Spatiotemporal Variations of Extreme Precipitation Events in the Jinsha River Basin, Southwestern China

Climate extremes have attracted widespread attention for their threats to the natural environment and human society. Based on gauged daily precipitation from 1963 to 2016 in four subregions of the Jinsha River Basin (JRB), four extreme precipitation indices developed by the Expert Team on Climate Change Detection and Indices (ETCCDI) were employed to assess the spatiotemporal variations of extreme precipitation events. Results show the following: (1) Max one-day precipitation amount (RX1day), max consecutive five-day precipitation amount (RX5day), precipitation on very wet days (R95p), and number of heavy precipitation days (R10mm) showed increasing trends in four subregions except for the decline of R10mm in the southeastern and RX5day in the midsouthern. Extreme precipitation has become more intense and frequent in most parts of the JRB. (2) In space, the four extreme precipitation indices increased from the northwest to the southeast. Temporal trends of extreme precipitation showed great spatial variability. It is notable that extreme precipitation increased apparently in higher elevation areas. (3) +e abrupt change of extreme precipitation in the northwestern, midsouthern, and southeastern mainly appeared in the late 1990s and the 2000s. For the midnorthern, abrupt change mainly occurred in the late 1980s.+is study is meaningful for regional climate change acquaintance and disaster prevention in the JRB.


Introduction
Extreme precipitation events are important indicators to measure intensity and frequency of precipitation which are very sensitive to climate change [1]. Global warming, increasing atmospheric moisture content, and surface evapotranspiration accelerate the water cycle and enhance extreme precipitation events [2][3][4][5][6]. It is very important for the ecological environment, the lives and property of people, and the national economy to understand the changing law of extreme precipitation events [7][8][9]. e occurrence of extreme precipitation may cause disaster for the people in a region. In recent years, the occurrence of quite a few extreme precipitation events in China has caused serious casualties and property losses. For example, in 1998, the extreme floods in the Yangtze River basin caused by extreme precipitation events claimed thousands of lives and caused billions of CNY in economic loss. On June 24, 2017, a catastrophic landslide destroyed the Xinmo village in Sichuan province under the influence of prolonged extreme precipitation, and 83 deaths were attributed to this landslide activity [10,11].
Global extreme precipitation had a wetter tendency throughout the 20th century [12]. Griffiths and Bradley found increased precipitation extremes in North America during 1926-2000 [13]. Extreme precipitation events presented great regional differences in South America due to the influence of large scale events [6]. In Asia, extreme daily precipitation events increased in both frequency and intensity in the Philippines during 1951-2010 [14]. Over the past three decades, precipitation showed a downward trend across India [15], but extreme precipitation events increased in some regions [16,17]. Almazroui and Saeed found that the accumulation of simulated precipitation related to extreme events (RR > 99th percentile) had an increasing trend and the contribution of extreme precipitation events to the accumulated rainfall enhanced over the Arabian Peninsula [18]. Similar to the global trend, extreme precipitation events in China are increasing and have great regional differences [19][20][21][22][23][24]. It is necessary to investigate the spatial and temporal variations of extreme precipitation events at regional scale.
As the uppermost part of the Yangtze River Basin, the Jinsha River Basin (JRB) is characterized by complex terrain, fragile ecosystems, and underdevelopment [25]. is region is an important zone for water resources and hydropower development, while climatic variability has greatly influenced the socioeconomic development and ecosystem [26][27][28][29]. Previous studies observed that precipitation in this region is distributed unevenly in both the spatial and temporal scale. Wang et al. found that the precipitation in the JRB increased slightly and insignificantly and the temperature showed a tendency to warmer [30][31][32]. An increase trend of continuous multiday extreme precipitation was found under 20-and 50-year return period over the JRB by the GCMs during 2020-2050 [27]. Nevertheless, most of previous studies only focus on annual and seasonal characteristics of precipitation in the JRB and regional extreme precipitation changes in the Yangtze River basin [33,34]; few studies have regarded the entirety of JRB as an investigated area to comprehensively analyze the spatiotemporal variations of extreme precipitation events.
is study investigated the spatiotemporal variations of extreme precipitation events in the JRB based on daily precipitation during 1963-2016 and provided a detailed assessment of observed extreme precipitation. Such a study will contribute to providing a scientific reference for reducing the risk of disaster in the JRB and regional responses of precipitation extremes to global climate change. e remainder of this paper is structured as follows: the study area and daily precipitation data are described in Section 2; Section 3 presents analysis techniques and methods; obtained results are presented in Section 4; the discussion and conclusions are given in Section 5.
e JRB is mainly covered by the Qinghai-Tibet Plateau, the Yunnan-Guizhou Plateau, and the Sichuan Basin.
e elevation increases from about 249 m in the southeast to 6575 m in the northwest. e JRB encompasses diverse climate patterns, including many types of monsoons and a typical plateau climate. e spatiotemporal variations of precipitation present unique regional characteristics due to the great differences in local altitude and climate conditions.

