Moisture sources of summer heavy precipitation in two spatial patterns over Northeast China during 1979–2021

This study classifies the spatial distribution of heavy precipitation in summer (June–August) from 1979 to 2021 in the three provinces of Northeast China (TPNC) into two patterns by using the self‐organizing maps (SOM) neural network, and then quantitatively analyzes their moisture transport channels and sources using the Lagrangian model. The results show that the summer heavy precipitation in TPNC can be divided into the northern and southern patterns according to the distribution of the heavy precipitation. Both patterns of heavy precipitation are affected by the low‐level vortex west of TPNC, but the strength and shape of the low vortex are different. The northern pattern is mainly influenced by the westerly flow in the vortex in the mid‐high latitudes, which transports moisture from the upstream westerly region into TPNC. The southern pattern is mainly affected by the southerly jet stream southeast of TPNC, which conveys a large amount of moisture from the East Asian summer monsoon region into TPNC. In terms of the summer climatological mean, the northern pattern has a higher precipitation recycling rate, while the southern pattern has a lower recycling rate.


| INTRODUCTION
Northeast China, located at the northern edge of the East Asian monsoon region, is a typical climate-vulnerable area of China (Song et al., 2022).Summer is the peak season for precipitation in Northeast China, with the highest frequency of heavy precipitation (Wang & Ding, 2008) and is also the primary growing season for local crops (Zhou & Wang, 2014).Heavy precipitation is closely related to local floods and results in substantial economic losses for society (Yang et al., 2022).During 1961During -2018, the extreme summer precipitation in Northeast China showed a nonsignificant upward trend (Tang et al., 2021).The frequency of extreme precipitation caused by specific circulation patterns also showed a decrease from the 1960s to the 1990s, and then increased to match the changes in the intensity of the East Asian summer monsoon, although this trend was insignificant (Ding et al., 2008;Liu et al., 2012;Tang et al., 2021).In addition, it has been predicted that as the water cycle intensifies, heavy precipitation in Northeast China will increase by the late 21st century and the related flood risk will rise (Tabari, 2020).Therefore, understanding the physical mechanism of summer heavy precipitation in Northeast China can provide scientific support for longterm predictions, as well as disaster prevention and mitigation.
Northeast China summer precipitation is closely related to accompanying atmospheric circulations at low and high latitudes, such as the western North Pacific subtropical high, the East Asian summer monsoon, the Northeast China cold vortex, the Okhotsk blocking high, and the meridional and zonal propagations of the teleconnection wave train (Li et al., 2018;Shen et al., 2011;Zhao & Sun, 2007).For instance, a strong Okhotsk high favors the convergence of humid air flow from the south and cold air flow carried by the Northeast China cold vortex, which increase precipitation in Northeast China (Chen et al., 2020).As the northward-moving tropical system and the Northeast China cold vortex combine, rainstorms tend to occur (Zhao & Sun, 2007).These atmospheric circulation factors work together to affect summer precipitation, especially heavy precipitation (Fang et al., 2017;Xu & Qi, 2022), which can lead to various precipitation patterns.Clustering the heavy precipitation in Northeast China with similar spatial distributions can be used to effectively investigate the mechanisms of the leading patterns of heavy precipitation in Northeast China.
A sufficient moisture supply is a necessary condition for heavy precipitation (Drumond et al., 2011;Newell et al., 1992;Zhou & Yu, 2005).It has been shown that the East Asian summer monsoon region is the main source for Northeast China summer precipitation, and the moisture first converges in eastern China and nearby seas and then moves northward to Northeast China with the rain belt of the East Asian summer monsoon (Han et al., 2019;Sun et al., 2007).Sun et al. (2010) indicated that the main moisture source of large-scale summer heavy precipitation was the East Asian monsoon region, while the moisture source of local heavy precipitation was the upstream westerly region during 1961-2005.In addition, the moisture transport characteristics of cold vortex heavy precipitation have been examined using the Hybrid Single-Particle Lagrangian Integrated Trajectory model (Ma et al., 2017;Wei et al., 2015;Yang et al., 2022).These studies mainly focused on a few heavy precipitation cases and qualitatively analyzed the channels and sources of their moisture supplies.However, little attention has been given to the general characteristics of the moisture source for Northeast China summer heavy precipitation and the related quantitative analysis of moisture sources.Therefore, it is of interest to examine moisture sources for summer heavy precipitation in Northeast China.
Based on the above premise, the following questions are important: What are the main spatial patterns of summer heavy precipitation in Northeast China?What circulation systems affect the different patterns of heavy precipitation?What are the similarities and differences between the moisture transport process and the moisture source of different patterns of heavy precipitation?To investigate the aforementioned questions, the remaining content is structured as follows: Section 2 describes the data and methods.Section 3 uses the self-organizing maps (SOM) method to classify the heavy precipitation in Northeast China during summer.The moisture transport and sources of the circulation patterns are analyzed in Section 4. The conclusion and discussion are presented in Section 5. Our study covers Northeast China, including three provinces-Heilongjiang, Jilin, and Liaoning (Figure 1), abbreviated as the three provinces of Northeast China (TPNC).

