Assessing the Spatial and Temporal Patterns of Seasonal Precipitation Extremes and the Potential Influencing Factors in Dongting Lake Basin, China

Abstract

The Dongting Lake Basin (DTLB) of China is a flood prone area. Knowledge of the spatiotemporal characteristics and risks of precipitation extremes is essential for flood mitigation. Based on the precipitation amount (R), precipitation intensity (Ri), max 1 day precipitation amount (Rx1) and max 5 day precipitation amount (Rx5), this study analyzed the spatial-temporal patterns, risks and investigated the influences of the precipitation extremes at seasonal scale. The distributed high values of R, Ri, Rx1, Rx5 and their 5-year return levels (R₅, Ri₅, Rx1₅, Rx5₅) indicated high flood risks in the eastern and northern parts of the basin, and the general upward trends of these indices suggested increasing flood risks, except for some areas in southwestern part in spring and autumn. The precipitation extremes were related to the topographic and circulation factors, within which the latter might have greater roles. Furthermore, the trend directions of the 5-year return levels (R₅, Ri₅, Rx1₅, Rx5₅) were not always the same as the initial indices (R, Ri, Rx1, Rx5), suggesting that overall decreasing (increasing) precipitation extremes do not always represent decreasing (increasing) risks of floods. Hence, policy makers should pay more attention to the risks of precipitation extremes rather than their overall tendencies.

1. Introduction

There is growing concern about extreme climate events due to the vulnerability of our society to the adverse impacts of such events [1,2,3,4]. Climate extremes, such as droughts, floods, cold weather, have brought huge damages on human society and the ecological environment around the world [1,5,6,7,8]. However, in a gradually warming environment, the intensity and frequency of climate extremes have changed considerably in the past decades [8].

The Dongting Lake Basin (DTLB), located in the middle reach of the Yangtze River, is an important grain and cash crop planting base in China. However, this region is sensitive to natural hazards [9], and is also a well-known flood prone region [10]. Records show that during the period of 1950–2009, floods (regardless of the severity) occurred almost every year in this basin [11], and caused economical loss of about $0.3 billion per year in agriculture [9,12]. For example, the 1996 flood affected 38 cities, 1.1 × 10⁶ hectare farmland, and about 7.2 million people’s livelihood and wealth, causing direct economic loss of $4.95 billion [13]. While in 1998, a similar severe flooding hit this basin again and brought out huge damages to the water conservancy facilities, agriculture, and influenced deeply the people’s lives. However, climate change in this region is also relatively considerable [14,15]. Previous studies pointed out that the frequency of floods in DTLB increased significantly in the last half century [16]. For example, in the periods of 1931–1990 and 1991–2000, there were 14 and 6 major floods, with the return periods of 4.3 and 1.7 years, respectively.

Indeed, there exist many factors that can influence the formation of flood, such as the water conservancy facilities, land cover and land usage, precipitation, etc. Nevertheless, paroxysmal large amount of water could be the prerequisite, which mainly attribute to heavy precipitation [1]. Therefore, knowledge of the spatial-temporal characteristics of precipitation extremes is very important in disaster mitigation, water resources management, agriculture and policy making [1]. Recently, many studies have focused on climate extremes at regional and global scale [1,3,4,8,17,18], and at varying time scales [7,19]. Similarly, climate extremes have also been intensively studied in China [20,21,22,23]. Overall, many of these studies focused on the changing patterns of several climate extreme indices [20,22], and some of them analyzed the probability distributions of the climate extremes [1,23,24,25] and investigated the potential underlying mechanisms or influencing factors, such as the ENSO events, SST, atmospheric circulation and topography [15,23,26,27]. A number of previous studies have attempted to characterize climate extremes in the DTLB. For example, Wang, Jiang, Wang and Yu [14] reported that precipitation in DTLB exhibited apparent upward trends after 1990s (especially in summer), and the storm frequency in summer also increased obviously, while the storm intensity showed no significant changes. However, Xu, et al. [28] concluded that during the period of 1960–2011, annual precipitation decreased gradually and had different changing directions in different seasons, but the trends were not significant; Song, et al. [29] also pointed out that the precipitation extremes changed insignificantly in DTLB.

