A Comparison of Different Station Data on Revealing the Characteristics of Extreme Hourly Precipitation Over Complex Terrain: The Case of Zhejiang, China

Both long‐term but small number of national stations and short‐term but large number of regional stations have been frequently used to study the extreme hourly precipitation (EXHP) in China. However, few studies focus on the differences of the two for revealing the features of EXHP. In this study, the characteristics of EXHP in Zhejiang Province are investigated using three rainfall data sets at three threshold criteria. The comparison between different data sets shows that increasing the station density can better reflect the climatic spatial distribution of EXHP thresholds if long‐term data is absent. The majority of EXHP can be classified into four weather types: the southwesterly wind type (30.7%–48.5%), the trough type (12.2%–23.6%), the tropical cyclone (TC) type (11.4%–17.5%) and the easterly wind type (4.9%–17.9%). The selection of stations is more sensitive to the proportions of the four weather types than the statistical years and threshold criteria. The monthly and diurnal variations of EXHP, as well as their differences revealed by the three data sets, are varied by weather type. Only using national stations cannot distinguish the spatial differences between the TC type and the easterly wind type, and there is an underestimation for southwesterly wind type and trough type in the mountainous area of southwest Zhejiang. The statistical year and station height are the main reasons for the differences in the duration of EXHP events calculated by different data sets, with the TC type having the largest effect and the southwesterly wind type the smallest.

With the availability of long-term (1951-present) hourly precipitation data based on national rain gauge observations, many aspects of the extreme hourly precipitation (EXHP) in China have been explored, including the definition of thresholds (Li et al., 2013a), the synoptic weather backgrounds (Luo et al., 2016;M. Wu et al., 2017), the diurnal variations (Zhang & Zhai, 2011;Zheng et al., 2019), the long-term trends (Fu et al., 2016;Xiao et al., 2016) and its possible link with urbanization (Liang & Ding, 2017;J. Wang et al., 2021;M. Wu et al., 2019). The seasonal and diurnal variations of EXHP in China varied among different regions. In general, the lower the latitude, the greater the span of seasons in which EXHP occurs (Li et al., 2013b). The diurnal variation of EXHP is similar to that of total precipitation (≥0.1 mm hr −1 ), which is mainly influenced by the large-scale circulations and local underlying surfaces (R. Yu et al., 2007;Yuan et al., 2012). The average duration of EXHP events is generally longer in the southeast coast of China and the Yangtze River Delta than that in northern China (Li et al., 2013b). The evolution of EXHP events is asymmetric, that is, precipitation peaks rapidly and then weaken slowly to an end (R. Yu et al., 2013). The spatial, seasonal, and diurnal variations of EXHP under different weather backgrounds are also different (Luo et al., 2016M. Wu et al., 2017). For example, in mainland China, EXHP events caused by tropical cyclones (TCs) account for more than 30% of the total cases over the southeast coast and show less preferred diurnal variation characteristics than those of other weather types. While EXHP of weak-synoptic forcing type is more dispersed and show a clear unimodal structure of diurnal variation.
Although the long-term hourly precipitation data provides us with great opportunities for understanding the characteristics of EXHP in China, there are only about 2,400 national stations in mainland China and the average distance between stations is about dozens of kilometers. Besides, most of the national stations are distributed in plain areas rather than mountainous regions where very intense rainfall has been frequently observed. It is reasonable to presume that the analysis results based on the few national stations may not be adequately representative of the EXHP in China. On the other hand, the number of regional automatic weather stations (AWSs) in China has significantly increased since 2005. The average distance between surface stations, therefore, has narrowed to a few kilometers (<10 km) in East China nowadays. However, compared to the national stations with several decades of records, the observation years of AWSs is relatively short. In Taiwan, C. Wu et al. (2016) have documented the differences between the two sources of rainfall data (long-term national stations vs. short-term AWSs) for revealing the main features of extreme rainfall caused by TCs, however, such studies are relatively lacking in mainland China.