| Data
The atmospheric data were obtained from the Climate Forecast System Reanalysis (CFSR; Saha et al., 2010, Saha et al., 2014; https://rda.ucar.edu/datasets/ds093.0/, https://rda.ucar.edu/datasets/ds094.0/) and the European Centre for Medium-Range Weather Forecasts Reanalysis v5 (ERA5, Hersbach et al., 2020; https://www.ecmwf.int/en/forecasts/dataset/ecmwf-reanalysis-v5).In parallel to previous studies (Liu et al., 2022;Pisso et al., 2019;Yamamoto & Palter, 2016), CFSR was used as the input data for the FLEXible PARTicle Dispersion Model (FLEXPART, Stohl et al., 2005), including geopotential height, temperature, specific humidity, relative humidity, horizontal wind, and vertical velocity at 42 levels from the surface to 1 hPa.The FLEXPART was used to generate physical quantities related to the motion of five million air particles worldwide.To prevent dataset dependence, CFSR and ERA5 datasets were applied to validate the conclusions of circulation analysis.Both datasets include a 6-h interval on a 0.5 Â 0.5 grid from 1979 to 2021.The results of ERA5 and CFSR in this study were generally consistent, and the analyses based on CFSR are presented in the subsequent sections.
Daily precipitation amounts were obtained from the CN05.1 dataset with a horizontal grid of 0.25 Â 0.25 (Wu & Gao, 2013).During 1979-2021, following earlier studies (Cao et al., 2018;Gao et al., 2017;Xu et al., 2022), the summer (June-August) heavy precipitation events were identified as the days with the top 25th percentile of the daily precipitation amounts averaged over TPNC.Notably that only the days with at least 1 mm precipitation were considered.For 43 summers, 776 heavy precipitation events were selected.

| SOM
SOM is a neural network-based method of clustering (Kohonen, 1998).This method is widely used in clustering meteorological events (Gao et al., 2019;Hewitson & Crane, 2002;Loikith et al., 2017;Mewes & Jacobi, 2020;Sheridan & Lee, 2011;Zhang et al., 2023).In this study, the input data were the spatial distribution of precipitation amounts for heavy precipitation events, and after SOM training, similar precipitation patterns were classified into the same node.The shape of the input array of nodes (i.e., map size) and the number of nodes in the SOM are required to be predefined by the user.An input array of nodes that is too large causes redundant information, which is inconvenient for analysis, while an array that is too small is unable to allocate input data with vastly different characteristics to the right node.Here, we tested five different map sizes (1 Â 2, 1 Â 3, 2 Â 2, 2 Â 3 and 3 Â 3) and found that heavy precipitation events were mainly distributed in two nodes.Even if the number of nodes increased, only a few heavy precipitation events were divided into other nodes.Therefore, the 1 Â 2 size (2 nodes) was selected.In short, according to the spatial distribution of precipitation, 776 heavy precipitation events were classified into two clusters.