From these existing literatures, most of the researchers concluded that the precipitation had overall downward trends and precipitation extremes changed insignificantly in DTLB [28,29,30,31,32]. The insignificant changing precipitation extremes could not reflect the facts that floods occurred more frequently in DTLB [11,13]. The explanation is that the previous studies mainly used some trend detection methods (e.g., MK test) to examine the trends of the precipitation [28,29,31]. However, though trend detection can point out the overall changing patterns of a time series, it would also miss some information of individual events, e.g., the floods. In recent years, scholars tried to investigate the risks of precipitation extremes by employing probability distributions [1,23,24,25]. They found that although the initial precipitation indices had no significant trends, the risks of precipitation extremes always exhibited significant changes [1,23]. The risk of precipitation extreme shows the severity of event would reach at a given probability or the possibility that an extreme event would occur. Hence, the risk (or probability) characteristics of a precipitation index may be more effective in implying the evolution of extreme events. However, few studies have focused on this issue in DTLB. Therefore, this study aimed to (1) analyze the trends and probabilistic characteristics of precipitation extremes at seasonal scale; (2) investigate the potential influencing factors for the changing patterns of the precipitation extremes; and (3) evaluate the changing patterns of floods implied by the precipitation extremes.

2. Data and Methods

2.1. Study Area

The Dongting Lake Basin (DTLB) is situated in the middle reach of the Yangtze River (approximately 24°38′ N–30°26′ N, 107°16′ E–114°17′ E) [33]. Within which the Dongting Lake is the second largest freshwater lake in China, and is also one of the most important international wetlands [33]. The lake is fed by four other major rivers, namely the Xiangjiang River, Zishui River, Yuanjiang River and Lishui River (Figure 1). The total area of DTLB is about 2.62 × 10⁵ km², and occupies about 14% of the Yangtze River Basin. This region has a subtropical monsoon climate, jointly influenced by the southeast monsoon, southwest monsoon, subtropical anticyclone and the westerly. The wet season in DTLB is between July and September while the dry season is between November and the next February. The topography in this basin is relatively complex (Figure 1), with mountains located in the eastern, southern and western parts, and low hills and flatlands in the central and northern part (Figure 1).

2.2. Data and Indices

This study used a 54-year (1960–2013) dataset of daily meteorological observations [34] at 43 stations (Figure 1), which has gone through quality control procedures. For the purpose of representing the extreme precipitation events, four precipitation indices, including the seasonal total precipitation (R), simple precipitation intensity (Ri), max 1 day precipitation amount (Rx1) and the max 5 day precipitation amount (Rx5), were employed [1]. These indices were calculated at the seasonal scale. The definitions and units of these indices are summarized in

Table 1. Precipitation indices and their definitions in this study.

Indices	Unit	Definition
seasonal total precipitation (R)	mm	Total precipitation in a season
simple precipitation intensity (Ri)	mm/day	The mean precipitation in the rainy days in a season
max 1 day precipitation amount (Rx1)	mm	Maximum daily precipitation in a season
max 5 day precipitation amount (Rx5)	mm	seasonal maximum consecutive 5 days precipitation

. Note that here the four seasons, the spring, summer, autumn and winter, were defined as the periods of March–May, June–August, September–November and December–February, respectively. Therefore, a total of 53 data (1960–2012) for each precipitation indices were obtained.

The circulation factors were reflected by using the atmospheric moisture flux (AMF) and moisture flux divergence (AMD). The vertically integrated seasonal mean AMF, mean AMD of the layer (surface–300 hPa) [35] in different seasons were calculated using the NCEP/NCAR reanalysis dataset (1960–2013) [36]. Note that due to the resolution of NCEP/NCAR is 2.5 degrees, the atmospheric moisture flux at each station was generated using the linear interpolation method. In order to represent the topographic conditions (topographic factors), the mean elevation (ME), standard deviation of elevation (SE), mean slope (MS), and standard deviation of slope (SS) were used (

Table 2. Definition and calculation methods of the topographic factors.