Zhejiang Province is located on the east coast of China, which is jointly affected by the westerly and easterly weather systems. Extreme rainfall caused by TCs, Mei-Yu fronts, and convective storms is the main cause of severe meteorological disasters in Zhejiang. Besides, the complex underlying surface, including plain, hill, basin, mountain, and long coastline, makes the distribution of heavy precipitation in Zhejiang more complicated. Several previous studies used AWSs to analyze the spatiotemporal distribution of heavy hourly precipitation (≥20 mm hr −1 ) in Zhejiang during the recent 10 years. Key relevant findings include heavy hourly rainfall occurring more frequently in coastal areas than inland areas with a high-frequency center in southeast Zhejiang (P. Yu, 2022). The seasonal variation of heavy hourly rainfall is closely related to the active weather systems at different times (Tao et al., 2021). Both the urban agglomeration along the Hangzhou Bay and the mountainous areas in southern Zhejiang are high-frequency centers where subdaily heavy precipitation occurs in the afternoon (J. Lu et al., 2019;. However, most of these studies used a fixed threshold to define heavy hourly rainfall, which could hardly reflect the variability of the spatial distribution of hourly precipitation intensity under complex topography in Zhejiang. Meanwhile, few studies focus on the characteristics of extreme hourly rainfall under different weather backgrounds in Zhejiang.
In this study, the percentile method is used to define EXHP in Zhejiang. The characteristics of EXHP under different synoptic situations are thoroughly discussed. More importantly, this study compares the similarities and differences of using two sources of rainfall data, that is, long-term data from national stations and short-term data from AWSs, for revealing the characteristics of EXHP in Zhejiang. In Section 2, the thresholds of EXHP in Zhejiang using different station data, different criterion, and different methods are presented. The general features of EXHP under different weather types are described in Section 3. Section 4 compares the differences of different rainfall data sets for revealing the characteristics of EXHP. A summary and conclusions are given in Section 5.

Data Description
The construction of national stations in Zhejiang began in the 1950s (Figure 1a). The number of stations increased rapidly afterward and stabilized at about 67 stations in the early 1970s (but some stations were unavailable in the 1990s). The hourly rainfall data from these national stations receive a strict quality-control procedure, including a climatological limit value test, a station extreme value test, an internal consistency test, and a comparison with manually checked daily rainfall data (Luo et al., 2016). On the other hand, the number of AWSs in Zhejiang has shown a consistent increasing trend since 2005. Both data sets have been extensively used to investigate the characteristics of extreme precipitation in Zhejiang (e.g., Dong et al., 2019;Jiang et al., 2020;Luo et al., 2016). The analysis of the station heights shows that the national stations are mainly (>93%) located in the plain areas (≤200 m) and no station is located in high mountains (>1,000 m). Compared to national stations, the proportion of AWSs in hilly and mountainous areas (200-1,000 m) has increased by 4.5 times, and there are 19 stations located on high mountains (>1,000 m). For better comparison, three hourly precipitation data sets are formed in this study by extracting and integrating the two original data sources: (a) data from 60 national stations (red dots in Figure 1c) during 1979-2020 (referred to as 42yr-Nstations) with an average spacing of about 40 km; (b) data from the same 60 national stations but during 2011-2020 (referred to as 10yr-Nstations); (c) data from 1,537 surface stations (60 national stations + 1,477 AWSs; red dots and blue dots in Figure 1c) during 2011-2020 (referred to as 10yr-Astations) with an average spacing of about 8 km. Notably, only hourly rainfall data from April to October is used in this study. Percentages (%) of different station heights at 60 national stations (red columns) during 1979-2020 and at 1,477 regional stations (blue columns) during 2011-2020 in Zhejiang. (c) Spatial distribution of 60 national stations (red dots) during 1979-2020 and 1,477 regional stations (blue dots) during 2011-2020 in Zhejiang. Shading represents the topography. SH, JS, AH, and FJ represents Shanghai, Jiangsu, Anhui, and Fujian, respectively.