| Moisture analysis method
This study used the Lagrangian method to analyze the moisture sources of heavy precipitation in TPNC.The Lagrangian method quantitatively analyzes the source of moisture in precipitation by monitoring the specific humidity changes of air particles from the source area to the precipitation area.The "WaterSip" algorithm (Sodemann et al., 2008) was used in this study for moisture tracking.The algorithm relies on particle data generated by FLEXPART to analyze the moisture changes of particles related to precipitation.The algorithm can quantitatively trace the precipitation moisture back to specific source regions and correlates well with the isotope observational results (Baker et al., 2015).To better depict the movement process of the air particles carrying moisture, the particle aggregation degree (Yao et al., 2021), which is the number of particles at each grid point divided by the total number of particles in this precipitation process, was defined.

| TWO SPATIAL PATTERNS OF HEAVY PRECIPITATION AND RELATED CIRCULATION
Based on the SOM results, two nodes of heavy precipitation patterns are obtained, namely N1 and N2 (Figure 2).Their spatial distributions are identified as the northern pattern (N1) and the southern pattern (N2), which differ from the distribution of all heavy precipitation events.The northern pattern has 442 days, accounting for 57% of the total days, and this number is slightly more than that of the southern pattern (334 days).For the northern pattern, heavy precipitation is mainly concentrated in Heilongjiang Province.For the southern pattern, precipitation is mainly concentrated in the Liaoning and Jilin provinces.The intensity of the precipitation center in the southern pattern is higher than that in the northern pattern.Correspondingly, the regionally averaged precipitation amount of the southern pattern (11.7 mm day À1 ) is also greater than that of the northern pattern (10 mm day À1 ).
For the two patterns of heavy precipitation, the 500 hPa geopotential height (Figure 3) is characterized by the circulation pattern of two ridges and one trough, with the trough located west of TPNC.According to previous The spatial patterns of the summer heavy precipitation events (mm day À1 ) for two SOM ("N1" and "N2") and for all heavy precipitation events ("ALL") in TPNC during 1979-2021.The area-averaged precipitation amount is shown in the top right.The number and node frequency of the heavy precipitation events are shown in the bottom right of the two SOM panels.
studies, this circulation pattern is favorable for Northeast China cold vortex activities that directly lead to summer precipitation there (Fang et al., 2021;Xie & Bueh, 2015).The difference between the two patterns is that the trough of the northern pattern is deeper, with a lowpressure center and a negative anomaly center of the geopotential height north of TPNC.In addition, for the northern pattern, the eastern ridge extends farther north.There are differences in the position and intensity of lowlevel vortices at 850 hPa geopotential height between the two patterns (Figure 3).Specifically, the center of the 850 hPa vortex in the northern pattern is stronger than the other, combined with an obvious high-pressure ridge over the Okhotsk Sea.For the southern pattern, due to southeastern tilt of the low vortex and abnormally high pressure over Japan, the contour lines of the 850 hPa geopotential height are denser south of TPNC than those of the northern pattern.

| Moisture transport
Figure 4 shows the main activity range of precipitable air particles at various backtracking times.In terms of the summer climatology, five main moisture transport channels are identified: Siberia-Lake Baikal-Northeast Asia channel (the northwestern channel), Balkhash Lake-Hexi corridor-Inner Mongolia channel (the western channel), Okhotsk Sea-Japan Sea channel (the eastern channel), South China Sea-Central China-Shandong channel (the southern channel), and Philippine Sea-Ryukyu Islands channel (the southeastern channel).In addition, these two patterns of heavy precipitation have different moisture transport processes (Figure 4).For example, at 5 days before the occurrence of heavy precipitation, the southern pattern (N2) has a wider range of particle activity in the southern and southeastern channels, while the range in the northwestern and western channels is smaller than the northern pattern (N1).
Figure 5 further shows the moisture transport process in detail through the distribution of the particle aggregation degree at multiple times before heavy precipitation.The particles from the two patterns of heavy precipitation gathered in Shandong Province and the surrounding areas at 4 days before the precipitation and then entered TPNC to form precipitation. Compared to the southern pattern, the northern pattern (N1) has fewer particles from the southern channel, while particles from Siberia and Northwest China are significantly more.For the southern pattern (N2), a large number of particles developed in the northwestern Pacific during 4-8 days before precipitation and subsequently gathered in East Asia.
To understand the process for the above differences in air particle transport, we further investigated the 850 hPa wind (Figure 6).The wind and geopotential height of the two patterns have consistent cyclones in the mid-low troposphere over TPNC.The wind in the northern pattern reflects a horizontally oriented cyclone, with strong westerlies south of TPNC.The southern pattern has a weaker cyclone, showing the obvious southerly jet southeast of TPNC.The cyclone of the northern pattern sends a large number of particles from the west into TPNC, while the jet stream of the southern pattern brings air particles from East Asia into TPNC.
F I G U R E 4 Activity range of main precipitation particles for the summer heavy precipitation patterns at different backtracking times for each SOM node ("N1" and "N2") and all summer days ("JJA") during 1979-2021 (e.g., dark red indicates the activity range of particles 9-10 days before precipitation).Note that the colors overlap, which means that the range of colors on the outside includes all the colors on the inside.