Factors	Abbreviations	Equations *
Mean elevation	ME (m)	$\mathrm{ME}=\frac{1}{n}{\sum }_1^{n}El{e}_i$ (1)
standard deviation of elevation	SE (m)	$\mathrm{SE}=\sqrt{\frac{1}{n}{\sum }_1^n{\left(El{e}_i-\overline{Ele}\right)}^2}$ (2)
Mean slope	MS (°)	$\mathrm{MS}=\frac{1}{n}{\sum }_1^n{\mathsf{\beta }}_i$ (3)
standard deviation of slope	SS (°)	$\mathrm{SS}=\sqrt{\frac{1}{n}{\sum }_1^n{({\mathsf{\beta }}_i-\overline{\mathsf{\beta }})}^2}$ (4)

Notes: * β is the slope, Ele_i is the elevation of each element (i), and n is the number of elements in the circular area.

) [23,37]. The topographic factors were calculated using the digital elevation model (DEM) (30-m spatial resolution) in a circular area within 50 km of each meteorological station (the area is 1962.5 km²) [23].

2.3. Methods

The trends of the four precipitation indices, the atmospheric moisture flux (AMF) and atmospheric moisture divergence (AMD) and the corresponding regional mean of these indices in various seasons were firstly analyzed using the modified pre-whitening Mann–Kendall trend test (MK-TFPW) [38,39,40]. The regional mean series were calculated as an arithmetic mean of values at all stations:

$$x_t=\frac{1}{n}{\displaystyle \sum }_1^n{x}_{i,t}$$

(5)

where the x_t is the regional mean index at year t, x_i,t is the index for station i at year t.

For estimating the probability characteristics of these precipitation indices, 13 probability distributions were used to fit the distributions of the seasonal series. These probability distributions were the Gaussian, Student’s t-test, Poisson, exponential, Rayleigh, Weibull, generalized extreme value (GEV), binomial, negative binomial, lognormal, geometric, generalized Pareto (GP), and extreme value (EV) distribution. These probability distributions were widely used in hydrometeorological studies [1,23,41,42]. The parameters of these probability distributions were estimated using the maximum likelihood method, and the Kolmogorov Smirnov test was used to choose the optimal distribution for each index. Then based on the optimal distribution, the 5-year return levels of the four precipitation indices at each station were determined. Here the 5-year return levels for R, Ri, Rx1, Rx5 were denoted as the R₅, Ri₅, Rx1₅, Rx5₅, respectively. Note that the reason for using 5-year return levels is that the return period of floods (or extreme precipitation events) was shorter than 5 years. As pointed out in Reference [16], there were 14 and 6 floods occurred during 1931–1990 and 1991–2000, with the flood return periods being 4.3 and 1.7 years, respectively. Hence, 5-year return level could represent well floods in this region. Note that in order to facilitate the distinction between the precipitation indices and their 5-year return levels, here we denote the precipitation indices defined in Table 1 as the ‘initial indices’ in the next context.

For examining the changing patterns of the 5-year return levels (R₅, Ri₅, Rx1₅, Rx5₅), a 30-year moving window was used [23]. Therefore, for each precipitation indices, a total of 24 windows could be obtained in the 53-year period (1960–1989, 1961–1990, …, 1983–2012). In each window, the 13 probability distributions and the same procedures mentioned above were used to determine the 5-year return levels. Hence, a total of 24 5-year return levels for each indices (R₅, Ri₅, Rx1₅, Rx5₅) at each station were obtained, and the trends of these 5-year return levels were examined also using the modified pre-whitening Mann–Kendall trend test.

Correlation analyses were used to study the relationships between the precipitation extremes and the topographic/circulation factors. Particularly, in each moving window (30-year), correlation analyses were conducted between the 5-year return levels of each extreme index and the topographic (ME, SE, MS, SS) and circulation factors (AMF and AMD) in the same window. For example, the correlation between Rx5₅ and MF in the 1960–1989 window was conducted between the 43 Rx5₅ (at the 43 stations) in the whole basin and the 43 AMF in the same window (1960–1989) (spatial correlation), while the correlation between Rx5₅ and ME (e.g., in the 1960–1989 window) was conducted between the 43 Rx5₅ and the 43 ME (the topographical factor at each station was a constant). Therefore, 24 values of correlations could be obtained (such as Rx1 vs. AMF, Rx1 vs. ME, etc.), and then the MK-TFPW was used to examine the trends of these correlations.

