Spatiotemporal variation of flash floods in the Hengduan Mountains region affected by rainfall properties and land use

Understanding the spatiotemporal characteristics of flash floods is important for reasonable and accurate identification of high-risk regions and for prediction of future hydrological regimes. Therefore, using time series datasets (1979–2015) of historical flash flood events, rainfall, and land use in the Hengduan Mountains region (HMR) of China, this study analyzed the effects of changes in precipitation and land use on spatiotemporal variation of flash floods. Analysis of the linear trend revealed that flash floods increased by 12 events/10 years and that 82% of events occurred in the flood season (June–August). The increase in flash flood events was found closely correlated with the increased frequency (rainstorm days increased by 3.5 d/10 years) and magnitude (rainstorm amount increased by 10.2 mm/10 years) of heavy rainfall as well as the expansion of the area of artificial surfaces (increase of 999 km2). Morlet wavelet analysis revealed significant periodic variation in the occurrence of flash floods on scales of 3–7, 8–15, and 21–31 years. On the basis of the standard elliptical difference, we identified displacement of the center of gravity of flash flooding from the northwest toward the southeast. Generally, more disasters were recorded in southern parts of the HMR owing to the frequent occurrence of rainstorms and the increase in area of both cultivated land and artificial surfaces with higher runoff potential. These findings improve understanding regarding the spatiotemporal dynamics of flash floods in the HMR and could support policymakers in identifying high-risk areas in mountainous watershed.


Introduction
A flash flood is a sudden-release event with high peak discharge and short response time in a mountainous watershed from a few tens to hundreds of square kilometers, which is generally triggered by heavy rainfall (Borga et al. 2010;Hapuarachchi et al. 2011;Lumbroso and Gaume 2012). Globally, flash floods represent one of the deadliest natural hazards (Špitalar et al. 2014) that often result in loss of life and substantial economic damage (Kumar et al. 2018;Saharia et al. 2017;Yu et al. 2018b). Earlier studies demonstrated that both the formation and the distribution of flash floods are affected by rainfall and the characteristics of the underlying surface (Atif et al. 2021;Gao et al. 2020;Hu et al. 2018). Land use and climate change, which are often referred to as "global change" (Lopez-Tarazon et al. 2019), are two of the most critical drivers of hydrological variations (Kim et al. 2013;Kundu et al. 2017), affecting the formation and development of flash floods via complex interaction of the water-vegetation-soil system Swain et al. 2021;Wang et al. 2020;Zhang et al. 2018). Globally, the influence of climate change is causing weather systems to become more and more unstable, which is increasing the occurrence of extreme weather events (Roy et al. 2020;Yu et al. 2018a). One of the main impacts of such change appears to be increase in both the frequency and the magnitude of extreme rainfall events (Allan and Soden 2008;Barredo 2006;Debortoli et al. 2017;Li et al. 2021;Llasat et al. 2021).
In recent decades, land use has undergone tremendous transition through urbanization and intensification of human activities (Abellan et al. 2019; Avashia and Garg 2020; Wan Mohtar et al. 2020). Flash floods susceptibility in many regions is particularly heightened owing to variation in both the frequency and magnitude of extreme rainfall events and the rate of land use change, particularly in association with urban growth (Borga et al. 2010;Llasat et al. 2016;Penna and Borga 2013). Therefore, the recognized increase in flash flood events attributable to changes in rainfall and land use has attracted worldwide attention. Quantitative assessment has been undertaken regarding the impact of changes in climate and land use on streamflow variation (Bronstert et al. 2002;Shahid et al. 2017;Swain et al. 2021), and a recent study reported that certain discharges of watersheds with the size of 10-201 km 2 increased significantly from the baseline land use scenario to the most urbanized scenario (i.e., from ≈50, 190, and 380 m 3 /s to ≈235, 385, and 940 m 3 /s, respectively) (Abellan et al. 2019). It has also been demonstrated that the occurrence of flash flood events shows an increasing trend owing to rainfall and land use variation (Llasat et al. 2014;Zhang et al. 2019b). For example, the risk of flash floods in the Yesanpo Scenic Area of, Beijing, China, has increased by 28% following tourism-related development and land-use changes .
With mountains covering two-thirds of its land area, China is at considerable risk of flash floods, and more than 60,000 such events occurred during 1949-2015 (Liu et al. , 2018. The climatological sensitivity of the Tibetan Plateau makes it a national and even a global "natural laboratory" reflecting its response to the effects of climate change (Cui et al. 2014(Cui et al. , 2015Yang et al. 2020). The Hengduan Mountains region (HMR), located in the eastern margin of the Tibetan Plateau, also makes an obvious response to climate change (Wang et al. 2013;Xu et al. 2018), as evidenced by the regional increase in extreme precipitation Shi et al. 2014;Wu et al. 2017;Zhang et al. 2014). Regionally, the distribution of heavy rainfall, which is affected by the complex topography, presents a pattern with substantial spatial discrepancy Ma et al. 2013). Moreover, the morphology of the mountainous basins with small and steep river catchments can transform the generation of intense runoff into severe devastating flash floods (Cui and Guo 2021;Liu et al. 2020;Penna and Borga 2013). The HMR has experienced dramatic alterations of land use over the past 20 years ); However, the effects of these changes on flash floods remain little studied. Additionally, active strong earthquakes, frequent extreme rainfall events, and the expansion of artificial surfaces associated with major projects such as the Sichuan-Tibet Railway markedly affected the hydrology of the HMR, possibly leading to the increased risk of flash floods (Zhang et al. 2019a(Zhang et al. , 2021a. In the context of the above, it is essential to conduct comprehensive documentation of past events when performing a flash flood research . Gaume et al. (2009) emphasized the importance of collecting information regarding historical flash flood events and creating an event database. On this basis, intensive investigation has been conducted on the temporal evolution, spatial distribution and patterns, seasonality, and numbers of fatalities and injuries associated with flash floods (Kaiser et al. 2021;Llasat et al. 2014), and analysis has been performed on the influence of environmental factors on the distribution of flash floods Xiong et al. 2019). Also, numerical models have been used to predict the characteristics of flash flood events in the future (Avashia and Garg 2020;Zhang et al. 2021b).
Owing to data constraints, synchronized analysis of the long-term spatiotemporal evolution of flash floods with the consideration of rainfall and land use patterns is lacking. Therefore, this study collected contemporaneous time series datasets  of historical flash flood events, rainfall, and land use with regard to the HMR. The study have two principal objectives: (1) to characterize the spatiotemporal variability in flash floods in the HMR during 1979-2015 and (2) to explore the effects of rainfall and land-use characteristics on the spatiotemporal variation of flash floods. The findings could provide scientific reference and decision-making support with regard to the formulation of reasonable measures for disaster prevention and the effective flood risk management.