Data and Indices of Extreme Precipitation Events.
e daily precipitation data of 24 gauged stations within the JRB from 1963 to 2016 was provided by China Meteorological Data Service Center managed by National Meteorological Information Center (http://data.cma.cn/). e data had been subjected to quality control procedures of the China Meteorological Data Service Center. e linear regression method was used to interpolate values for the little missing data from adjacent stations. Some information on meteorological stations is listed in Table 1. Meteorological stations are distributed over the basin as shown in Figure 1 and can describe precipitation characteristics in space. Subregional average precipitation data was obtained by the arithmetic mean method. e precipitation time series spanning 54 years    [35][36][37][38][39]. Four indices were employed to comprehensively analyze extreme precipitation events in this study (Table 2). Max one-day precipitation amount (RX1day) and max consecutive five-day precipitation amount (RX5day) were two important indicators for flood producing [40]. Precipitation on very wet days (R95p) is a percentile-based indicator and accounts for 20-40% of annual total precipitation in all stations of the JRB. RX1day, RX5day, and R95p were three intensity indicators. Number of heavy precipitation days (R10mm) was selected to analyze the frequency of extreme precipitation.

Rotated Empirical Orthogonal Function Analysis.
Empirical orthogonal function (EOF) analysis is a decomposition of a signal or dataset based on orthogonal basis functions determined from the data. It is an efficient method to split the spatial-temporal field into both space and time and is widely used in the study of meteorology and climatology [33,41,42]. e output of the EOF method includes the time coefficient (PCs), spatial pattern (EOFs), eigenvalues, and others. However, the obtained patterns are not physically meaningful due to orthogonality in both space and time. To overcome the limitation of EOF, the VARIMAX method was used to obtain rotated EOFs [43,44]. Rotated empirical orthogonal function (REOF) relaxes the orthogonality of the spatial patterns and yields localized structures. REOF was good at dividing climatic patterns [45]. In this study, regionalization was obtained by the distribution of loadings.

Estimation of Tendency.
e MK test is a distributionfree rank-based method [46,47] and utilized routinely in trend detection in hydrological and meteorological time series [16,48,49]. However, some time series have serial correlation, which may lead to incorrect conclusions for MK tests. us, trend free prewhitening MK (TFPW-MK) was  employed to remove correlation [50,51]. If there is a linear trend in hydroclimatic time series, the magnitude of the trend can be estimated using nonparametric Sen's slope method [52][53][54].

Ordered Clustering
Method. e ordered clustering method is an effective method to estimate the most likely abrupt point of time series by statistical analysis [56]. e basic idea of this method is to find a time point to divide the time series into two subsequences and to make the sum of deviation squared of the subsequences minimum and the sum of deviation squared between subsequences maximum [57].

Pettitt Test.
Pettitt test is a nonparametric method which does not need to consider the distribution type of variables. It is based on the Mann-Whitney two-sample test.
is method can detect whether there is a change point and find the location of the change point [58]. If the value of approximate probability, p, is less than 0.05, then the change point is considered statistically significant.

Mann-Kendall Mutation Test.
e Mann-Kendall mutation method seems to be an appropriate method for analyzing abrupt change and has been commonly used in time series such as streamflow, precipitation, and water quality. If curves of the two statistics of the Mann-Kendall mutation test (UF k and UD k ) have cross points and the points occur within the confidence interval, the cross points can be viewed as significant abrupt change points [59].

Precipitation Features and Regionalization.
Impacted by monsoon activities and landform, the precipitation in the JRB is unevenly distributed, which gradually increased from the northwest to the southeast. Multiyear average annual precipitation ranged from 293 to 1135 mm for the 24 stations. A one-way analysis of variance (one-way ANOVA) test was applied to compare the annual total precipitation (ATP) in all stations. e results showed that the difference was significant (p < 0.001). e study on regionalization of precipitation and extremes can help to furtherly understand the space differences. To our knowledge, no regionalization study specifically for precipitation in the Jinsha River Basin has been done yet.
Considering the close correlations between extreme precipitation indices and annual total precipitation (Table 3), the EOF and REOF analyses were applied to partition the study area according to the annual total precipitation. e first 7 EOFs together represented 80% of the total variance, and the first 4 EOFs together represented 68% of the total variance (Table 4). e fact that the first 4 EOFs captured more than 60% of the total variability indicated that the complexity of the spatial pattern of precipitation over the Jinsha River Basin can largely be explained by a small number of spatial structures. e "scree test" was used to determine the number of significant orthogonal functions [60]. e first four eigenvalues were well separated from the rest. e first four EOFs were retained to be rotated using VARIMAX method. e obtained spatial field was handled by the inverse distance weighting interpolation with the GIS technique. Figure 2 shows the distributions of the field of the first four REOFs for the annual total precipitation in the JRB. Based on distributions of these high loading eigenvectors (the absolute values of the contours equal to or larger than   Advances in Meteorology 0.5), four subregions areas for variations of precipitation in the JRB were identified (Figure 1).