3. Results

3.1. Trends of the Seasonal Precipitation Indices

As shown in Figure 2, the four precipitation indices had different changing patterns and spatial distribution in the four seasons. In particular, the R (seasonal total precipitation) exhibited descend trends at almost all the stations (about 93%) in spring and in autumn, and upward trends in summer and winter (Figure 2a–d). The Ri decreased in the central part of the DTLB and increased in other parts in spring, and showed uptrends in summer, autumn and winter (Figure 2e–h). While the Rx1 and Rx5 had generally the same changing patterns in summer, autumn and winter (Figure 2j–l,n–p), and differed slightly in spring. Particularly, the Rx1 and Rx5 increased in summer, autumn and winter in major part of the basin; whereas in spring, the Rx1 decreased in the central part and increased in the south and north part (Figure 2i), while the Rx5 had general tendencies to decline except at few stations located in the southeastern mountainous regions (Figure 2m).

Overall, the regional mean R (seasonal) decreased by −9.30 and −10.29 mm∙season⁻¹∙decade⁻¹ in spring and autumn, and increased by 7.40 and 5.91 mm∙season⁻¹∙decade⁻¹ in summer and winter, respectively; data also showed that the regional annual mean R (annual) decreased by −5.6 mm∙year⁻¹∙decade⁻¹ in this basin. The Ri increased by 0.06, 0.34, 0.23 and 0.25 mm∙day⁻¹∙decade⁻¹ in spring, summer, autumn and winter, respectively; the Rx1 decreased by −0.28 and −0.26 mm·day⁻¹·decade⁻¹ in spring and autumn, and increased by 0.87 and 0.97 mm∙day⁻¹∙decade⁻¹ in summer and winter; while the Rx5 decreased by −1.74 and −0.79 mm∙(5·days)⁻¹∙decade⁻¹ in spring and autumn, and increased by 1.66 and 2.30 mm∙(5·days)⁻¹∙decade⁻¹ in summer and winter (

Table 3. Changing trends for the regional mean R (mm∙season⁻¹∙decade⁻¹), Ri (mm∙day⁻¹∙decade⁻¹), Rx1 (mm∙day⁻¹∙decade⁻¹) and Rx5 (mm∙(5·days)⁻¹∙decade⁻¹) in different seasons.

Indices	Spring	Summer	Autumn	Winter	Indices	Spring	Summer	Autumn	Winter
R	−9.30	7.40	−10.29	5.91	Rx1	−0.28	0.87	−0.26	0.97
Ri	0.06	0.34 *	0.23	0.25	Rx5	−1.74	1.66	−0.79	2.30

Note: * The bold number in the table means significance at p < 0.05.

). However, these changing trends mentioned above were generally not significant (p > 0.05) (Figure 2, Table 3).

3.2. Spatial-Temporal Patterns of the 5-Year Return Levels

Overall, the order of these 5-year return levels from high to low were summer > spring > autumn > winter (Figure 3). Furthermore, the distribution of these return levels also exhibited apparent regional features (Figure 3). Figure 3a–d showed that relatively high R₅ were mainly distributed in the eastern part in spring and winter, and in the western part in summer and autumn (Figure 3a–d). As for the Ri₅, high values were distributed in northeastern, northern, northeastern and eastern part in the four seasons (Figure 3e–h). The distribution of Rx1₅ and Rx5₅ were generally the same to each other (Figure 3i–p), with relative high values being observed in the eastern, northern, eastern and southeastern parts.

The changing patterns of the 5-year return levels (Figure 4) were overall the same to the initial precipitation indices (Figure 2). However, there still existed some differences. The first is that the trends for the return levels were statistically significant (p < 0.05) at most of the stations (Figure 4). Second, the trends of the return levels were not always the same to those of the initial indices defined in Table 1. For example, Figure 2a showed that the R (initial index) increased at most of the stations except at 3 stations located in the southeastern part, while the R₅ (probability-based index) exhibited uptrends at 6 stations in the northeast part. The number of the stations that had opposite trends in the initial indices and the 5-year return levels are summarized in

Table 4. Numbers of stations that had increasing primary precipitation indices and had increasing 5-year return levels, and numbers of the stations that had different trend directions in initial indices and 5-year return levels in different seasons.