Study area
The HMR (24°39′-33°34′N, 96°58′-104°27′E) is located in southwest China where neotectonic movement is active. It lies in the transition area between the Tibetan Plateau and the Sichuan Basin, and regional elevation decreases from the northwest to the southeast (Fig. 1). The terrain is steep under the control of the uplift and orogeny of the Tibetan Plateau. Within a horizontal distance of 70-150 km from east to west, the vertical height difference can be as great as 4,000 m, and the terrain gradient can be up to 2-6%. The HMR has a typical monsoon climate. The area is affected not only by the East Asian monsoon from the Western Pacific Ocean and the South Asian monsoon from the Bay of Bengal and Indian Ocean, but also by the Tibetan Plateau monsoon and westerlies. Owing to the effects of climate change as well as the strength and retreat of the monsoons, severe rainstorm events occur frequently (Kumar et al. 2018;Li et al. 2012). Under the influence of the various monsoons and owing to the complex terrain, the HMR has prominent seasonal and vertical climatic zones (Yu et al. 2018a). The substantial differences in climate conditions between the different regions contribute to various regional vegetation types that include shrubs, forests, and grassland ).

Flash flood events
Flash flood events data were obtained from the National Flash Flood Investigation and Evaluation Project database, which was launched to investigate Chinese historical flash flood events. The database comprises records of more than 60,000 flash flood events that occurred in China during 1949-2015, including details of the occurrence time, location, associated precipitation, and recorded injuries for each disaster event. Owing to incomplete records of flash flood events before 1979, this study considered only those data corresponding to the 2044 flash flood events that occurred in the HMR during 1979-2015.

