J. Meteor. Res.  2019, Vol. 33 Issue (1): 138-148 PDF
http://dx.doi.org/10.1007/s13351-018-8083-9
The Chinese Meteorological Society
0

#### Article Information

HE, Shuangshuang, Jun WANG, and Huijun WANG, 2019.
Projection of Landslides in China during the 21st Century under the RCP8.5 Scenario. 2019.
J. Meteor. Res., 33(1): 138-148
http://dx.doi.org/10.1007/s13351-018-8083-9

### Article History

Received May 21, 2018
in final form September 10, 2018
Projection of Landslides in China during the 21st Century under the RCP8.5 Scenario
Shuangshuang HE1,2, Jun WANG1,4, Huijun WANG1,2,3
1. Nansen–Zhu International Research Centre, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing 100029;
2. Collaborative Innovation Center on Forecast and Evaluation of Meteorological Disasters, Nanjing University of Information Science & Technology, Nanjing 210044;
3. Key Laboratory of Meteorological Disaster, Nanjing University of Information Science & Technology, Nanjing 210044;
4. University of Chinese Academy of Sciences, Beijing 100049
ABSTRACT: More and more rainstorms and other extreme weather events occur in the context of global warming, which may increase the risks of landslides. In this paper, changes of landslides in the 21st century of China under the high emission scenario RCP8.5 (Representative Concentration Pathway) are projected by using a statistical landslide forecasting model and the regional climate model RegCM4.0. The statistical landslide model is based on an improved landslide susceptibility map of China and a rainfall intensity–duration threshold. First, it is driven by observed rainfall and RegCM4.0 rainfall in 1980–99, and it can reproduce the spatial distribution of landslides in China pretty well. Then, it is used to forecast the landslide changes over China in the future under the RCP8.5 scenario. The results consistently reveal that landslides will increase significantly in most areas of China, especially in the southeastern, northeastern, and western parts of Northwest China. The change pattern at the end of the 21st century is generally consistent with that in the middle of the 21st century, but with larger increment and magnitude. In terms of the probability, the proportion of grid points that are very likely and extremely likely to experience landslides will also increase.
Key words: landslides     projection     statistical landslide forecasting model     regional climate model
1 Introduction

Global warming is unequivocal, as the Fifth Assessment Report (AR5) of the Intergovernmental Panel on Climate Change (IPCC) claimed, and it leads to changes in extreme weather and climate events. It is likely that the frequency and intensity of heavy precipitation events have increased in more land regions in recent years. Moreover, precipitation is the main triggering factor of landslides (Guzzetti et al., 2007). Therefore, it is imperative to project the changes of landslides in the context of global warming. While IPCC stated that heavy precipitation changes could affect landslides in some areas, it did not provide a global overview on landslides, indicating that projection of landslides is still difficult (Salciarini et al., 2016).

The term landslide refers to a slope failure caused by rainfall, rivers, groundwater activity, earthquake or artificial slope adjustment such as soil and rock sliding downslope due to gravity. Landslides occur in a very short time, destroying villages and towns and bringing serious loss of life and damage to infrastructure (Segoni et al., 2018). Given that landslides are widely distributed in China and can cause great damage and loss (Liu et al., 2013; Li et al., 2017), it is necessary to project the landslide trends in China in the future under the background of climate change. Moreover, in the last one hundred years, the average surface temperature has increased by 0.5–0.8°C in China, and the annual amount of precipitation experienced an increasing trend during 1856–2002 (Ding et al., 2006). An evaluation of the landslide changes in China under global warming is still absent.

To project the changes of landslides, a landslide forecast model and predicted rainfall data in the future are needed. Recent approaches to forecast landslide generally include empirical models (Caine, 1980; Guzzetti et al., 2008; Segoni et al., 2015) and physics based models (Fredlund et al., 1996; Iverson, 2000; Montrasio and Valentino, 2008; Liao et al., 2010). Although physical models are more reasonable in some aspects as they employ the physical processes in the model, they are only suitable for small spatial scales for they need detailed geological information and considerable computational resources (Kirschbaum et al., 2012). Therefore, an empirical statistical landslide model is used in this paper to project the changes of landslides in China.