Seasons	R			Ri			Rx1			Rx5
	R5 (1)	R (2)	NDiff (3)	Ri5	Ri	NDiff	Rx15	Rx1	NDiff	Rx55	Rx5	NDiff
Spring	6	3	7	27	23	12	20	20	10	18	9	11
Summer	33	34	13	37	41	4	33	34	9	31	32	7
Autumn	6	4	4	39	36	5	19	17	14	24	20	10
Winter	43	42	1	43	43	0	38	39	3	42	39	3

Note: ⁽¹⁾ Number of stations that had increasing R₅; ⁽²⁾ number of stations that had increasing R; ⁽³⁾ number of the stations had different trends in R₅ and R, the same in the other columns.

. Clearly, the number of stations with increasing initial indices were not always equal to those with increasing 5-year return levels. In general, the number of the stations that had different trends in initial indices and 5-year return levels were higher in the spring, summer, autumn than in winter (Table 4).

At the regional scale,

Table 5. Trends for the 5-year return levels for the four precipitation indices (mm∙year⁻¹∙decade⁻¹).

Indices	Spring	Summer	Autumn	Winter	Indices	Spring	Summer	Autumn	Winter
R5	−16.95 *	11.56	−14.29	12.82	Rx15	0.06	1.11	0.02	1.71
Ri5	0.13	0.36	0.37	0.41	Rx55	−0.85	2.38	0.61	3.11

Note: * The bold number in the table means significance at p < 0.05.

shows that the regional mean R₅ decreased by −16.95 (p < 0.05) and −14.29 (p < 0.05) mm∙season⁻¹∙year⁻¹ in spring and autumn, and increased by 11.56 (p < 0.05), and 12.82 (p < 0.05) mm∙season⁻¹∙year⁻¹ in summer and winter. The regional mean Ri₅ increased in all the seasons by 0.13 (p < 0.05), 0.36 (p < 0.05), 0.34 (p < 0.05), and 0.41 (p < 0.05) mm∙day⁻¹∙year⁻¹ respectively. In addition, the regional mean Rx1₅ increased in all the seasons by 0.06 (p > 0.05), 1.11 (p < 0.05), 0.02 (p > 0.05), and 1.71 (p < 0.05) mm∙day⁻¹∙year⁻¹, respectively. While the Rx5₅ decreased by −0.85 (p < 0.05) mm∙day⁻¹∙year⁻¹ in spring and increased by 2.38 (p < 0.05), 0.61 (p > 0.05), and 3.11 (p < 0.05) mm∙(5·days)⁻¹∙year⁻¹ in summer, autumn and winter, respectively.

3.3. Relationships between Precipitation Extremes and Environmental Factors

Figure 5 showed that the four topographic factors (ME, SE, MS and SS) have apparent decline trends from the southwestern part to the northeastern part. These distributions are coincided with the topography of this catchment, with mountains located in the eastern, southern and western parts, and low hills and flatlands in the central and northern part (Figure 1). As for the circulation factors, the regional mean atmospheric moisture flux (SMF) decreased significantly in all the seasons (p < 0.05), and the atmospheric moisture divergence (AMD) decreased (p < 0.05) in summer and increased (p < 0.05) in other seasons (Figure 6).

Table 6. Correlation between the precipitation extremes and the topographic and circulation factors and the corresponding trends.