Rainfall
Owing to difficulty in obtaining hourly rainfall data, this study used daily rainfall data to analyze the relationship between the spatiotemporal variation of flash floods and rainfall. Meteorological station data were used to analyze the temporal evolution of rainfall in the HMR. However, owing to the low density of meteorological stations in the HMR, accurate representation of regional precipitation cannot be obtained through interpolation of station data. Therefore, the well-recognized and widely used approach of using gridded rainfall data was adopted to analyze the spatial variation of rainfall.
On the basis of their contemporaneous record length  with the flash flood data, meteorological data recorded at 25 national standard meteorological stations in the HMR were obtained from the China Meteorological Data Network (http:// data. cma. cn). The gridded data comprised the Global Unified Gauge-Based Analysis of Daily Precipitation (0.50° × 0.50°) dataset, which forms part of a suite of products from the Unified Precipitation Project undertaken by the Climate Prediction Center of the National Oceanic and Atmospheric Administration (available from: https:// psl. noaa. gov/). The temporal coverage of this dataset extends from January 1, 1979, to the present.
Rainfall events are often classified in terms of depth of rain delivered in a 24-h period (Breugem et al. 2020). In this respect, heavy rainfall is defined as daily precipitation of ≥ 50 mm, and the rainstorm days represent the number of days on which a heavy rainfall event occurs. This indicator is not only a "climate change indices" (http:// etccdi. pacifi ccli mate. org/ list_ 27_ indic es. shtml) used by the World Meteorological Organization Climate Commission for Climatology, but it is also a widely used rainstorm classification standard in China Zhang et al. 2014;Zhou et al. 2011). Moreover, this threshold is adopted as the critical value for warning of precipitation-related mountain disasters in the HMR ).

Land use
This study used land use data (1-km resolution) obtained from Resource and Environment Science and Data Center of Chinese Academy of Sciences (http:// www. resdc. cn/). The dataset comprises land use data at 5-or 10-year intervals during 1980-2015 in six classes: cultivated land, forest, grassland, water bodies, artificial surfaces, and unused land. Table 1 lists specific details of the classification system.
To ensure the temporal consistency between the three datasets, the period of 1979-2015 was selected for analysis in this study.

Linear trend and wavelet analysis
The trend of change of both flash floods and rainfall can be expressed applying a linear equation (Mudelsee 2019;Xu et al. 2018): where y represents a flash flood or rainfall events, t is time  in this study), and b is the linear trend term.
Wavelet analysis (Ciria and Chiogna 2020) is an effective method for signal processing that has been used widely in time-frequency analysis of rainfall series. The Morlet wavelet function, which is a type of complex-valued wavelet used widely for systematic analysis, can be expressed as follows: (1) y = a + bt where i is a complex and c is a constant. For a given discrete time series f (kΔt)(k = 1, 2, … , N;Δt is the sampling interval), its wavelet transform is where a is the period length of the wavelet, b is the time shift of the wavelet, and W f (a, b) is the wavelet transform coefficient.
On this basis, the wavelet square difference can be calculated as follows: Equation (3) can be used to calculate the wavelet variance and to draw the wavelet variance graph, from which the main period of the time series can be determined.

Center of gravity model and standard deviation ellipse
The center of gravity (mean center) was used to systematically analyze t historical flash flood and rainfall events that occurred in the HMR during 1979-2015. According to the moving track and distance of center of disasters in 10 years, the directional distribution of disasters was analyzed in combination with the standard deviation ellipse (Xiong et al. 2019). The standard deviational ellipse was used to identify the spatial distribution of historical flash flood events and to represent the change of location and the movement trend of the center of disasters. The x-axis of the standard deviational ellipse represents the directionality of the spatial distribution of flash flood events. In contrast, the y-axis represents the degree of dispersion of the spatial distribution of the disasters.
The mean center, which is the average x-and y-coordinates of all the features in the study area, is helpful for tracking changes in the distribution or for comparing the distributions of different components. The mean center is given as follows: where x i and y i are the x-and y-coordinates for feature i, respectively, and n is the total number of features.
The weighted mean center extends to the following: where w i is the weight for feature i. The standard deviational ellipse is expressed as follows: where x i and y i are the x-and y-coordinates for feature i, respectively, X,Y representing the mean center of the features, and n is the total number of features. The angle of rotation is calculated as follows: where x i and ỹ i are the deviations of the x-and y-coordinates from the mean center, respectively.