Temporal Variation of Extreme Precipitation
Events. e temporal variations of extreme precipitation indices from the TFPW-MK test and Sen's slope in the four regions are summarized in Table 5. Max one-day precipitation amount (RX1day) and precipitation on very wet days (R95p) increased in all four regions of which RX1day in region IV (the southeastern) and R95p in region I (the northwestern) reached a significant level (α � 0.05). e magnitude of the trend in RX1day increased from the upper basin to the lower        basin. is may be related to the uneven precipitation in the JRB which increased gradually from the northwest to the southeast. R95p increased much more in regional IV than the other three regions; R95p in region II (the midnorthern) just showed a slightly increasing trend at a rate of 0.35 mm/ decade. Max consecutive five-day precipitation amount (RX5day) increased nonsignificantly in regions I, II, and IV and decreased nonsignificantly in region III (the midsouthern). Number of heavy precipitation days (R10mm) showed increasing trends in regions I, II, and III with slopes of about 0.55, 0.12, and 0.22 d/decade, respectively, of which region I reached a significant level (α � 0.05). Decreasing trend was detected for R10mm in region IV.

Advances in Meteorology
In addition, percentages of stations that showed increasing and decreasing trends in each extreme precipitation index in the JRB were calculated and plotted in Figure 3. Most of the stations experienced increasing in RX1day, R95p, and R10mm though few trends were statistically significant. About 63% of stations had increasing trends for RX1day, and this index owned the largest significance as 17% of stations experienced a significant increase. For R95p, 71% of stations had increasing trends, but only about 13% of these trends were statistically significant; 29% of stations had decreasing trends for R95p. R95p owned the largest percentage of stations which showed increasing trends. 58% of stations showed increasing trends for R10mm with only 8% being significant, and 42% of stations showed decreasing trends in this index. ese indicated that extreme precipitation events have become more intense and frequent over the study period. As for RX5day, the percentages of stations which showed increasing trends and decreasing trends were awfully equal.

Extreme Precipitation Indices.
e spatial distributions of the four indices are illustrated in Figure 4. RX1day, RX5day, R95p, and R10mm shared a similar spatial pattern, gradually increasing from northwest to southeast (Figures 4(a)-4(d)). For instance, the annual values of RX1day varied from 20.4 to 82.0 mm with lower values in the upper basin and higher values in the lower basin, which was also detected in a previous study [61]. e mean value of each extreme precipitation index in each region was calculated (Table 6). For all four indices, the largest mean values emerged in region IV and the smallest mean values appeared in region I.

Decadal Trends of Extreme Precipitation Indices.
e TFPW-MK test and Sen's slope were employed to investigate the spatial variations of each index's changing trend of the individual stations. Figure 5 gives the spatial distributions of decadal trends of extreme precipitation indices over the JRB. About 63% of the stations exhibited increasing trends in RX1day with four statistically significant stations (α � 0.05) in the lower basin ( Figure 5(a)). e change trend of RX5day was spatially heterogeneous (Figure 5(b)).
ere are some regional differences in the spatial distribution of the R95p and R10mm change trends. For R95p, the upper and lower basin     Advances in Meteorology were dominated by an increasing trend of which three stations were statistically significant ( Figure 5(c)). In the middle basin, the number of stations showing increasing trends was almost equal to stations with decreasing trends. For R10mm, the upper and middle basin were dominated by an increasing trend, while the lower basin was dominated by a decreasing trend. In sum, trends of extreme precipitation indices showed great spatial variability across the JRB. Most region of the JRB showed increasing in RX1, R90p, and R10mm.

Abrupt Change Analysis.
Abrupt change analysis of extreme precipitation events was conducted by the ordered clustering analysis, Pettit test, and Mann-Kendall mutation test. Figure 6 shows the diagnosed graphs of the ordered clustering analysis and Mann-Kendall mutation test in  Table 7 shows the results of abrupt change analysis for the time series of extreme precipitation indices in the four subregions. From Table 7, it can be seen that the abrupt change of extreme precipitation indices in region I mainly appeared in 2007. In region II, RX1day abruptly changed in 1978 and RX5day and R95p abruptly changed in 1989; there were no abrupt change points for RX10mm. For region III, abrupt changes of extreme precipitation indices were more likely to occur around 1984, 1997, and 2002. For region IV, RX1day and R95p abruptly changed in 1997; RX5day and R10mm made no abrupt change points in this region.