Definition of EXHP Thresholds
Using the percentile method, the threshold of EXHP at each station is defined according to the cumulative density function of hourly precipitation during a certain period time. To examine the variations of EXHP threshold intensity obtained using different years of hourly precipitation data, we used Equation 1 to calculate the relative percentage of threshold difference: where f is the relative percentage of threshold difference (%), T is the threshold value obtained using the hourly precipitation data during 1979-2020, and t is the threshold value obtained using hourly precipitation data for any year during 1979-2016 (e.g., the year of 2011) to 2020. Notably, t is calculated by default using a minimum of 5 years of hourly precipitation data (i.e., 2016-2020). Figure 2 shows the variation of value f with the length of the selected precipitation years for the 60 national stations at six threshold criteria, respectively. In general, the fewer years of precipitation data used, the higher the EXHP threshold, the larger the value of f. Comparing the statistical results of 2011-2020 with those of 1979-2020, the value of f remains below 30% from 99th to 99.95th percentile thresholds. However, it becomes much larger in the threshold criteria of 99.98th and 99.99th, with the minimum exceeding 30% and the maximum exceeding 100%. This indicates that the strongest hourly precipitation records vary widely from year to year The horizontal coordinate number (e.g., 2011) indicates using the hourly precipitation data from that year (e.g., year of 2011) to 2020, and the vertical coordinate is the relative percentage difference (f; %) between two EXHP thresholds obtained using precipitation data from that period (e.g., 2011-2020) and from 1979 to 2020, respectively. More details see Equation 1 in Section 2.2. and using short years of precipitation data may not be representative to define the most EXHP. Furthermore, a comparison of EXHP thresholds defined using the generalized extreme value distribution method (GEV) with those defined using the percentile method is made ( Figure 3). The GEV method could objectively extrapolate the extreme values and identify the intensity thresholds for rainfalls with different return periods based on a mathematical foundation (Li et al., 2013a;M. Wu et al., 2017;Zheng et al., 2016). The intensity of the 99.5th percentile threshold (19.3-27.5 mm hr −1 ) is slightly higher than the intensity of heavy hourly precipitation defined by China Meteorological Administration (20 mm hr −1 ). The intensity of the 99.9th percentile threshold (29.3-48.3 mm hr −1 ) is similar to the 2-year return value (28.6-45.0 mm hr −1 ) and the intensity of the 99.95th percentile threshold (34.1-59.1 mm hr −1 ) is comparable to that of the 5-year return value (35.6-61.2 mm hr −1 ). From the analysis above, three threshold criteria (i.e., 99.5th, 99.9th, and 99.95th percentiles) are used to select the EXHP records in this study.

Spatial Distribution of EXHP Thresholds
Using the three sets of hourly precipitation data (42yr-Nstations, 10yr-Nstations, and 10yr-Astations), the spatial distribution of EXHP threshold intensities under the three criteria (99.5th, 99.9th, and 99.95th) are given in Figure 4. The EXHP threshold from the station data are interpolated to a grid of 0.1° × 0.1° by using the inverse distance weighted method. Previous work suggest that the interpolation methods are insensitive to the results (Ikeda et al., 2010;Xu et al., 2017). The EXHP threshold in Zhejiang generally decreases from the coastlines to the inland area, with large threshold gradient belt locating 50-100 km away and almost parallel to the coastlines. in the southeast coast of Zhejiang is reflected in both 10yr-Nstations and 10yr-Astations, but is not highlighted in the results of 42yr-Nstations, indicating that EXHP in this region has increased significantly in the recent 10 years compared to the last 42 years. Furthermore, according to the standard deviation (STD) and the spearman correlation coefficient (Fieller et al., 1957), the spatial distributions of EXHP threshold (99.9th and 99.95th) of 10yr-Astations and 42yr-Nstations are more similar compared to those of 10yr-Nstations and 42yr-Nstations (Table 1). This indicates that the climatic characteristics of EXHP threshold in Zhejiang can be reflected to some extent by significantly increasing the station density despite the relatively short observation period.