Temporal variation in flash floods
The interannual variability of flash flood events in the HMR during 1979-2015 was analyzed as having an undulant rising characteristic of 12 events/10 years (Fig. 2a). There were  (1991: 138 events, and 1998: 230 events) when catastrophic floods were triggered by prolonged extreme heavy rainfall in the region of the Yangtze River Basin link between the ENSO (El Ni˜no-Southern Oscillation) cycle and summer climatic anomalies (Bell et al. 1999;Kundzewicz et al. 2020), which resulted in 4,150 fatalities and direct economic losses of RMB16.6 billion ($2.57 billion). It was found that flash floods in the HMR occurred mainly in summer (June-August), accounting for 82.0% (1594 events) of the events throughout the entire year (Fig. 2b). Statistically, the largest proportion occurred in July (612 occurrences), accounting for 34.8% of the total. This finding was verified annually from 1979 to 2015, indicating the month with the most frequent occurrence of flash flood disasters.
Periodic changes in the time series flash flood events, identified on the basis of the wavelet coefficients (Fig. 3a), could be used to predict future changes in the trend of flash floods on different timescales. Potential periodicities of 3-7, 8-15, and 21-31 years in the time series of flash floods from 1979 to 2015 (Fig. 3a) were identified, and significant periodicity with the locally higher energy density was found in relation to the periods of 3-7 and 8-15 years. For the period of 21-31 years, weak energy was distributed widely during 1979-2015. Three prominent peaks were found in the wavelet deviation diagram (Fig. 3b), and the fluctuation of the three periods efficiently controlled the variation characteristics of flash floods during the time series of 1979-2015. The maximum value of the wavelet peak appeared in the period scale of 9 years, indicating the strongest periodic oscillation.

Spatial distributions in flash floods
Analysis of the interannual variation of flash flood events was difficult owing to the long timescale ; therefore, historical flash flood events were counted for each decade (1979-1985,1985-1995, 1996-2005, and 2006-2015) (Yuan and Ren 2021). Although the first period was < 10 years, its impact was smaller thanks to the occurrence of fewer flash floods. For the different periods, the spatial moving track and distance of the center of gravity of the flash floods were analyzed using the center of gravity model (Fig. 4), and the disaster distribution and direction were comprehensively estimated using the standard deviation ellipse. The calculated critical parameters of the center of gravity and the standard deviation elliptical are listed in Table 2, including the coordinates and the movement distance of centroids, and the direction angle and x-as wel as y-axes distances of the standard  , the center of gravity of flash flood events was determined as lying in the region of 26.99°-28.03°N, 101.40°-101.59°E, i.e., located near the city of Panzhihua and Liangshan Yi Autonomous Prefecture, Sichuan province.
It was found that the center of gravity initially moved 74 km southward, then 3 km westward, and finally 174 km northeastward. The change of direction angle  was not significant, ranging from 3.44° to 174.13°. The flash flood events presented a north-south pattern, and the pattern showed a trend of successive strengthening, strengthening, and weakening. The length of the semi-major axis along the x-axis showed an overall trend of decrease and the data showed high-low-high fluctuation, indicating that the directionality of the spatial distribution of flash floods first weakened and then strengthened. The north-south pattern changed from significant to nonsignificant and then to substantial. The average length of the semi-minor axis along the y-axis was 264 km. The short half axis of 1979-1985 was the longest, indicating that the centripetal force of the distribution of flash floods was weak and that the degree of dispersion of the disasters was large (Fig. 5). Affected by the spatial distribution of heavy rainfall and land use, the distribution of flash floods during January-December presented a pattern with more (less) frequent occurrence in the south (north). The center of gravity of the flash flood events mainly moved in the southeast-northwest and southwest-northeast directions, reflecting the periodic changes of rainfall attributable to expansion and retreat of the southeast and southwest monsoons. Unlike the interannual movement of the center of gravity of flash floods, the monthly movement of flash floods was seldom affected by land use change because there was very little variation during the year.