Indices	Correlations						Changing Trends of Correlations (× 10−4∙Year−1) **
	MS	SS	ME	SE	AMF	AMD	MS	SS	ME	SE	AMF	AMD
R5	−0.26	−0.09	−0.50 *	−0.14	0.55	−0.60	5.2	6.1	4.0	5.2	0.1	4.5
	0.54	0.44	0.46	0.42	−0.49	0.06	2.1	4.2	−3.7	3.4	−11.8	−10.4
	0.28	0.37	0.14	0.32	0.05	−0.03	−11.8	−12.3	−9.1	−11.4	3.9	5.7
	−0.37	−0.20	−0.56	−0.21	0.59	−0.75	−1.1	−1.5	−1.1	−1.2	0.5	1.0
Ri5	−0.46	−0.30	−0.71	−0.36	0.24	−0.66	5.9	7.0	5.2	5.8	0.4	3.5
	−0.37	−0.35	−0.51	−0.40	−0.14	−0.65	−6.6	−3.6	−3.5	−4.2	1.1	0.6
	−0.31	−0.21	−0.52	−0.26	−0.41	−0.38	7.6	12.2	7.2	9.1	−8.7	13.0
	−0.49	−0.34	−0.69	−0.37	0.39	−0.75	0.8	1.1	1.2	1.1	−2.3	0.4
Rx15	−0.14	−0.09	−0.31	−0.11	0.13	−0.27	3.0	4.6	2.7	4.7	−2.5	5.1
	0.01	0.04	−0.13	−0.04	−0.24	−0.35	5.7	6.4	10.0	−1.6	5.6	2.2
	0.06	0.16	−0.06	0.10	−0.05	−0.26	−4.5	−13.5	12.3	−7.7	3.9	16.2
	−0.34	−0.17	−0.49	−0.18	0.61	−0.76	0.9	−0.2	0.1	0.2	1.1	−4.2
Rx55	−0.24	−0.11	−0.44	−0.13	0.45	−0.47	−0.5	1.0	−0.9	0.1	6.4	−1.4
	0.16	0.12	−0.03	0.08	−0.21	−0.32	6.3	6.4	0.6	5.2	0.2	4.0
	0.18	0.29	0.07	0.26	0.15	−0.16	−17.6	−21.6	−4.4	−16.1	13.2	18.3
	−0.36	−0.22	−0.55	−0.22	0.55	−0.78	0.4	0.4	−0.3	0.7	0.2	−0.6

Notes: * The bold number in the table means that the correlation or the trends were significant at p = 0.05; ** the changing trends was calculated based on the absolute value of the correlations.

showed that the four return levels had relationships with the topographic and circulation factors. Particularly, the R₅, Ri₅, Rx1₅ and Rx5₅ were negatively correlated to the ME, SE, MS and SS in spring and in winter. While in summer and autumn, the R₅, Rx1₅, Rx5₅ were generally positively and Ri₅ negatively correlated to the topographic factors (Table 6). As for the circulation factors, R₅, Ri₅, Rx1₅ and Rx5₅ were negatively correlated to AMD in all the seasons, and generally positively correlated to the AMF (Table 6). However, negative correlation for AMF could also be observed in summer and autumn (Table 6). Relatively, the R₅ and Ri₅, especially the Ri₅, had higher correlations than the Rx1₅ and Rx5₅ (Table 6).

Furthermore, the correlations generally exhibited significant (p < 0.05) trends in the past decades. As for the topographic factors (ME, SE, MS and SS), the R₅, Rx1₅ and Rx5₅ mainly had increasing correlations in spring, summer, and winter, and had downward correlations in autumn; while the Ri₅ mainly had increasing correlations in spring, autumn, and winter, and had decreasing increasing correlations in summer. As for the circulation factors (AMF and AMD), the correlations were mainly increasing except that for the R₅ in summer (Table 6).

4. Discussion

Humans and natural environment are vulnerable to the impacts of climate and weather extremes [1]. The Dongting Lake Basin (DTLB) is a flood prone area in China, and floods have resulted in dramatic damages to the society and natural environment in this region [43]. On the other hand, the Dongting Lake is the secondary largest freshwater lake in China, which plays important roles in regional economy development, regional environmental and ecological issues [33]. Intensive precipitation events also have impacts on soil erosion and bring huge sediments or pollutants into the lake and result in high pressure of lake degradation and water pollution [44,45]. Comprehensive knowledge of the probability or risks of the precipitation extremes and the potential influencing factors is essential for the management and mitigation of natural hazards and for the Dongting wetland protection. This study analyzed the trends and probabilistic characteristics of four precipitation extremes, including the total precipitation amount (R), simple precipitation intensity (Ri), max 1 day precipitation amount (Rx1) and max 5 day precipitation amount (Rx5) in different seasons in the DTLB, and investigated the relationship between the precipitation extremes and the local topographic and circulation factors. It would be meaningful for evaluating the changing patterns of floods or droughts and improving water resources management in this region.