Landslide occurrence depends on complex interactions among a large number of factors. These factors usually fall into two categories: 1) static factors, which are basic characteristics of the surfaces that determine landslide susceptibility, including slope, soil properties, elevation, aspect, land cover and lithology, 2) dynamic factors, including earthquake, heavy rainfall and glacier outburst (Dai et al., 2002). Since heavy rainfall is one of the main triggering factors of landslide and the predictability of other factors like earthquake is low, this study mainly focuses on the relationship between rainfall and landslide occurrence.

Static factors decide the places that are prone to landslides according to the landslide susceptibility (Bălteanu et al., 2010). Researchers have made considerable efforts to identify the static factors that determine landslide occurrence (Carrara et al., 1991; Lee and Min, 2001; Fabbri et al., 2003; Coe et al., 2004; Sarkar and Kanungo, 2004). Six factors (slope, soil type, soil texture, elevation, land cover and drainage density) that are closely associated with landslide occurrence are identified in previous studies (Hong et al., 2007). Satellite remote sensing data, geographic information system (GIS) techniques and weighted linear combination method are used to establish a global landslide susceptibility map.

Dynamic factors decide the time of landslide occurrence. Currently, rainfall intensity–duration threshold is most commonly used in predicting the time of landslide (Caine, 1980; Hong et al., 2005; Kirschbaum et al., 2012). Wang et al. (2016) developed a landslide susceptibility map for China (“location”), and combined it with the rainfall-threshold (“time”) in establishing a new landslide forecast model to detect landslide hazards as function of time and location in China.

The predicted rainfall data from a regional climate model RegCM4.0 are used in this paper, as many studies have found that the regional climate model is superior to the global climate model in the simulation of regional precipitation (Gao et al., 2012; Niu et al., 2015). IPCC released four Representative Concentration Pathways (RCPs) defined by their approximating total radiative forcing in 2100 relative to 1750. These four RCPs include one mitigation scenario that results in a very low forcing level RCP2.6, two stable scenarios RCP4.5 and RCP6, and one very high greenhouse gas emissions scenario RCP8.5 (IPCC, 2014). The RCP8.5 is used in this paper to project how landslides change under the high emission scenario.

Projecting future landslide changes is of significant importance in protecting human lives and properties, and it deserves considerable attention, especially in the context of global climate change. However, previous studies rarely include projections of future landslides in China. Therefore, by using an empirical statistical model of landslides and future rainfall data from a regional climate model, this paper aims to project landslide changes in China in the 21st century under the high emission scenario RCP8.5, and to provide a reference for the operational landslide forecast.

2 Data and model 2.1 Statistical landslide forecasting model

Wang et al. (2016) developed a new landslide susceptibility map of China, which performed better than the previous one in reflecting the spatial distribution of the observed landslide occurrences in China. By combining this susceptibility map with rainfall intensity–duration threshold, a statistical landslide model is set up. The conceptual framework of the statistical landslide model is presented in Fig. 1. The model calculates susceptibility by using the statistic factors of a region. In addition, empirical statistical methods are used to obtain the rainfall thresholds as the criteria for landslide occurrence in the area. Cases where precipitation exceeds thresholds in landslide-prone areas are considered as possible landslides.

 Figure 1 The framework of the statistical landslide model for rainfall-triggered landslides.

Wang et al. (2016) combined six factors (slope, elevation, soil type, soil texture, land cover and drainage density) to calculate the landslide susceptibility map of China (Fig. 2) with 1 km × 1 km resolution. The map indicates that landslides are likely to occur in the areas around southwest of China and the Tibetan Plateau. The rainfall is considered to meet the conditions for triggering landslides when the actual precipitation intensity in D hours is greater than the rainfall threshold I. Many studies on rainfall intensity–duration thresholds have been carried out (Jibson et al., 1989; Hong et al., 2005), and the global threshold developed by Hong et al. (2006) is used in this paper. The expression is:

 $I = 12.45 \times {D^{ - 0.42}}\;\;\left( {3 < D < 300} \right),$ (1)
 Figure 2 Landslide susceptibility map and the locations of eight sub-regions in China including the western part of Northwest China (NWW), the eastern part of Northwest China (NWE), Tibetan Plateau (TP), Southwest China (SW), Southeast China (SE), Yangtze River basin (YRB), North China (NC), and Northeast China (NE).

where I is the rainfall intensity (mm h–1), and D is the duration between the beginning of the rainfall and the landslide occurrence (unit: h). Taking landslide susceptibility S into account, the possibility of landslide PS can be determined by:

 $P_S = \frac{{{R_{\rm{d}}}}}{I} \times S,$ (2)

where Rd is the actual rainfall intensity (mm h–1), S is the landslide susceptibility, and PS is the index describing the possibility of landslides. The landslide occurrence can be predicted based on the value of PS (PS ≥ 6, extremely likely; 5 ≤ PS < 6, very likely; 4 ≤ PS < 5, likely; and PS < 4, rarely likely).