Rainfall characteristic
A decreasing trend of 10.48 mm/10 years for average annual rainfall was found in the HMR, and the maximum value of precipitation appeared in 1998 (Fig. 6a). However, increasing trends of 3.5d/10 years and 10.2 mm/10 years were found for the number of rainstorm days and rainstorms precipitation, respectively, with peaks also occurred in 1998 (Fig. 6b, c). Additionally, the monthly rainfall depth and the number of rainstorm days varied substantially within the year, and the maximum values of both occurred in July (Fig. 6d). Specifically, rainstorm days and rainstorms precipitation during June-August accounted for 80% of the annual total of each year (Fig. 6e, f). This characteristic mirrored the temporal variation of flash flood events. The annual distribution of precipitation showed clear seasonality reflecting the strong influence of monsoon.
Three over-centers (i.e., 1986, 2006, and 2014) and two under-centers (1979 and 2015) were found during 1979-2015. Figure 7a shows that during the evolutionary process of rainfall during 1979-2015, there was certain periodicity on the scale of 3-8, 9-17, and 21-32 years. In contrast, the energy density for 21-32 years was relatively high, but the periodic variation was only significant from 1989 to 2011, whereas the other periodic time domain was fairly broad but not significant. Two evident peaks were found in the wavelet square deviation graph (Fig. 7b), which correspond to timescales of 22 and 28 years. The maximum peak corresponds to the timescale of 28 years, indicating that the periodic oscillation of approximately 28 years was strongest and represents the first principal period of rainfall variation. The timescale of 22 years corresponds to the second peak, i.e., the second principal period.
The spatial distribution of rainfall in the HMR region was different, affected by the topography , which decreased from southwest to northwest (Fig. 8a-d). The rain belt was generally aligned in a northeast-southwest direction, and the rainfall center moved over a distance of 3.8-16.2 km during 1979-2015 (Fig. 8e). The maximum precipitation occurred in the east-north-east of the study area, e.g., around Ya'an, Sichuan. The minimum precipitation occurred in the west-north-west of the study area, e.g., around Qamdo, Tibet. It was found that heavy rainfall occurred mainly in areas with high precipitation, i.e., at lower elevations in the south of the region where there is frequent occurrence of flash flood (Fig. 9).

Land use change
A map of land use in the HMR in 2015 is presented in Fig. 10f, various land use types during 1980-2015 are listed in Table 3. Apparent spatial heterogeneity is evident in the map of land use distribution. Forest and grassland were the main land use types, distributed widely in the mountainous, accounting for approximately 88% of the total. Grassland was mainly concentrated in the north of the region in areas with high elevation, whereas forest dominated in the central and southern areas with low elevation. Other land use types were 1 3 scattered and small in area (accounting for only 0.2%-7.7%). Artificial surfaces and cultivated land were mainly distributed in the southeast of the mountains at lower elevation and concentrated in the river valleys owing to topographical restriction. Unused land was distributed mainly in the northwest of the region, primarily around Ganzi and Qamdo, while water bodies were primarily distributed in the central areas. From 1980 to 2015, following urban development, the area of artificial surfaces expanded rapidly (i.e., from 727 to 1726 km 2 ), an increase of 137%. Before 2005, rampant growth of population led to significant reduction in the area of forest and increase in the area of grassland. Then, in 2015, forest recovered, increasing by 6685 km 2 in comparison with the area in 1980, while the area of grassland decreased by 2932 km 2 . The areas of cultivated land, water bodies, and unused land decreased by 138, 2284, and 927 km 2 , respectively, during the 35-year study period. Consideration of the increase in flash flood events and land use change in the study area during 1980-2015 (Fig. 10a-e), it revealed that the urbanization and the intensification of human activities, which caused the areas of cultivated land and artificial surfaces to expand substantially, led to increase in the occurrence of flash floods.
The land use changes of 1980 and 2015 were superimposed to obtain the transfer area matrix shown in Table 4. Land use in the HMR has changed abruptly and dramatically over the past 35 years (Fig. 11); specifically, the conversion among cultivated land, forest, and grassland is most noteworthy. The decrease in cultivated land is attributable to the transformation of the land into artificial surfaces, forest, and grassland. The main reason for the decline of grassland and the increase of forest is that a large area of grassland (49,291 km 2 ) has been changed to forest. These changes reflect national policies intended to protect the ecological environment, such as returning cultivated land to forest and grassland. Additionally, unused land has largely been turned into grassland (1557 km 2 ) and forest (6,957 km 2 ), indicating that the ecological environment of the HMR has improved. The increase in the area of artificial surfaces is attributed mainly to the conversion of cultivated land (910 km 2 ) and grassland (381 km 2 ). Owing to the acceleration of urbanization driven by socioeconomic development, other land use types are also being transformed into artificial surfaces. Conversion of various land use types into cultivated land and artificial surfaces has occurred mainly in the south of the region, where the elevation is low and human activities are intense. Conversion to forest and grassland has occurred primarily in the northern parts, especially the northwest, which are primarily high-elevation mountainous areas.