Classification of EXHP
The large-scale circulations of EXHP are classified into several types. The TC type is first identified as any EXHP record occurs when a TC is influencing Zhejiang. The start and end of TCs affecting Zhejiang during 1979-2020 is determined by the comprehensive analysis of Shanghai Typhoon Institute (cdata.typhoon.org.cn; (d-f) 60 national stations during 2011-2020; (g-i) 1,537 all stations (national + regional) during 2011-2020.The left column (a, d, g) is for the threshold of 99.5th percentile, the middle column (b, e, h) is for the threshold of 99.9th percentile, and the right column is for the threshold of 99.95th percentile. Note, the color bar is the same for each column. Ying et al., 2014;X. Lu et al., 2021) and Zhejiang Meteorological Observatory. The remaining EXHP records are further classified using the obliquely rotated principal component analysis (PCA) in T mode (PCA-T hereinafter; Huth, 1996Huth, , 2000. This method organizes the input data matrix with grid points of each sample (circulation pattern) in rows and samples in columns, to find typical synoptic patterns more efficiently . Previous studies have suggested a reasonably good performance of the PCA-T in the synoptic-pattern classification in China (Bai et al., 2021;Chen et al., 2022;Liu et al., 2022). The fifth generation of the European Centre for Medium-Range Weather Forecasts atmospheric reanalysis (ERA5) data are used as the input data for the weather type classification analysis, which are hourly available at a horizontal resolution of 0.25° (Hersbach et al., 2020).
As the majority (65.1%-74.9%) of the EXHP records occurred during 14-23 local solar time (LST) and the weather pattern prior to the occurrence of EXHP may be more indicative, the geopotential height at 850 hPa at 14 LST on the day that EXHP occurred is selected as the objective typing element.

EXHP of Four Weather Types
Although the number of EXHP days varied among different sets of precipitation data and thresholds, four weather types that are common and EXHP belong to these four types accounted for an absolute proportion of total records (83.1%-95.2%). Taking the classification results of EXHP using the 42yr-Nstations at the 99.5th percentile as an example ( Figure 5), for the TC type, the location of TC centers and the movement path of TCs during the occurrence of EXHP are indicated accordingly (Figure 5a). For the first non-TC-type (Figure 5b), the superimposed mean wind field and water vapor flux at 850 hPa show that Zhejiang is mainly influenced by consistent southwesterly winds, and the area with large values of water vapor flux is located in western Zhejiang and the area from the Taiwan Strait to the East China Sea. This type is referred to as the southwesterly wind type in the later analysis. In the second non-TC-type (Figure 5c), there is a northeast-southwest composite trough in the north-central part of Zhejiang, and the trough line extends southward to affect Zhejiang with a large value zone of water vapor flux ahead. This type is abbreviated as the trough type. In the third non-TC-type (Figure 5d), Zhejiang is influenced by the easterly mean wind and the water vapor flux in Zhejiang decreases from the coastline to the inland. This type is abbreviated as the easterly wind type.
The percentages of EXHP under the four weather types (TC type, southwesterly wind type, trough type, and easterly wind type) are shown in Figure 6. The TC type accounts for 11.4%-17.5% of the total records, and its proportion decreases slightly with the increase of the threshold criterion. The southwesterly wind type is the most dominant weather type, and EXHP of this type ranges from 30.7% to 48.5%. It is worth noting that with the increase of the threshold criterion, the proportion of this type increases correspondingly when only using the 60 national stations (42yr-Nstations and 10yr-Nstations; Figures 6a-6f), while this feature is reversed in the 10yr-Astations statistics (Figures 6g-6i). The second most common type is the trough type, which has a relatively stable share (20.0%-23.6%) when only national station data are used, while its share increases from 12.2% to 27.3% with the threshold criterion in the 10yr-Astations statistics (Figures 6g-6i). The percentage of the easterly wind type is equal to that of the TC type (10.0%-17.9%), except for a small percentage (4.9%) at the 99.5th percentile threshold of the 42yr-Nstations (Figure 6a). The classification of the remaining weather types may vary under different data sets and threshold criteria. But since their total proportions are relatively small (4.8%-16.9%), they are collectively referred to as no tropical cyclone Others in this study and will not be discussed in detail.