The statistical landslide forecasting model has been evaluated in previous studies. Hong and Adler (2008) collected 25 landslides happened in 1998–2006 from news reports, 19 landslides were predicted successfully; Wang et al. (2016) verified and analyzed 37 landslides during the period of 2008–11, 32 were successful predicted, indicating that this model is certainly capable on simulating landslides.

2.2 Data

Daily accumulated precipitation dataset based on interpolation of data from more than 2400 observational stations in China (hereafter observed rainfall) is used in this study to test the performance of the statistical landslide model driven by regional climate model rainfall (Wu and Gao, 2013). The model has a spatial resolution of 0.25° longitude by 0.25° latitude, covering the period 1961–2007.

Model rainfall data are collected from a regional climate model RegCM4.0 (hereafter RegCM4.0 rainfall), driven by the global model of Beijing Climate Center Climate System Model version 1.1 (BCC_CSM1.1). The final model output has a horizontal resolution of 50 km and covers the period of 1961–2099. Model output from 1961 to 2005 is considered as the historical simulation results, whereas the model output from 2006 to 2100 is taken as the future projection under the high emission scenario RCP8.5. Present-day observed and simulated annual mean precipitation is compared to assess the model performance by using this regional model. In addition, the simulated rainfall spatial pattern is a good representation of the observed data (Gao et al., 2013).

Since it was quite difficult to obtain all the landslide events in China, only 322 landslide events in 1991–99 were collected from literature and online news (Ma et al., 2009). The data of landslide events includes time and location. The failure location of city is known for 6 events, the county is known for 31 landslides events, and the information of village, country or site for the rest 285 events is available. The landslides collected in this study are mainly large scale. This landslide inventory will be compared with the simulated distribution of landslides to provide a reference for the performance of the model.

To facilitate the analysis, model outputs and observed precipitation are interpolated to a common grid with a resolution of 0.1° longitude by 0.1° latitude. 1980–99 represents the present day, 2040–59 represents the middle of the 21st century, and 2080–99 represents the end of the 21st century.

3 Validation of the present-day simulation

To validate the performance of the statistical landslide forecasting model, the simulated landslide distribution is compared with that in the landslide inventory. Figure 3 shows the simulated distribution from 1991 to 1999 by using observed rainfall. As the landslide inventory does not contain all of the landslide events during 1991–99, Fig. 3 only reflects the qualitative performance of the landslide model. The areas with landslide records in the inventory are analyzed, while some landslides in other areas may not be catalogued in the inventory. As shown in Fig. 3, the actual distribution of landslides during 1991–99 shows consistence with the simulated result in the southwest of China, southeast coast and north of China. However, the model does not successfully simulate the landslide occurrence in Xinjiang Region and west of China. Overall, the simulated result modeled by observed rainfall is consistent with the landslide distribution in the areas with landslide records. Therefore, to quantitatively analyze the performance of the model dri-ven by RegCM4.0 rainfall, the results simulated by observed rainfall can be considered as a reference of “landslide inventory” to provide the reliability of projections.

 Figure 3 Landslide inventory from 1991 to 1999 and the simulated frequency distribution of landslides in China in 1980–99 driven by observed rainfall.

Figure 4 shows the spatial distribution of landslides from 1980 to 1999 by adding up the frequency on each 0.5° × 0.5° grid, and the simulations use both the observed rainfall and RegCM4.0 rainfall to drive the statistical landslide model. Both of the simulations show that the landslides are concentrated in the southwestern and southeastern parts of China and the lower reaches of the Yangtze River. High occurrences are seen along the southeastern coast of China and the area surrounding Sichuan basin, while very few landslides occur in northwest of China (Fig. 4a). These areas are also risk zones for high landslide occurrence according to historical records. Besides, the landslides simulated by RegCM4.0 rainfall cover larger areas, and the simulated frequency is higher, especially at the edge of the Tibetan Plateau (Fig. 4b). The spatial correlation coefficient (SCC) of the simulated landslide occurrence between observed rainfall and RegCM4.0 rainfall is 0.33. In general, the landslide model driven by RegCM4.0 rainfall can capture the spatial pattern reasonably well.

