Changes in Hourly Extreme Precipitation Over Eastern China From 1970 to 2019 Dominated by Synoptic‐Scale Precipitation

Because of its dense population, extreme precipitation, in particular hourly extreme precipitation (HEP), is receiving increasing attention from both academic and public bodies in eastern China. Based on a continuous 50‐year record of hourly precipitation and reanalysis data, we show here for the first time that changes in the HEP occurrence are dominated by changes in the duration of the Meiyu front system. Further analyses reveal that greater occurrence of HEP in northeastern China, the lower reach of Yangtze River, and southern China during the warm season is largely due to a longer duration of the post‐Meiyu I stage when Meiyu front stays in northern China and meridional circulation dominates the eastern coastal area of China. These results improve our understanding of the changing behavior of extreme rainfall in China and shed light on the prevention of flash floods.


Data and Method
An hourly rainfall data set was obtained from the National Meteorological Information Center, China Meteorological Administration, which maintained rain-gauge records at 2,435 stations in mainland China from 1951 to 2019. We adopted the criteria of the relocation problem and the continuous valid record after Zhai et al. (2005) and H. Zhang and Zhai (2011). Observational data were not included if a station was moved more than 20 horizontal kilometers from the original location or reinstalled in elevation 50 m or more during the study period, or the availability rate, defined as the rate of correct and valid records out of the entire records in the warm season (May-September) in any year, was less than 80%. Since the availability rate of data is greater than 97.5% and more stable after 1970, the analyses were restricted to the period 1970-2019 to ensure that the records were reliable and covered sufficient observatories in eastern China from 1970 to 2019. A total of 741 stations passed quality control ( Figure 1a).
As the primary monsoon system affecting China, East Asian summer monsoon (EASM), a subtropical monsoon in which the low-level winds reverse primarily from northerly in winter to southerly in summer (Ding & Chan, 2005;B. Wang & LinHo, 2002), has different rainfall stages, i.e., spring, pre-Meiyu, Meiyu, mid-summer, and fall Kong et al., 2017). To quantify the synoptic-scale circulation at different monsoon stages, we used reanalysis data from the National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) (Kalnay et al., 1996). This reanalysis data covered the period from 1948 to the present day, with four values per day and a spatial resolution of 2.5° latitude × 2.5° longitude. The pseudo-equivalent potential temperature ( e ) (Bolton, 1980) and horizontal winds at 850 hPa were calculated at 6-h basis.

Definition of HEP
In the study, HEP events were defined at each station as precipitation amounts greater than the 95th percentile of the hourly rainfall (rain rate greater than 0.1 mm h −1 ) in the warm season (May-September) during 1970-2019 ( Figure S1). The overall results will not change much if we choose the 90th percentile as the criteria to define HEP (not shown).

Determination of Local- and Synoptic-Scale HEP (L-HEP and S-HEP)
We defined L-HEP and S-HEP following Guo et al. (2017) with modifications. An HEP was considered as an L-HEP at a weather station when the HEP met the following two criteria: within a 200 km radius 1 around the station, there had to be more than five stations besides the station itself (criteria 1), and the percentage of stations with rainfall record had to be less than 20% (criteria 2). If an HEP event only satisfied the first 1 To include more stations within the radius for effectively calculating different scales of hourly precipitation, 200-km radius is chosen here instead of 150-km in Guo et al. (2017) although the overall results do not change. criteria but not the second, it was considered as an S-HEP. Note, only stations at which the precipitation records meet the first criteria can be used to distinguish between L-HEP and S-HEP events; as a result, of the 741 stations passing quality control, 716 stations are chosen (Figures 1b-1e, and 2).