Comparing the proportions of the four types obtained by the three rainfall data sets, it is found that the results of the 10yr-Nstations (Figures 6d-6f) are more like those of the 42yr-Nstations (Figures 6a-6c) -6i), indicating that the selection of stations is more sensitive to the proportion of each type than the statistical years.

Evolutionary Characteristics of Four Weather Types
According to the variation of hourly precipitation intensity in the 3 hr before the occurrence of EXHP, EXHP events are further classified into four major categories, namely, abrupt, growing, continuous, and other types ( Table 2). For the abrupt type (Figure 7a), the hourly precipitation amount in the hour before the occurrence of EXHP (peak) does not exceed 10% of the peak amount, and the hourly precipitation intensity decreases the fastest among these four categories after the peak. For the growing type (Figure 7b), its evolution is quite symmetrical with a process of gradual growth and weakening before and after the peak. For the continuous type (Figure 7c), the hourly precipitation intensity in the 1 hr before the peak is usually greater than the peak (>100%), that is, a situation where two consecutive hours of precipitation reach the EXHP intensity. Those that do not belong to the first three types are collectively referred to as the other type, which has a process of increasing (but not exceeding the peak) and then weakening before the occurrence of EXHP (Figure 7d).
For the three rainfall data sets under the three threshold criteria, it is found that both abrupt and growing types are the main types (Figures 8a and 8b), accounting for 43.6%-53.3% and 42.8%-46.8% of the total records, respectively. The continuous type accounts for less than 10% of the total records, and its proportion decreases as threshold criteria increases (Figure 8c). The other type is the least to occur (Figure 8d), accounting for less than 3% of the total records. A comparison of the percentages of abrupt and growing types of EXHP under four weather types shows that the TC type is dominated by the growing type, followed by the easterly wind type, the trough type, while the southwesterly wind type occurs more abruptly (Figure 9). And the higher the threshold criterion, the more pronounced this feature is. The percentage (%) of EXHP under each type is also marked on the side, and the gray circles indicate the 5%-50% frequency scale from the inside to the outside, respectively.

The Monthly and Diurnal Variations
To compare the similarities and differences of the three rainfall data sets for revealing the characteristics of EXHP under four weather types, EXHP under the 99.5th percentile threshold is used for further analysis. The semimonthly and diurnal variations of the occurrence frequency of EXHP are first analyzed ( Figure 10). For the convenience of comparison, the occurrence frequency of EXHP is normalized using Equation 2: Temporal evolution type Hourly rainfall rates in 3 hr prior to the extreme rainfall hour (R −1 , R −2 , R −3 , respectively) compared with the extreme hourly precipitation (R 0 ) Abrupt type R −1 < 0.1R 0 and R −2 < 0.1R 0 and R −3 < 0.1R 0 Growing type (R −1 > R −2 or R −1 > R −3 或 R −2 > R −3 ) and at least one of (R −1 , R −2 , R −3 ) > 10% but < R 0 Continuous type At least one of (R −1 , R −2 , R −3 ) > R 0 Other type Not belong to the above three types Table 2 Classification where Z′ is the normalized variable, Z is the original variable, and Z min /Z max is the minimum/maximum value in the array, respectively.