 Figure 4 Simulated distributions of landslides in China during 1980–99 driven by observed rainfall for (a) PS ≥ 4, (c) 4 ≤ PS < 5, (e) 5 ≤ PS < 6, and (g) PS ≥ 6; and by RegCM4.0 rainfall for (b) PS ≥ 4, (d) 4 ≤ PS < 5, (f) 5 ≤ PS < 6, and (h) PS ≥ 6. The plotted variable x is the frequency on a 0.5° × 0.5° grid. The spatial correlation coefficients (SCC) between the simulation results using observed rainfall and RegCM4.0 rainfall are labelled on the top center of the figures.

As mentioned in Section 2.1, the PS in the statistical landslide model describes the three likelihood levels of landslides, based on which the result is further analyzed. As shown in Table 1, the proportions of grid points with 4 ≤ PS < 5 (likely), 5 ≤ PS < 6 (very likely), and PS ≥ 6 (extremely likely) are 85.22%, 12.54%, and 2.23%, respectively, when driven by observed; and are 82.22%, 13.84%, and 3.94%, respectively, when driven by RegCM4.0 rainfall. The proportions of these three likelihood levels are essentially the same in the two simulations with the deviation less than 3%. Both of the simulations reveal that 4 ≤ PS < 5 accounts for the largest proportion and covers the largest area. The spatial distribution of this likelihood level essentially coincides with the area of PS ≥ 4 (Fig. 4c, d). Areas with 5 ≤ PS < 6 are mainly distributed along the southeastern coast of China, the southern part of Yangtze River Valley and the lower reaches of the Yellow River ( Fig. 4e, f). The proportion of areas with PS ≥ 6 is minimal, which is mainly concentrated along the southeastern coast of China, the middle and lower reaches of the Yangtze River, and the lower reaches of the Yellow River (Fig. 4g, h). The spatial correlation coefficients of the three different likelihood levels of landslide occurrence between the two simulations are 0.37 (4 ≤ PS < 5), 0.52 (5 ≤ PS < 6), and 0.69 ( PS ≥ 6), respectively. Thus, the results simulated by the landslide model driven by RegCM4.0 rainfall perform well when the landslide distribution simulated by the landslide model driven by observed rainfall is taken as the reference.

Table 1 The percentages of different likelihood levels of landslides
 PS Driven by observed rainfall Driven by RegCM4.0 rainfall Observed – simulated 4 ≤ PS < 5 85.22% 82.22% 3% 5 ≤ PS < 6 12.54% 13.84% –1.3% PS ≥ 6 2.23% 3.94% –1.71%
4 Future projection 4.1 Changes in spatial distribution

The statistical landslide model in this paper is driven by RegCM4.0 rainfall under the RCP8.5 scenario to study the changes of landslide in China in the future. Figure 5 shows the simulated percent change in landslides during the middle and the end of the 21st century by using RegCM4.0 rainfall under the RCP8.5 scenario. As shown in Fig. 5a, during the middle of the 21st century, the landslide occurrences in most areas are projected to increase over 50% when compared with the present day (1980–99). In general, the frequency of landslides is projected to significantly increase, especially in Southwest China, the Yangtze River basin, and Northeast China. At the end of the 21st century, the spatial distribution of landslides (Fig. 5b) is projected to be similar to the pattern in the middle of the 21st century. However, the increase in the magnitude tends to be larger, and the increasing grid points tend to be greater than that in the middle of the 21st century. Table 2 shows that the percentages of increasing and decreasing grid points are 59.30% and 40.70% in the middle of the century, 65.58% and 34.42% at the end of the 21st century, respectively.