Self-Organizing Maps (SOMs)
To further understand the changing behavior of the duration and averaged daily HEP frequency in the warm season from 1970 to 2019, a clustering method SOM analysis (Johnson, 2013;Kohonen, 2001;Kohonen et al., 1996) was applied to the HEP for all the 741 stations. The occurrence of HEP is first counted in a day, and calculated into a 50-year average or a 25-year average for the 1970-1994 and 1995-2019 period for each day; a 5-day moving average is then conducted before utilizing the SOMs technique to avoid high-frequency noise and better capture the characteristics of HEP patterns at different monsoon stages. The HEP events were classified according to the five monsoon stages in the warm season. The SOM method is a neural network-based cluster analysis that classifies a high-dimensional data set into representative patterns, using a neighborhood function to topologically order the high-dimensional input and group similar clusters together (Kohonen & Somervuo, 1998). Unlike other clustering methods (e.g. k-means [Lin & Chen, 2006] and Ward's methods [Bao & Wallace, 2015;Lin & Chen, 2006]), SOMs could extract more robust and distinctive topology information. This method is ideally suited given the discreteness of intraseasonal stages and abrupt transitions evident in the EASM . The SOMs function in the MATLAB 2018b Deep Learning Toolbox (https://www.mathworks.com/help/deeplearning/index.html) was used to cluster the data and reduce their dimensionality.
NG ET AL.
10.1029/2020GL090620 3 of 9  Figure 1a shows the mean occurrence of HEP in the warm season for the 741 stations over eastern China during 1970-2019. The HEP occurrence decreases from south to north of eastern China and varies between 7.20 and 41.18 h across the region. S-HEP events account for more than 85.0% of total HEP in most areas, in particular over the lower reach of the Yangtze River (LYR) (Figure 1b). The minimum percentage of S-HEP over total HEP is about 61.7%, in northern China (NC).

Characteristics of HEP
In eastern China, the main precipitation system in the warm season is usually associated with the shift of the Meiyu front and occasionally tropical cyclones (TCs), we thus analyze the latitude variation of the occurrence of HEP, S-HEP, and L-HEP, respectively, from May to September during 1970-2019. Along with the monsoon onset, HEP starts in May at the low latitude region (south of 30°N). During June and July, the high-occurrence of HEP (i.e., HEP is greater than 10-h) moves from southern to northern China and retreats southward quickly in late August. During the warm season, HEP events appear more frequently in the low latitude region (south of 30°N) compared to the middle (30°N-38°N) and high latitude (north of 38°N) regions. The latitude variation of S-HEP (Figure 1d) is similar to that of HEP ( Figure 1c) from May to September, indicating that changes in synoptic-scale HEP dominate changes in total HEP.  (Kendall, 1975;Mann, 1945;Sen, 1968). Positive trends of total HEP are observed in the SC, LYR, and northeastern China (NEC) regions, generally in consistent with the results of H. Zhang and Zhai (2011) who analyzed fewer gauge data ending at 2000, but more stations are showing an upward trend in our study. Specifically, HEP increases with time in 489 stations, the trends are significant at a 5.0% level for 101 stations in the SC, LYR, and NEC regions. Downward trends are found at 66 stations, but none of them is significant. Of the 741 stations, 161 stations show that the slope is 0 (Figure 2a). Comparing Figures 2a, 2b, and 2c, we find that the positive trend of HEP is mainly due to the increase in S-HEP (Figure 2b). The correlation between the trend in HEP and S-HEP is 0.82, which is much higher than that between the HEP and L-HEP, with a value of 0.17 (Figures 2b-2c).

Changes in the Occurrence of HEP
We further analyze the occurrence changes of HEP, S-HEP, and L-HEP in different rainfall intensity categories and the total amount of rainfall from the first  to the second (1995-2019) 25-year periods (Figure 2d). Among the 11 different rainfall intensities, the highest HEP occurrence occurs when precipitation intensity is within 10.10-15.00 mm h −1 in both periods.
Compared with the first 25-year period, the occurrence of total HEP grows by about 9.1% in the second period (left-most bars Figure 2d), and the growth in HEP number is mainly due to the increase in S-HEP. The increasing features in S-HEP and L-HEP are observed in nine rainfall intensity bins, except for the intensity bins in 0.10-5.00 and 5.10-10.00 mm h −1 where L-HEP is decreasing. Figure 2d also shows that the occurrence of S-HEP increases the most within the 10.10-15.00 mm h −1 rainfall intensity bin, same as total HEP. In summary, the changed occurrence of S-HEP dominates the number of growths in total HEP during the second 25-year period.
NG ET AL.
10.1029/2020GL090620 4 of 9   and second  25-year periods. The left of the Y-axis (blue) indicates the occurrence and the right of the Y-axis (gray) indicates the difference between the first and second periods. The difference with statistical significance at the 95.0% confidence level are represented by solid dots. The X-axis shows the total amount of rainfall for an hour, rainfall intensity in 0.1-5, 5.1-10, 10.1-15 mm/h, …, 45.1-50 mm/h, and >50 mm/h, respectively.