The results showed that the selected precipitation indices at most of the stations generally had insignificant changing trends in the four seasons (Figure 2) [29,31]. However, the direction of the trends differed in different seasons. At the regional scale, the R, Rx1 and Rx5 decreased in spring and autumn, and increased in summer and winter, while the Ri exhibited upward trends in all the seasons. These results were consistent with other regions in China [46]. Together with the facts that the annual precipitation amount decreased in the same decades, the results above indicate that the precipitation would have tended to concentrate in summer and winter, especially in summer. Furthermore, the trends of the Rx1 and Rx5 in winter were 0.97 mm∙days⁻¹∙decade⁻¹ and 2.30 mm∙(5·days)⁻¹∙decade⁻¹, respectively, being generally higher than those (absolute value) in other seasons, implying that extreme precipitation events in winter were enhanced more significantly.

The distribution of 5-year return levels of the four precipitation indices were overall similar to each other, with high value being in the eastern and northern parts of the basin (Figure 3). The 5-year return level represents the value that the precipitation extremes could reach at a probability of 80% in a period. Higher the value is, higher the risk of floods is [1]. Please note that there are many factors affect the occurrence of floods besides the extreme precipitation, such as the land use and vegetation change, dam construction, which can change the response of a region (catchment) to the storm [47]. In the Dongting Lake Basin, ecological restoration was implemented (e.g., the “Green for Grain” program [48]) and water conservancy facilities [13,49] were constructed and improved for alleviating flood hazards in the past decades. However, these factors (e.g., land use or cover change) can decrease the occurrence frequency of floods, but cannot control the flood risks brought by the extreme precipitation, since the latter is mainly determined by the climate (weather) change though feedbacks also existing between the land surface and precipitation. Due to the improvement of floods prevention measures, a relative extreme precipitation event may not result in a real flood in the region, however, this extreme events would also bring high pressure to the water conservancy facilities. In other words, the flood risk brought by the extreme precipitation still exists even in this case. On the other hand, in this study the flood risk was implied using the 5-year return levels of the precipitation extremes. According to the historical records [16], the return period of floods was shorter than 5 years in this catchments. Therefore, the 5-year return levels can indicate floods risks well. From this perspective, the distribution of the 5-year return levels indicated that the eastern and northern part of DTLB were in higher risks of floods. Moreover, in spring and autumn, the decreasing trends of the 5-year return levels (Figure 4) suggested that flood risks generally decreased (decreasing 5-year return levels) in the southwestern part and increased (increasing 5-year return levels) in other parts of the basin. While in summer and winter, the overall increasing 5-year return levels indicated that the flood risks increased in the entire basin. Indeed, the wet season in DTLB generally covers the period from April to September [33], within which the floods mainly occurred. However, the results indicated that risks of floods in winter increased significantly, especially in the southeastern part of the basin (Figure 3 and Figure 4). This is consistent with the records that in the winter in 1997, 2002 and 2015, terribly unusual floods occurred in the southeastern part of the basin [50,51].

Unlike the initial indices, the changing trends for the probability-based indices (5-year return levels) were mainly significantly (Figure 4). These indicated that though the precipitation indices changed insignificantly, the risks of floods had been significantly changed in this region. Furthermore, the direction of the trends for the probability-based indices were not always the same as those for the initial indices (Table 5). These means that overall decreasing precipitation not always represent decreasing risks of floods, and vice versa. Results in Table 4 also showed that in spring, autumn and winter, the number of the stations that had increasing 5-year return levels was generally higher than that had increasing initial precipitation extremes, while in summer, the results were opposite. These alert us that in water resources management, water conservancy project construction and hazards mitigation, the policy maker should pay more attention to the risks of the precipitation extremes.

Circulation and topography are two main influencing factors for the precipitation extremes [4,17,19,23,52,53]. As showed in this study, the changing patterns of these precipitation extremes have relationships (reflected using the correlation) with these two types of factors (Table 6). In general, the R₅ and Ri₅ had higher correlations than the Rx1₅ and Rx5₅. The reasonable explanation is that the R and Ri were the general characteristics of the precipitation, while the Rx1 and Rx5 were individual events in a period. The influencing factors selected here were the overall characteristics of the topography or the circulation features. Furthermore, the correlations exhibited significant trends in the past decades (Table 6). Actually, correlation measures the strength of a relationship between two variables [23]. The trends of the correlation reflect the strength of the relationship between two variables, or the strength of the impacts that the independent variables (the environmental factors here) have on the dependent variable (the precipitation extremes here) [23]. From this perspective, the overall increasing (absolute) correlations for the circulation factors (AMF, AMD) indicated that the large circulation would have increasing influence on the precipitation extremes (Table 6). While the mixed changing correlations for topographic factors (ME, SE, MS and SS) (Table 6) in different seasons suggested that the influence of topography have different behaviors.