Table 2 The percentages of increasing and decreasing grid points with different likelihood levels of landslide occurrence in the 21st century of China projected by using RegCM4.0 rainfall
 Period The middle of the 21st century The end of the 21st century PS Increase Decrease Increase Decrease PS ≥ 4 59.30% 40.70% 65.58% 34.42% 4 ≤ PS < 5 59.06% 40.94% 63.21% 36.79% 5 ≤ PS < 6 55.75% 44.25% 62.16% 37.84% PS ≥ 6 56.29% 43.71% 66.76% 33.24%

Figure 5 also shows the percentage changes in different likelihood levels of landslides in the middle and the end of the 21st century, and the corresponding increasing and decreasing percentage of grid points are shown in Table 2. The increasing grid points with 4 ≤ PS < 5 ( Fig. 5c) account for 59.06% in the middle of the century, while the decreasing grid points account for 40.94%. In addition, magnitude associated with the increasing grid points is much larger than that associated with the decreasing ones. At the end of the century (Fig. 5d), the increasing and decreasing grids of 4 ≤ PS < 5 account for 63.21% and 36.79%, respectively. The increasing grids are 4% more than those in the middle of 21st century. The percentages of increasing and decreasing grid points with 5 ≤ PS < 6 are 55.75% and 44.25% in the middle of the century ( Fig. 5e), and are 62.16% and 37.84% at the end of the century (Fig. 5f), respectively. A significant increasing trend is shown in southwest of China and along the southeastern coast of China. The percentages of increasing and decreasing grid points with PS ≥ 6 are 56.29% and 43.71% in the middle of the century, respectively, mainly in Southwest of China, along the southeastern coast of China, and along the lower reaches of the Yangtze and Yellow Rivers (Fig. 5g). At the end of the century (Fig. 5h), the increasing and decreasing grids are 66.76% and 33.24% respectively. The percentages of increasing grid points for all of the likelihood levels of landslides at the end of the 21st century are higher than those in the middle of the 21st century. Thus, the risks of landslides will be larger at the end of 21st century.

 Figure 5 Landslide percentage changes in 2040–59 for (a) PS ≥ 4, (c) 4 ≤ PS < 5, (e) 5 ≤ PS < 6, and (g) PS ≥ 6; and in 2080–99 for (b) PS ≥ 4, (d) 4 ≤ PS < 5, (f) 5 ≤ PS < 6, and (h) PS ≥ 6. The plotted variable x is the percentage change in frequency when compared with the present day on a 0.5° × 0.5° grid.

By using a regional climate model to drive the statistical landslide model, the percentages of different probabilities of landslide occurrence in the 21st century in China are shown in Table 3. Associated with the three likelihood levels, i.e. likely, very likely, and extremely likely, the percentages are 83.12%, 13.26% and 3.61% in the middle of the 21st century, and are 79.94%, 15.48% and 5.05% at the end of the 21st century, respectively. Compared with the middle of the 21st century, the number of grid points where landslides are very likely and extremely likely to occur are projected to increase by 2.22% and 1.44% at the end of the 21st century, which indicates that more areas will be exposed to high risks of landslides.

Table 3 The percentages of different likelihood levels of landslides in the 21st century of China projected by using RegCM4.0 rainfall
 PS The middle of the 21st century The end of the 21st century Middle – end 4 ≤ PS < 5 83.12% 79.46% –3.62% 5 ≤ PS < 6 13.26% 15.48% 2.22% PS ≥ 6 3.61% 5.05% 1.44%
4.2 Temporal variation projections

Figure 6 shows the annual total frequency from 2006 to 2099 of landslides simulated by the statistical landslide model driven by RegCM4.0 rainfall under the RCP8.5 scenario. As shown in Fig. 6a, landslides show a general significant increasing trend (a t-test yields the p-value of 0.01 or less) in China. This increasing trend is especially significant during the periods 2020–40 and 2080–90 with strong interannual variability and interdecadal oscillations. The frequencies of different likelihood levels of landslides are shown in Fig. 6. All of the three likelihood levels of landslides are projected to increase. The likelihood level with 4 ≤ PS < 5 is projected to increase significantly between 2020 and 2030 ( Fig. 6b). The increasing trend in 5 ≤ PS <6 is most significant in 2020–30 and 2070–90 ( Fig. 6c), and the likelihood level with PS ≥ 6 is projected to increase considerably in 2070–90 (Fig. 6d).

 Figure 6 Annual frequency of landslides from 2006 to 2099 simulated by the landslide model driven by RegCM4.0 rainfall under the RCP8.5 scenario for (a) PS ≥ 4, (b) 4 ≤ PS < 5, (c) 5 ≤ PS < 6, and (d) PS ≥ 6. The blue line denotes the frequency, the red line is 3-yr moving average, and the black line is the trend.