Controls of rainfall properties to spatiotemporal variability in flash floods
The occurrence and distribution of flash floods are controlled by rainfall, which is the main triggering factor, and are associated primarily with short high-intensity rainstorms (Borga et al. 2010). The occurrence of heavy rainfall over a short period can cause rapid rising flash floods (Roy et al. 2020). The contribution of convective precipitation is approximately 50% for 75% of flash floods, but it can increase to 70% for 50% of such events   (Llasat et al. 2016). The HMR is highly affected by flash floods, where convective activity causing severe weather and intense rainfall is favored owing to the combination of multiple monsoons and the complex topography . Flash floods are linked directly to rainfall, and this study indirectly analyzed whether the magnitude of the trigger itself varies temporally and spatially to enable consideration of the occurrence of such events in the context of climatic changes. It demonstrated that the spatiotemporal variations of both flash flood events and the triggering rainfall are highly consistent. The annual distribution of rainfall showed clear seasonality reflecting the strong influence of monsoons Yuan et al. 2017). Precipitation is concentrated in summer (June-August), and the occurrence of flash floods found highly similar to that of heavy rainfall because the other influencing factors of flash floods have only slight annual variation. There was a decrease in the precipitation (10.48 mm/10 years) during the study period, but increases in the number of days (3.5 d/10 years) and amount of precipitation (10.2 mm/10 years) of heavy rainfall events under the influence of climate change. The trend of increase in the occurrence of flash floods (12 events/10 years) was more dramatic than that of rainstorms, which might reflect the synergistic effect of increases in the frequency and magnitude of rainstorms triggering more flash floods in a rainstorm event. It might also reflect that land use changes increase flash flood sensitivity. It is worth noting that a relationship between the flash flood period and the rainfall period was found in this study. Resonance periods on the scale of 3-7 and 21-31 years were identified in the evolution of flash floods and rainfall during the time series , and they were controlled by the same period at 22 years.
The distribution pattern of rainfall or extreme heavy rainfall usually shows high consistency with regional topographical changes (Anna et al. 2020;Li et al. 2021). The spatial distribution of flash floods in the HMR showed a more (less) frequent occurrence in southern (northern) areas and the least frequent occurrence in the northwest, similar to the spatial distribution of rainfall and rainstorms. Comparison of the migration trajectory of the center of gravity of rainfall and flash floods revealed no overlap from 1979-1985 to 1986-1995. However, remarkable correlation was found between the migration trajectories of the center of gravity of precipitation and flash floods longitudinally, i.e., mainly from north to south. From 1986From -1995From to 1996From -2005, the center of gravity of flash floods migrated westward, whereas the center of gravity of rainfall migrated southwestward, which might reflect the impact of two heavy rainfall events (corresponding to the 1991 and 1998 floods) in the southeast of the area during this period. From 1996From -2005From to 2006From -2015, the centers of gravity migrated in very similar directions, i.e., from the southwest towards the northeast. These movements indicate excellent correlation between the trajectory of the center of gravity of flash floods and the trajectory of the center of gravity of precipitation, confirming that the occurrence of flash floods is affected substantially by changes in rainfall. Furthermore, the change in the underlying surface conditions was also found to affect the occurrence of flash floods. These results indicate that the spatiotemporal evolution of flash floods in the HMR is predominantly influenced by rainfall variation, which is subject to the effects of climate change.