of Extreme Hourly Precipitation Events According to the Temporal Evolution of Hourly Precipitation in the 3 hr Prior to the Hourly Extreme
For the TC type, all the three data sets show that EXHP peaks in early August (August 1-August 15) ( Figure 10a), while its proportion in the 10-year statistics (10yr-Nstations and 10yr-Astations) is significantly lower than in the 42-year statistics during August 15-September 15. In terms of daily variation (Figure 10e), although the TC type has less preferred characteristics compared to the other weather types, still a relatively higher proportion occurs in both the early morning (02-06 LST) and late afternoon (16-18 LST). The late afternoon peak is more pronounced in the 42-year analysis, while the two peaks are comparable in the 10-year statistics. For the southwesterly wind type (Figures 10b and 10f), all the three data sets reflect similar characteristics in semimonthly and diurnal variations, that is, a single-peak structure, peaking in late June (June 15-June 30) and around evening (16-18 LST), respectively. For the trough type, its semimonthly variation shows a bimodal structure in late June (June 15-June 30) and late August (August 15-August 30), with the most pronounced bimodal structure for the 10yr-Astations. These two peaks coincide with the period when the East Asian summer monsoon is in the process of northward and southward movement passing through Zhejiang, and the cold and warm air are confronting each other in this region. The diurnal variation also shows a bimodal structure (Figure 10g). The major peak occurs around 18 LST when using only 60 national stations, and 2 hr earlier (16 LST) in 10yr-Astations. The secondary peak is around early morning (06-08 LST) and is most obvious in 10yr-Astations, followed by 10yr-Nstations, and least obvious in the 42yr-Nstations. This indicates that EXHP of trough type occurred in the early morning has increased in the recent 10 years compared with the past 42 years, and the early morning peak is better captured by encrypting stations. For the easterly wind type, the seasonal variation (Figure 10d) shows that the peak occurs in late August (August 15-August 30). Its occurrence frequency is significantly overestimated in September and underestimated in June-July if only the national stations are used. In terms of the diurnal variation (Figure 10h), this type shows similar characteristics to the southwesterly wind type, that is, a consistent single-peaked structure (peaking around 18 LST).
For the TC type, although all three data sets reflect that the occurrence frequency decreases from the coastlines to inland areas (J. Wang et al., 2018), by encrypting the stations, the high-frequency center (>75%) is found to be located at the mountainous windward slope, about 10-30 km inward from the coastline (Figure 11c), rather than along the coastline when only national stations are used (Figures 11a and 11b). Previous studies have found that the windward slope effect of coastal topography (Ji et al., 2007) and the interaction of TCs with high-altitude mountains (C. Wu et al., 2002;Yang et al., 2014) are conducive to the occurrence of EXHP. For the southwesterly wind type, the high-frequency center is located in northwest Zhejiang when only the national stations are used (Figures 11d and 11e), while the 10yr-Astations shows that this type occurs frequently throughout the whole western Zhejiang (Figure 11f). Therefore, only using national stations could significantly underestimate this type in southwest Zhejiang. This is probably because most of the national stations are built on plain areas (≤200 m), which could not reflect the interaction between low-level air flow and mountains. However, local convection is easily stimulated when low-level water vapor is transported by southwesterly winds and blocked, crossed, or bypassed by mountains here (Bai et al., 2021;Chow et al., 2012;Houze, 2012). For the trough type, the occurrence frequency of EXHP is also underestimated in southwest Zhejiang but overestimated in north central Zhejiang when only national stations are used (Figures 11g-11i). For the easterly wind type (Figures 11j-11l), the high-frequency center is along the coastlines, especially along the Hangzhou Bay and the southeast coast. And only using national stations is difficult to distinguish between the easterly wind type and TC type in terms of spatial distribution. Figure 9. Distribution of the percentage (%) of extreme hourly precipitation for the abrupt type (horizontal coordinate) and growing type (vertical coordinate), with four symbols representing four weather types (tropical cyclone type, southwesterly wind type, trough type, and easterly wind type), and different colors as in Figure 8, representing the statistical results using different hourly rainfall data sets and thresholds.
From the analysis above, it is clear that the differences in the spatial distribution of EXHP revealed by dense stations and sparse national stations are mainly due to the different topographic heights at which the stations are deployed. If we further analyze the proportion of station heights in the top 25% of EXHP of 10yr-Astations (red dots in the third column of Figure 11), it can be found that the proportion (46.5%-50.5%) of these high-frequency stations located at certain topography (>200 m) is much higher than the original proportion (30.8%) of station deployments ( Figure 12). This once again illustrates the importance of placing automatic stations at certain topography to reveal the spatial distribution of EXHP. The exception is the easterly wind type, which has an even slightly higher proportion (73.1%) of high-frequency stations located in the plains (0-200 m). This probably because the easterly wind type mainly occurs along the coast (especially near the Hangzhou Bay), where the topography in Zhejiang is flatter.