However, given the complex topography in China, the landslide distribution varies among regions. The method of Chen et al. (2012) to divide China into eight sub-regions is adopted in this paper to analyze the changes of landslides in different regions. The eight sub-regions are the western part of Northwest China (NWW), the eastern part of Northwest China (NWE), Tibetan Plateau (TP), Southwest China (SW), Southeast China (SE), Yangtze River basin (YRB), North China (NC) and Northeast China (NE) (Fig. 2). In addition, the trends of landslides in the eight sub-regions are analyzed. Figure 7 shows the time series of annual landslide frequency and the trends in the eight sub-regions in 2006–99. In terms of frequency, landslides are projected to be concentrated in Southwest China, Southeast China, and Yangtze River basin, and significant increasing trends are in Southeast China (Fig. 7e), Northeast China (Fig. 7h), and the western part of Northwest China (Fig. 7a). The statistical data have passed the 0.01 significance level of a t-test. The landslides in North China (Fig. 7g) and Yangtze River basin (Fig. 7f) also show an increasing trend, but it is not as significant as that mentioned previously. Tibetan Plateau (Fig. 7c) and Southwest China (Fig. 7d) show a decreasing trend. There is no significant trend but a strong interdecadal oscillation in eastern part of Northwest China (Fig. 7b).

 Figure 7 The annual frequency of landslides from 2006 to 2099 in eight sub-regions: (a) the western part of Northwest China (NWW), (b) the eastern part of Northwest China (NWE), (c) Tibetan Plateau (TP), (d) Southwest China (SW), (e) Southeast China (SE), (f) Yangtze River basin (YRB), (g) North China (NC), and (h) Northeast China (NE). The blue line denotes the frequency, the red line is 3-yr moving average, and the black line is the trend.
5 Discussion

This paper provides a preliminary method for landslide projection. It can be seen that the risks of landslides will largely increase at the end of 21st century in China by using the landslide model that combines the landslide susceptibility, the rainfall threshold, and the projected rainfall from RegCM4.0.

The changes of rainfall have huge impacts on landslides since rainfall is the main triggering factor of landslides. Gao et al. (2013) suggested a general increase in precipitation based on simulations of RegCM4.0 from 2080 to 2099, especially in the western part of North-west China (increase of 25%–50%), Northeast China (5%–10%), and the Yangtze River basin (5%), which is consistent with the change pattern of landslides revealed in this study. In addition, Southwest China shows a decrease by about 5%–10%. The changes in precipitation in North China and Tibetan Plateau are not consistent, and there is no significant trend of landslides in these regions, either. Thus, the frequency of landslide change is consistent with the precipitation change pattern as expected. However, only the rainfall is considered here, while the change of geological environment and human activities are not considered. Moreover, as the landslide inventory is incomplete, only visual comparison is made without quantitative analyses. Because the landslide records and the simulated landslide distribution based on observed rainfall are consistent in some areas, the latter is used in this paper as baseline to validate the simulated results driven by RegCM4.0 rainfall. However, if more landslides data become available, the model can be better validated.

There is still so much to be done for a better projection of landslides, such as cataloging a more complete landslide inventory to obtain a better justification of the model, and using the ensemble model rainfall to drive the landslide model under different scenarios to provide more comprehensive results.

6 Conclusions

Future changes of landslide have a critical impact on people’s life and the nation’s economy. Thus, based on the statistical landslide model and the regional climate model RegCM4.0, this paper first evaluated the performance of the statistical landslide model driven by RegCM4.0 rainfall by comparing its simulation with that using observed rainfall. Then, the RegCM4.0 rainfall under the RCP8.5 scenario was used to drive the statistical landslide model to project future changes of landslides in China. The main conclusions are as follows.

(1) The simulated landslide distribution shows consistence with landslide inventory in most areas with landslide records.

(2) The spatial correlation coefficients of different likelihood levels of landslides between simulations using observed rainfall and RegCM4.0 rainfall to drive the model are 0.37, 0.52, and 0.69 respectively. Thus, the statistical landslide model driven by RegCM4.0 rainfall can capture the spatial pattern of different likelihood levels of landslides pretty well.

(3) Projected results driven by RegCM4.0 rainfall under the RCP8.5 scenario show significant increasing trends of landslides, especially in Southwest China, Northeast China, and the western part of Northwest China.

(4) At the end of the 21st century, the increasing trend will be more evident, and the projections reflect more increasing grid points with larger magnitudes than in the middle of the 21st century. The proportion of points where landslides are very likely and extremely likely to occur will also increase.

