinjiang) based on HYSPLIT4 model (in Chinese). Plateau Meteor., 37, 68–77, 10.7522/j.issn.1000-0534.2017.00031.
Zhang, J., N. Li, H. Qin, J. Li, J. Liu, W. Liu, and B. Meili, 2016: The observational analysis and water vapor characteristics of a rainstorm process in Xinjiang (in Chinese). Torrential Rain Disasters, 35, 537–545.
Zhang, Y., B. Yu, Z. Wang, and L. Jia, 2018: Dynamic mechanism and water vapor transportation characteristics of two extreme rainstorm events in Ili River valley in summer of 2016 (in Chinese). Torrential Rain Disasters, 37, 435–444.
Zhou, Y., Z. Xie, and X. Liu, 2019: An analysis of moisture sources of torrential rainfall events over Xinjiang, China. J. Hydrometeor., 20, 2109–2122, https://doi.org/10.1175/JHM-D-19-0010.1.