| Moisture sources
Based on the "WaterSip" method (Sodemann et al., 2008), the distributions of moisture sources for the two heavy precipitation patterns were obtained (Figure 7).The similarity is that most of East Asia, the East China Sea, and the Japan Sea are the main moisture sources for the TPNC summer heavy precipitation, and the maximum centers are mainly distributed in the Circum-Bohai Sea.In terms of the differences in the distribution of moisture between these two patterns, the main moisture source in the northern pattern extends westward to 110 E, while the southern boundary only extends to northern the South China Sea.The moisture source in the southern pattern has a wider range of moisture centers, and the main moisture source extends southward to the whole South China Sea, but only westward to approximately 105 E. The positive anomalies of moisture contribution in the northern pattern are mainly distributed in TPNC and west of it, while the southern pattern has more moisture from the east China and the Northwest Pacific than the summer climatology (Figure 7c,d).
We further quantified the moisture contribution to precipitation from different moisture sources for the two patterns of heavy precipitation.First, five major subregions of moisture sources were divided (Figure 8a) based on the distribution discrepancies of moisture sources between the two patterns: TPNC, Siberia-Mongolia-Xinjiang, Central-East China, the Yellow Sea-Japan Sea, and the South China Sea-Philippine Sea.The moisture contribution rate of each subregion for the northern and southern patterns was further calculated separately (Figure 8b).The contribution rate is the ratio of moisture contributions from a source subregion to each heavy precipitation.The northern pattern (N1) has some characteristics in common with the summer climatology.For example, Siberia-Mongolia-Xinjiang accounts for more than 30% of the precipitation moisture, while the South China Sea-Philippine Sea accounts for less than 10%.However, the local contribution rate is significantly higher than that of the summer climatology, which indicates that the heavy precipitation in the northern pattern has a higher precipitation recycling ratio.The moisture source of the southern pattern (N2) is significantly different from that of the northern pattern or summer F I G U R E 5 Particle aggregation degree (units: 10 À5 ) for each SOM node ("N1" and "N2") and the difference between N2 and N1 ("N2-N1") at different backtracking times for each SOM node (e.g., "-12 h" denotes the time 12 h prior to precipitation).The particle aggregation degree is defined as the number of particles in each 0.5 Â 0.5 grid divided by the total number of particles in a precipitation process.Note that the color bar is logarithmic, following Bohlinger et al. (2017) and Huang et al. (2018).
climatology.The most important source of precipitation moisture is Central-East China, while the contribution rate of Siberia-Mongolia-Xinjiang decreases to approximately 20% and the South China Sea-Philippine Sea increases to the same level as the level of Siberia-Mongolia-Xinjiang. The recycling ratio of the southern pattern is also significantly lower than that of the northern pattern and the summer climatology.
The above difference in moisture contributions can be explained by the aggregation process of air particles carrying moisture (Figure 5).Compared with those of the northern pattern, more particles of the southern pattern are concentrated in Central-East China and the South China Sea-Philippine Sea, leading to an increase in the contribution rates of Central-East China and the South China Sea-Philippine Sea and a decrease in those of Siberia-Mongolia-Xinjiang. However, the South China Sea-Philippine Sea has not become the main moisture source (Han et al., 2019;Sun et al., 2010), because although particles can obtain a large amount of moisture from oceans, moisture is lost in the middle of longdistance transport (Hu et al., 2018;Sun & Wang, 2013).The difference in particle transport can be further explained by 850 hPa winds (Figure 6).We note that there is no significant difference in the 850 hPa wind over South China between the two patterns and the summer climatology.This means that moisture from the South China Sea and other oceans can be sent to Central-East China in both patterns.However, only the southerly jet stream of southern pattern (N2) can effectively transport air particles in Central-East China to TPNC.This indicates that the relationship between the strength of the East Asian summer monsoon, defined by wind in South China, and the heavy precipitation in TPNC is unstable.Only heavy precipitation patterns with strong southerly winds south of TPNC are associated with the East Asian summer monsoon (Tang et al., 2021).