In general, this paper makes a preliminary projection of the changes in landslides by using a statistical model and rainfall from a regional climate model. Landslide events are projected to significantly increase and to become much extremer at the end of the 21st century. Thus, it is crucial to propose more effective measures to reduce the landslide risks. And it is also important to develop more accurate landslide forecast systems to minimize the damage from landslides. As it is obvious that the quality of future rainfall estimates plays an important role in the evaluation, this paper adopts rainfall derived from a single set of model output that uses a high emission scenario. In future work, we will consider model ensembles or higher resolution rainfall data, as well as different emission scenarios to provide a more comprehensive evaluation for the future changes of landslide in China.

Acknowledgments. We thank Xuejie Gao for assistance with the data.

References
 Bălteanu, D., V. Chendeş, M. Sima, et al., 2010: A country-wide spatial assessment of landslide susceptibility in Romania. Geomorphology, 124, 102–112. DOI:10.1016/j.geomorph.2010.03.005 Caine, N., 1980: The rainfall intensity–duration control of shallow landslides and debris flows. Geogr. Ann., 62, 23–27. DOI:10.1080/04353676.1980.11879996 Carrara, A., M. Cardinali, R. Detti, et al., 1991: GIS techniques and statistical models in evaluating landslide hazard. Earth Surf. Proc. Land., 16, 427–445. DOI:10.1002/esp.3290160505 Chen, H. P., J. Q. Sun, and X. L. Chen, 2012: The projection and uncertainty analysis of summer precipitation in China and the variations of associated atmospheric circulation field. Climatic Environ. Res., 17, 171–183. DOI:10.3878/j.issn.1006-9585.2011.10137 Coe, J. A., J. W. Godt, R. L. Baum, et al., 2004: Landslide susceptibility from topography in Guatemala. Landslides: Evaluation and Stabilization, W. A. Lacerda, M. Ehrlich, S. A. B. Fontoura, et al., Eds., Taylor & Francis, London, 69–78. Dai, F. C., C. F. Lee, and Y. Y. Nagi, 2002: Landslide risk assessment and management: An overview. Eng. Geol., 64, 65–87. DOI:10.1016/S0013-7952(01)00093-X Ding, Y. H., G. Y. Ren, G. Y. Shi, et al., 2006: National assessment report of climate change (I): Climate change in China and its future trend. Progressus Inquisitiones de Mutatione Climatis, 2, 3–8. Fabbri, A. G., C. J. F. Chung, A. Cendrero, et al., 2003: Is prediction of future landslides possible with a GIS?. Nat. Hazards, 30, 487–503. DOI:10.1023/B:NHAZ.0000007282.62071.75 Fredlund, D. G., A. Q. Xing, M. D. Fredlund, et al., 1996: The relationship of the unsaturated soil shear to the soil-water characteristic curve. Can. Geotech. J., 33, 440–448. DOI:10.1139/t96-065 Gao, X. J., Y. Shi, D. F. Zhang, et al., 2012: Climate change in China in the 21st century as simulated by a high resolution regional climate model. Chinese Sci. Bull., 57, 1188–1195. DOI:10.1007/s11434-011-4935-8 Gao, X. J., M. L. Wang, and F. Giorgi, 2013: Climate change over China in the 21st century as simulated by BCC_CSM1.1-RegCM4.0. Atmos. Ocean. Sci. Lett., 6, 381–386. DOI:10.3878/j.issn.1674-2834.13.0029 Guzzetti, F., S. Peruccacci, M. Rossi, et al., 2007: Rainfall thresholds for the initiation of landslides in central and southern Europe. Meteor. Atmos. Phys., 98, 239–267. DOI:10.1007/s00703-007-0262-7 Guzzetti, F., S. Peruccacci, M. Rossi, et al., 2008: The rainfall intensity–duration control of shallow landslides and debris flows: An update. Landslides, 5, 3–17. DOI:10.1007/s10346-007-0112-1 Hong, Y., and R. F. Adler, 2008: Predicting global landslide spatiotemporal distribution: Integrating landslide susceptibility zoning techniques and real-time satellite rainfall estimates. Int. J. Sediment Res., 23, 249–257. DOI:10.1016/S1001-6279(08)60022-0 