What Lead to Changes in HEP?
It has been long known that rainfall in eastern China is closely related to monsoons in the warm season (Ding & Chan, 2005;Tao, 1987). It is necessary to find out if changes in EASM between the first and second 25-year period dominate the changes in the warm-season HEP occurrence over eastern China.
We first show the 50-year climatological pattern of HEP (Figure 3) based on SOMs analyses. The five SOMs HEP nodes reasonably capture the intraseasonal stages of EASM rainfall over eastern China Kong et al., 2017). The rainfall patterns present here are generally in agreement with those in Chiang et al. (2017) and Kong et al. (2017), with slight differences in onset time and duration because we use different precipitation data sets and up-to-date research periods (ending at 2019).
During the pre-Meiyu stage (May 1-June 7), HEP first appears in SC (mainly south of 30°N, Figure 3a). The southerly wind brings moisture from adjacent oceans to SC, favorable to generate HEP events. From June 8 to June 28, the Meiyu front migrates northward to the LYR region at the Meiyu stage (Figure 3b). At the same time, following the monsoon movement, HEP occurs at the east side of the Tibet Plateau, where the Indian summer monsoon (ISM), a tropical monsoon in which the low-level winds reverse from winter easterlies to summer westerlies (Ding & Chan, 2005;B. Wang & LinHo, 2002), joins the EASM to facilitate convection. After the Meiyu stage, the southerly wind continues to strengthen and moves further north. During the post-Meiyu I stage (Figure 3c), the Meiyu front migrates northward to the NC region, leading to the high daily HEP frequency there. Besides, the HEP can be observed in the NEC region, likely due to the mountain topography uplifting airflow (X. Zhang et al., 2012). The SC also has a relatively high averaged daily HEP frequency (around 0.15-0.20 h d −1 ) in the post-Meiyu I stage, associated with the meridional circulation accompanied by the strong low-level jet that provides moisture and dynamic forcing of convection Q. Zhang et al., 2017). The averaged daily HEP frequency in SC is not as high as in the pre-Meiyu stage, which is around 0.20-0.25 h d −1 . The ISM is strongest in July when the warm and humid NG ET AL.
10.1029/2020GL090620 5 of 9 air mass meets the air mass from the north in southwestern China (SWC) to form the southwest vortex, which may be responsible for the extreme precipitation (Figure 3c) (Chen et al., 2015;X. Wang e al., 2017).
In the post-Meiyu II stage, the 340-K isoline of  e , influenced by the weakening of the EASM and the ISM, retreats to LYR, leading to the declines of the averaged daily HEP frequency in both NC and NEC regions (Figure 3d). In the SC region, the dominant winds are southeasterly. The majority of HEP events are observed in the coastal area and SWC. As the monsoons retreat further south in fall (Figure 3e), except for the impact of TCs on the SC and some weak weather systems in individual areas (Luo et al., 2016), HEP events end in most of the areas north of the Yangtze River (Figure 3e). We note that the variation in L-HEP (unit: h d −1 ) is not obvious ( Figure S2), indicating that variation in HEP is mainly due to changes in S-HEP.
However, the onset and duration of HEP in the five stages change from the first  to the second (1995-2019) period. Figure 4 shows the differences in occurrence of HEP in hours (Figures 4a-4e), the beginning/ending dates and duration at each stage (Figure 4f). The pre-Meiyu, Meiyu, and post-Meiyu II stages end early by 4, 9, and 3 days, respectively, whereas the starting dates begin early for the Meiyu, post-Meiyu I, and fall stages in the latter period. Thus, compared to the first 25-year period, the duration of the post-Meiyu I and fall precipitation stages increase by 12 and 3 days, separately, but the pre-Meiyu, Meiyu, and post-Meiyu II stages shorten by 4, 5, and 6 days during 1995-2019. A nonparametric bootstrap method (with a hundred samples) is used for estimating the statistical significance of the difference in each stage's duration (Davison & Hinkley, 1997;Lupu & Maenhaut, 2002;Marchand et al., 2006). Results suggest that the changes in the durations of the five stages from 1970-1994 to 1995-2019 are all significant (Table S1).
Changes in the duration do not necessarily alter the total occurrence of HEP in each monsoon stage, the daily mean of HEP occurrence, that is, the rate of HEP change per day, could also alter; we thus further analyze the relative contributions of the changes in duration and rate of HEP to the total occurrence of HEP from the first to the second 25-year period. Results suggest that from 1970-1994 to 1995-2019, about 1970-1994 and 1995-2019, respectively. the post-Meiyu I and fall stages (Figures S3c and S3e), which contribute to the changes in HEP. Compared to the duration changes, HEP daily frequencies (i.e., the rate of change per unit time) increase at fewer stations, 5.5% and 7.8% stations passing the significance test at the post-Meiyu I and fall stages, respectively.
The patterns of differences in the total occurrence of HEP between 1970-1994 and 1995-2019 are dissimilar among the five stages. Increased HEP occurrence is observed across eastern China in the post-Meiyu I (Figure 4c), and over much of eastern China during the fall stage (Figure 4e), respectively. During post-Meiyu I and fall stages, S-HEP increases by 48.4% and 27.2%, separately ( Figures S4-S5). On the contrary, during the other three stages, i.e., the pre-Meiyu (Figure 4a), Meiyu (Figure 4b), and post-Meiyu II (Figure 4d) stages, HEP occurrence either slightly increases or decreases over eastern China from the first to the second 25-year period. HEP events occur more along coastal SC during the Meiyu stage. These phenomena demonstrate that change in the occurrence of HEP in the warm season from 1970-1994 to 1995-2019 is largely due to the changes in HEP occurrence during the post-Meiyu I and fall stages, especially in the post-Meiyu I stage (the national mean contribution is 66.7% in total HEP difference in the warm season), when partly associated with increased TC-induced precipitation ( Figure S6).