The Duration of EXHP Events
The EXHP events is defined as the continuous rainfall process (≥0.1 mm hr −1 ) with at least one EXHP record at a station. And the duration of EXHP event is defined as the continuous rainfall with at most 1-hr intermittence (Li et al., 2013b). Despite the use of different data sets and thresholds, about 63.2%-77.0% (80.2%-90.5%) of EXHP events occur within 12 (24) hr, and the higher the threshold criterion, the shorter the duration of EXHP events. Comparing the cumulative distribution function (CDF) curves for the 10yr-Nstations and the 10yr-Astations, the duration of EXHP events obtained by national stations is generally shorter than that by all stations (Figure 13a). Since the national stations are mainly located in the plains (0-200 m), while a larger proportion of the AWSs are built in places with some topography (>200 m), the 10yr-Astations are further divided into plain stations (0-200 m; solid line in Figure 13b) and mountain stations (>200 m; dashed line in Figure 13b). It is found that the TC type is most sensitive to station height, and the duration of EXHP events is significantly longer for mountain  Figure 11. (a) Spatial distribution of extreme hourly precipitation of tropical cyclone type at the 99.5th percentile threshold of 42yr-Nstations, with red, orange, dark blue, and light blue dots for stations with occurrence frequencies of 75%-100%, 50%-75%, 25%-50% and 0%-25%, respectively; (b) the same as (a), but of 10yr-Nstations; (c) same as (a), but of 10yr-Astations. (d-f) Same as (a-c), but for the southwesterly wind type; (g-i) same as (a-c), but for the trough type; (j-l) same as (a-c), but for the easterly wind type.
stations than for plain stations. M. Wu et al. (2017) found a similar feature in their study of TC-induced extreme rainfall in Taiwan, that is, the duration of TC precipitation events increases with the station height. The second most sensitive type is the easterly wind type. The proportion of EXHP events maintained within 1-12 hr is lower in mountain stations than in plain stations but higher after 24 hr. The duration of trough type and southwesterly wind type have the lowest sensitivity to station height. Comparing the two data sets of national stations, the duration of EXHP events is generally shorter for the 42yr-Nstations compared to the 10yr-Nstations (Figure 13a). To further analyze the possible reasons for this difference, thresholds obtained from the 42yr-Nstations (solid line in Figure 13c) and from the 10yr-Nstations (dashed line in Figure 13c) are used to select EXHP during 2011-2020, respectively. Results show that the CDF curves using these two kinds of threshold are quite similar, indicating that the thresholds obtained from different years of data has no significant effect on this difference (Figure 13c). On the other hand, if the 99.5 percentile thresholds of the 42yr-Nstations are used, but EXHP are divided into two periods, that is, 2011-2020 and 1979-2010 (Figure 13d), the differences are highlighted. The largest difference is for the TC type, followed by the easterly wind type, the trough type, and the smallest difference is for the southwesterly wind type. This comparison (Figure 13c vs. 13d) illustrates that using the same stations, the differences in duration results are more sensitive to statistical years than to using different thresholds (calculated based on different years), especially for EXHP events influenced by strong weather systems (e.g., TC type).

Summary and Discussion
In this study, the characteristics of EXHP under four weather types in Zhejiang are examined based on three sets of hourly precipitation data (42yr-Nstations, 10yr-Nstations, and 10yr-Astations) at three threshold criteria (99.5th, 99.9th, and 99.95th). While revealing the features common to the EXHP in Zhejiang, the differences of Figure 12. The proportion of 10yr-Astations with different heights, tropical cyclone (TC) represents the stations in the top 25% occurrence frequency of TC type of 10yr-Astations using the 99.5th percentile threshold (red dots in Figure 11c), SW_wind, Trough, and E_wind is similar to TC, but for the southwesterly wind type, trough type, and easterly wind type, respectively. using three rainfall data sets for revealing the characteristics of EXHP are also compared. The relevant findings are mainly as follows.