Effects of land use on the spatiotemporal characteristics in flash floods
Land use change directly impacts the underlying surface of the drainage basin, generally altering the runoff generation and confluence processes at hillslopes (Elfert and Bormann 2010), which can affect the formation and development of flash floods. It was demonstrated that the sensitivity of flash floods increased in areas of artificial surfaces and cultivated land but decreased in areas of forests and grassland (Yue et al. 2016).
Forest and grassland can intercept rainfall and regulate runoff, efficiently lowering the peak discharge and delaying the peak time of flash floods. In contrast, the frequency of flash floods is higher in areas with artificial surfaces and cultivated land (He et al. 2005;Yue et al. 2016). For example, for the Chabagou catchment (205 km 2 ) in Shaanxi Province, China, it was reported that grassland and forest could reduce flood peaks by 36% and 64%, respectively (Fu et al. 2020). In central and northwestern parts of the HMR, large areas of land have been changed to forest and grassland, effectively alleviating the risk of flash flood occurrence in this region (Fig. 11). However, destruction of soil structure by agricultural practices and moistening of the soil via irrigation contribute to runoff intensification, thereby exacerbating the risk of extreme precipitation causing flash floods and associated soil erosion (Fu et al. 2020;O'Donnell et al. 2011). Cultivated land accounts for 7.7% of the total land use in the HMR, but 31.9% of flash floods occur in areas of cultivated land.

3
High surface runoff following heavy rainfall events, attributable to increases in the extent of impervious surfaces and artificial surfaces that change the hydrological response to higher flow rate peaks and shorter concentration times, escalates the risk of urban flooding (Abellan et al. 2019;Wan Mohtar et al. 2020). For example, the risk of flash floods in the Yesanpo Scenic Area of Beijing, China, has increased by 28% following tourism-related development and other land-use changes ). Even small areas of artificial surfaces could trigger the large probability of flash flood events. For example, artificial surfaces accounting for only 0.8% of the land use in the mountain area, but 4.9% of flash floods occur in such areas. The highest susceptibility to flash floods is in the large areas of cultivated land and artificial surfaces, much of which reflect conversion from forest and grassland, distributed in southern parts of the HMR where the land has low elevation and human activities are intense. This reflects the impact of the pattern of land-use change on the spatiotemporal evolution of flash floods in the HMR.

Conclusions
The complex spatiotemporal distribution of flash floods is affected both by rainfall and by the underlying surface conditions, which increases the difficulty regarding the development of related disaster prevention and mitigation measures. Therefore, on the basis of linear trend, wavelet analysis, center of gravity model, and standard deviational ellipse, this study characterized the spatiotemporal variation of 2,044 flash flood events that occurred in the high-elevation HMR during 1979-2015. Significant increase of 12 events/10 years was found in the occurrence of flash floods during the study periods, which was attributable to the high frequency and magnitude of heavy rainfall and the expansion of areas of cultivated land and artificial surfaces. In terms of the monthly distribution, 82% of flash flood events occurred in summer (June-August), almost entirely because of the concentration of rainfall, which is a feature of the monsoon climate, and only slightly because of land-use characteristics, which have little annual variance. It was noted that historical events were characterized by significant periodicities (3-7, 8-15, and 21-31 years), which were subject to changes in rainstorm extremes resulting from the impacts of climate change and human activity.
Under the influence of the spatial distribution of heavy rainfall, and the expansion of areas of cultivated land and artificial surfaces, the spatial center of gravity of flash floods during 1979-2015 apparently migrated from the northwest toward the southeast. The center of gravity of flash floods, which mainly moved in the direction of southeast-northwest and southwest-northeast from January-December, was affected by the periodic change of rainfall caused by the expansion and retreat of the southeast and southwest monsoons. Higher frequencies of flash flood events generally appeared in the south of the HMR, reflecting the uneven spatiotemporal distribution of flash flood events. These findings could help governmental policymakers identify and predict regions of the HMR at high risk of flash flood hazards under global change scenarios.