Conclusions
Using long-term measurements of hourly rainfall of 741 stations during 1970-2019 in eastern China and NCEP/NCAR reanalysis data, we examined how and why HEP has changed during the warm season in the past half-century. Our analysis found that climatologically, HEP occurrence varies between 7.20 h and 41.18 h across eastern China in the warm season. During the past 50 years, the occurrence of HEP has increased over NEC, LYR, and SC; such changes of HEP is largely due to changes in occurrence of S-HEP in the warm season.
We carry out SOMs analysis applied to HEP data to identify the timing, duration, and changes of these intraseasonal stages from 1970-1994 to 1995-2019 in order to reveal the HEP changes. Among the five intraseasonal stages of the EASM, the increase of HEP occurrence (total hours in the second 25-year period) mainly occurs in the post-Meiyu I stage when southerly winds dominate eastern China. Compared to 1970-1994, the duration of the post-Meiyu I stage becomes 12 days longer (1995-2019), leading the Meiyu front to stay in NEC and NC longer and S-HEP increase by 48.4%. SC and LYR are dominated by meridional circulation that brings more moisture and HEP events to the regions. These results suggest the relatively important roles of synoptic weather systems and circulation patterns played in the HEP events in recent years.
This work does not analyze the impact of aerosols on changes in HEP due to the absence of a long-term record of aerosols concentration, we will further study this using model simulations.

Data Availability Statement
Data on hourly precipitation were obtained from the National Meteorological Information Center of the China Meteorological Administration, the best-track data of tropical cyclone were obtained from the China Meteorological Administration (http://tcdata.typhoon.org.cn/zjljsjj_zlhq.html), and the global reanalysis data set was provided by NCAR (https://psl.noaa.gov/data/gridded/data.ncep.reanalysis.html). The authors used MATLAB to perform calculations, SOMs analyses, and visualization.