1. The thresholds of EXHP in Zhejiang generally decrease from coastlines to inland areas and the large threshold gradient is located about 50-100 km away from the coast. Compared to the 10yr-Nstations, the spatial distribution characteristics of EXHP thresholds obtained by 10yr-Astations are closer to the 42-year statistics, suggesting that the climatic spatial distribution of EXHP threshold can be reflected in a certain extent by significantly increasing the station density if long-term precipitation data is absent. 2. Regardless of the rainfall data sets and threshold criteria used, four weather types can be classified in which the majority of EXHP occurs in Zhejiang: the southwesterly wind type (30.7%-48.5%), the trough type (12.2%-23.6%), the TC type (11.4%-17.5%), and the easterly wind type (4.9%-17.9%). Among the four types, the selection of stations is more sensitive to the proportions of the four types than the statistical years and threshold criteria. 3. The monthly and diurnal variations of EXHP in Zhejiang are varied by weather types. Comparing the three rainfall data sets for revealing the differences in the four weather types, the smallest difference is for the southwesterly wind type, the difference for the TC type mainly comes from the selection of statistical years, and the difference for the trough type and the easterly wind type mainly comes from the station selection. 4. The sparse national stations are mostly located in the plains while Zhejiang is mountainous. When only national stations are used, the high frequency centers of EXHP are biased in each weather type. The high-frequency center of TC type is identified on the coastlines rather than on the windward slope about 10-30 km from the coastline. And its difference with the easterly wind type cannot be distinguished. Besides, the southwesterly wind type is significantly underestimated in southwest mountainous area of Zhejiang. The trough type is also underestimated in southwest Zhejiang but overestimated in north central Zhejiang. 5. The majority (63.2%-77%) of the EXHP events occur within 12 hr. Among the four weather types, the TC type has the longest duration and is dominated by the growing type, while the southwesterly wind type has the shortest duration and is dominated by the abrupt type. The statistical year and station height are the main reasons for the differences in the duration of EXHP events calculated by different rainfall data sets, with the TC type having the largest effect and the southwesterly wind type the smallest.
The above results show that for EXHP events influenced by strong weather systems, such as the TC type, extending the statistical year and increasing the observation density are both important to reveal its regional climatic characteristics. While for weak synoptic-scale forcing events, such as the southwesterly wind type, increasing the mountain stations is critical while other factors are less important. Although this paper provides a comprehensive overview of the climatic characteristics of EXHP in Zhejiang over the past half century, there is still much to study further. For example, the observation data used in this paper is limited to station data and the research method stays in the basic statistical analysis. Based on the existing research, more observational data (e.g., radar and satellite observations) and numerical simulations will be used in the future study to specifically examine the triggering mechanism of short-term extreme precipitation, the fine three-dimensional structure of the convective system, the multi-scale system interactions, and the effect of local underlying surface (topography, large urban clusters and land-sea contrast, etc.) for each weather type. All this work will improve the understanding of the fine characteristics and physical mechanisms of EXHP in Zhejiang, and provide a scientific support for local shortterm heavy precipitation forecasting services and disaster prevention.

Data Availability Statement
The ERA5 data provided by the European Centre for Medium-Range Weather Forecasts (https://cds.climate. copernicus.eu/cdsapp#!/home) and two kinds of observational data at China Meteorological Administration (CMA) (http://data.cma.cn/) are used in this study. The terrain data used is available from https://www.usgs.gov/ centers/eros/science/usgs-eros-archive-digital-elevation-global-30-arc-second-elevation-gtopo30. To conduct the PCA-T analysis, the open source software package "cost733class" within the framework of COST Action 733 "Harmonisation and Applications of Weather Type Classifications for European Regions" (Philipp et al., 2016) was used. Figures were plotted with the Interactive Data Language (IDL) (http://www.idlworld.com). The observational data, terrain data, and the methods/scripts/code used in the paper can be downloaded from https://doi. org/10.5281/ZENODO.7758469.