Discussion and Conclusions
Humans and natural environment are easily affected by climate extremes. e influence of climate extremes on the society and natural environment is increasing under the global warming background [7,25,62,63]. e Jinsha River Basin is highly sensitive to climate change due to its geographic, ecological, and social background. is study investigated the spatiotemporal variations of four extreme precipitation indices such as max one-day precipitation amount (RX1day), max consecutive five-day precipitation amount (RX5day), precipitation on very wet days (R95p), and number of heavy precipitation days (R10mm) in the Jinsha River Basin. It would be helpful for evaluating  Advances in Meteorology changes of extreme precipitation at the regional scale and the management and mitigation of natural hazards. Based on daily precipitation records from 24 meteorological stations during 1963-2016, the Jinsha River Basin was divided into four subregions using rotated empirical orthogonal function analysis (REOF) and four extreme precipitation indices were selected to analyze the spatiotemporal variations of extreme precipitation events in the Jinsha River Basin in southwestern China. Results indicated that RX1day, RX5day, R95p, and R10mm had increasing trends in four subregions except that R10mm decreased in the southeastern (region IV) and RX5day decreased in the midsouthern (region III). Increasing in RX1day and RX5day may increase the risk of regional floods as these two indices largely represent the intense precipitation events occurring in a very short time [40,64].
e Jinsha River Basin located in middle-high altitude areas has a high risk of suffering summer flash floods due to extreme precipitation, which often occurs with secondary disasters such as landslides, debris flow, and erosion [65][66][67]. e JRB is an area of a high incidence of debris flow; about 13% of debris flow disasters in China occurred here [29]. Since the values of extreme precipitation indices in the JRB were low, their absolute increasing amount was not big. Increase of RX1day in the southeastern and increase of R95p and R10mm in the northwestern were statistically significant (α � 0.05).
us, extreme precipitation in the northeastern (region I) and midnorthern (region II) increased in both the intensity and frequency along with a strong warming trend in temperature extremes [32]. For the midsouthern, extreme precipitation became more frequent as R10mm increased in this region. Extreme precipitation increased in intensity but decreased in frequency in the southeastern. General increasing trends of extreme precipitation events in the JRB were also reported in previous studies [19,33,61]. Precipitation and snowmelt are the main source of water resource in the Jinsha River Basin. Changes in extreme precipitation may influence water management in this region [68].
All extreme precipitation indices shared a similar spatial pattern with an increasing trend from the northwest to the southeast. e interaction of intricate topography and complicated climate plays an essential role in determining the distribution of extreme precipitation events over this region [69]. Decadal trends of extreme precipitation indices demonstrated large spatial variability across the JRB, but extreme precipitation increased apparently in intensity and frequency in higher elevation areas. e increase in extreme precipitation in higher elevations was due to the cyclonic activity in the north of the Indian subcontinent which brings moisture to the Qinghai-Tibet Plateau combined with the airflow lifted by topography [70]. Recent studies have indicated that changes in extreme precipitation are related to elevation [38,71]. Spatiotemporal variations of extreme precipitation have critical impacts on ecohydrological processes [72][73][74]. erefore, the decision-makers need to realize the spatial heterogeneity and characteristic differences of the extreme precipitation events. e change point of the extreme precipitation indices in the northwestern mainly occurred in 2007. RX1day abruptly changed in 1978 and RX5day and R95p abruptly changed in 1989 in the midnorthern. e abrupt change points of extreme precipitation indices in the midsouthern and southeastern mainly appeared around 1997 and 2002. Abrupt changes in the 1980s may be related to the change in anomalous water vapors transport over the tropical Indian Ocean-western Pacific in the mid-1980s [75]. Gao et al. indicated that the majority of extreme precipitation indices over monsoon region in China displayed some abrupt shifts in the 1990s which temporally coincided with the abrupt shifts in the El Niño-Southern Oscillation, Indian Ocean Dipole, and Atlantic Multidecadal Oscillation around this period [24]. Xie et al. found that precipitation time series in the central area of southwest China showed an abrupt decrease around 2002-2008, being consistent with the abrupt decrease in 2007 in the Indian summer monsoon [56].
Global extreme precipitation showed a widespread increasing trend and has seriously affected human life and social development [12][13][14]36]. However, extreme precipitation events have great regional differences. is study provides a detailed analysis of spatiotemporal variations of extreme precipitation in the JRB. is is very important to study the characteristics of extreme precipitation events in the high mountains of Asia under the background of global climate change. Furthermore, the results can be useful for local government to devise better adaptation and mitigation measures to combat potential catastrophe in changed extreme precipitation.
Data Availability e daily precipitation series was provided by China Meteorological Data Service Center and can be found online at http://data.cma.cn/. Digital Elevation Model data were downloaded from the Geospatial Data Cloud of China and accessed via http://www.gscloud.cn/.

Conflicts of Interest
e authors declare that there are no conflicts of interest regarding the publication of this paper.