| CONCLUSION AND DISCUSSION
SOM was applied to classify the summer heavy precipitation in TPNC from 1979 to 2021, and then the connections between circulation patterns and heavy precipitation in TPNC were analyzed.To quantify the F I G U R E 6 850 hPa wind (a, b, unit: m s À1 ) for each SOM node ("N1" and "N2") and its anomaly (c, d, "ano" means anomaly) of the summer heavy precipitation patterns during 1979-2021.Only differences that are significant at the 95% confidence level are shown in the anomaly.
moisture transport and sources, the Lagrange method was used to analyze the moisture source for each SOM node.The main conclusions are as follows.
1.There are two spatial patterns of SOM-based heavy precipitation events, namely, the northern and southern patterns.Although both patterns of heavy precipitation are related to the vortex at 850 hPa west of TPNC, the strength and shape of the vortex are different.2. Five moisture transport channels during summer heavy precipitation are identified for TPNC, but the two spatial patterns of heavy precipitation show different moisture transport characteristics.For the northern pattern, more air particles carrying moisture are transported through the western and northwestern channels, while the southern pattern mainly highlights the importance of the southern and southeastern channels.3. Siberia-Mongolia-Xinjiang is the main moisture source for the northern pattern.The moisture contributions from Central-East China and the South China Sea-Philippine Sea are crucial to the southern pattern.The precipitation recycling rate of the northern pattern of heavy precipitation is also significantly higher than that of the southern pattern.
Previous studies on the moisture source of summer heavy precipitation in Northeast China have mainly emphasized moisture from the oceans (Han et al., 2019;Sun et al., 2010).Although this study found that the moisture sources of the two patterns of heavy precipitation are very different (Figure 8b), the total moisture contribution of the land is greater than that of the oceans.A similar conclusion was noted by Yang et al. (2022) in moisture tracking study of a cold vortex rainstorm over Northeast China on July 25, 2016.Even in the Yangtze River basin located in the East Asian monsoon region, 58% of the direct precipitation moisture is from land, according to Fremme and Sodemann (2019).Consequently, the land regions adjacent to the precipitation area could act as a rich moisture source.This finding also indicates the need to conduct quantitative research on precipitation moisture sources.
When using SOM to cluster the spatial pattern of heavy precipitation in Northeast China, we also noticed that the distribution of a few heavy precipitation events is different from the two patterns noted in the article.Moreover, the heavy precipitation circulation pattern we summarized is mainly related to the vortex or low trough.Other circulation systems (shear, cyclone, typhoon and cutoff low-pressure system) could still lead to heavy precipitation in Northeast China (Sun et al., 2010;Wu et al., 2022;Zhao & Sun, 2007).The heavy precipitation events caused by these circulation systems are also closely related to the two patterns identified in this study.For example, the northward movement of typhoons (Ma et al., 2017;Sun et al., 2010;Wang & Wang, 2021) can also lead to heavy precipitation in the southern part of TPNC, which resembles the southern pattern (N2).

F
I G U R E 1 Map of the study area (red line) and topography (shading; units: km).

F
I G U R E 7 The composite moisture uptake (a, b, unit: mm day À1 ) for each SOM node ("N1" and "N2") and its anomaly (c, d, "ano" means anomaly) of summer heavy precipitation patterns during 1979-2021.The 80th percentiles of the mass contributed by moisture sources are shown in (a, b) contours.Note that the color bar is logarithmic, followingBohlinger et al. (2017) andHuang et al. (2018).