The results were partly different from Liu, Xu and Sun [23], who reported that the correlations for topographical factors generally decreased and the correlations for circulation generally increased in southwestern China at the annual scale. These may be due to the differences in time scales, topography and the climate systems. The climate in DTLB is controlled by the subtropical anticyclone, southeast monsoon, southwest monsoon, and the westerly. The evolution of these climate systems result in a unsteady climate system in this basin [54]. While the climate in southwestern China is controlled by the South Asia monsoon but also influenced by the East Asia monsoon and the Tibetan Plateau monsoon and the westerlies [32]. As for the topography, mountainous landscape distributed in the eastern, southern and western part of the basin, and low hills and flat plain in the central and northern part in DTLB (Figure 1). These are quite different from that in southwestern China [23], which generally declines from west toward east and from north toward south. Therefore, the interaction of topography and climate systems would be different in the two regions, as Shi and Durran [55] pointed out that the sensitivities of extreme precipitation to global warming were lower over mountains than over oceans and plains.

However, precipitation formation is a complex processes, and is influenced by many local and large scale factors [17,56,57,58]. The topography and circulation factors are the main factors, but also cannot fully explain the behaviors of the precipitation extremes. For example, the land cover changes (e.g., deforestation), soil moisture and urbanization also exert potential influences [58,59,60,61]. In this study, the mixed changing trends of the correlations between topographic factors and the precipitation extremes in different seasons also demonstrated the complexity of the underlying mechanisms. Results also showed that the atmospheric moisture flux (AMF) and atmospheric moisture divergence (AMD) were decreasing in most of the seasons in the past decades (Figure 6) [62]. These trends of were not always the same to the precipitation extremes. For example, in the summer and winter, the precipitation and precipitation extremes had considerable upward trends. However, through the atmospheric moisture flux was weakening in the past decades, the atmospheric moisture budget in this region seemed to be increasing [63]. In the context of a warming environment, low-level moisture changes with warming atmosphere could fuel comparable changes in heavy precipitation events [18], many studies pointed out that global warming could lead to a global increase in extreme rainfall events [64,65]. Hence, the more detailed studies are still needed to fully understand the linkage between precipitation extremes and the local drivers, large circulation, temperature and other environmental factors.

5. Conclusions

This study investigated the changing patterns, probabilistic characteristics and the potential influencing factors of the precipitation extremes in Dongting Lake basin (DTLB) at a seasonal scale. Results showed that the DTLB had overall decreasing annual precipitation in the past decades. However, the precipitation extremes had apparent regional features and had different behaviors in different seasons. High precipitation extremes (R-total precipitation amount, Ri-precipitation intensity, Rx1-max 1 day precipitation amount, Rx5-max 5 day precipitation amount and their 5-year return levels R₅, Ri₅, Rx1₅, Rx5₅) were generally distributed in the eastern and northern parts of the basin, implying high risks of floods in this regions. Furthermore, overall upward trends of these precipitation extremes (5-year return levels) mainly occurred throughout the basin, indicating that the risks of floods increased in the past decades, except for some areas located in the southwestern part of the basin in the spring and autumn. Meanwhile, the directions of the trends of the 5-year return levels (R₅, Ri₅, Rx1₅, Rx5₅) were not always the same as the initial precipitation indices (R, Ri, Rx1, Rx5), suggesting that decreasing (increasing) precipitation extremes not always represent decreasing (increasing) risks of floods. Therefore, policy makers should pay more attention to the risks (probability characteristics) of the precipitation extremes. On the other hand, these precipitation extremes were related to the topographic (mean elevation, standard deviation of elevation, mean slope and standard deviation of slope) and circulation factors (atmospheric moisture flux (AMF) and atmospheric moisture divergence (AMD)). The gradually increasing correlations between AMF, AMD and the precipitation extremes implied that the circulation factors might have greater and increasing roles in influencing the behaviors of these precipitation extremes. However, further studies on the interaction of local drivers, large scale circulation, temperature and other potential environmental factors are in great need for understanding the underlying mechanisms of the behaviors of the precipitation extremes.