Hong, Y., H. Hiura, K. Shino, et al., 2005: The influence of intense rainfall on the activity of large-scale crystalline schist landslides in Shikoku Island, Japan. Landslides, 2, 97–105. DOI:10.1007/s10346-004-0043-z Hong, Y., R. Adler, and G. Huffman, 2006: Evaluation of the potential of NASA multi-satellite precipitation analysis in global landslide hazard assessment. Geophys. Res. Lett., 33, L22402. DOI:10.1029/2006GL028010 Hong, Y., R. Adler, and G. Huffman, 2007: Use of satellite remote sensing data in the mapping of global landslide susceptibility. Nat. Hazards, 43, 245–256. DOI:10.1007/s11069-006-9104-z IPCC, 2014: Climate Change 2014: Synthesis Report. Contribution of Working Groups I, II and III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change, R. K. Pachauri, et al., Eds., IPCC, Geneva, Switzerland, 151 pp. Iverson, R. M., 2000: Landslide triggering by rain infiltration. Water Resour. Res., 36, 1897–1910. DOI:10.1029/2000WR900090 Jibson, R. W., 1989: Debris flows in southern Puerto Rico. Spec. Pap.-Geol. Soc. Am., 236, 29–56. DOI:10.1130/SPE236-p29 Kirschbaum, D. B., R. Adler, Y. Hong, et al., 2012: Advances in landslide nowcasting: Evaluation of a global and regional modeling approach. Environ. Earth Sci., 66, 1683–1696. DOI:10.1007/s12665-011-0990-3 Lee, S., and K. Min, 2001: Statistical analysis of landslide susceptibility at Yongin, Korea. Environ. Geol., 40, 1095–1113. DOI:10.1007/s002540100310 Li, W. Y., C. Liu, M. Scaioni, et al., 2017: Spatiotemporal analysis and simulation on shallow rainfall-induced landslides in China using landslide susceptibility dynamics and rainfall I-D thresholds. Sci. China Earth Sci., 60, 720–732. DOI:10.1007/s11430-016-9008-4 Liao, Z. H., Y. Hong, J. Wang, et al., 2010: Prototyping an experimental early warning system for rainfall-induced landslides in Indonesia using satellite remote sensing and geospatial datasets. Landslides, 7, 317–324. DOI:10.1007/s10346-010-0219-7 Liu, C., W. Y. Li, H. B. Wu, et al., 2013: Susceptibility evaluation and mapping of China’s landslides based on multi-source data. Nat. Hazards, 69, 1477–1495. DOI:10.1007/s11069-013-0759-y Ma, L., P. Cui, G. B. Zhou, et al., 2009: Geological Meteorological Hazard. China Meteorological Press, Beijing, 156 pp. (in Chinese) Montrasio, L., and R. Valentino, 2008: A model for triggering mechanisms of shallow landslides. Nat. Hazards Earth Syst. Sci., 8, 1149–1159. DOI:10.5194/nhess-8-1149-2008 Niu, X. R., S. Y. Wang, J. P. Tang, et al., 2015: Multimodel ensemble projection of precipitation in eastern China under A1B emission scenario. J. Geophys. Res. Atmos., 120, 9965–9980. DOI:10.1002/2015JD023853 Salciarini, D., E. Volpe, S. A. Kelley, et al., 2016: Modeling the effects induced by the expected climatic trends on landslide activity at large scale. Procedia Eng., 158, 541–545. DOI:10.1016/j.proeng.2016.08.486 Sarkar, S., and D. P. Kanungo, 2004: An integrated approach for landslide susceptibility mapping using remote sensing and GIS. Photogramm. Eng. Rem. S., 70, 617–625. DOI:10.14358/PERS.70.5.617 Segoni, S., D. Lagomarsino, R. Fanti, et al., 2015: Integration of rainfall thresholds and susceptibility maps in the Emilia Romagna (Italy) regional-scale landslide warning system. Landslides, 12, 773–785. DOI:10.1007/s10346-014-0502-0 Segoni, S., L. Piciullo, and S. L. Gariano, 2018: A review of the recent literature on rainfall thresholds for landslide occurrence. Landslides, 15, 1483–1501. DOI:10.1007/s10346-018-0966-4 Wang, J., H. J. Wang, and Y. Hong, 2016: A Realtime Monitoring and Dynamical Forecasting System for Floods fand Landslides in China. China Meteorological Press, Beijing, 164 pp. (in Chinese) Wu, J., and X. J. Gao, 2013: A gridded daily observation dataset over China region and comparison with the other datasets. Chinese J. Geophys., 56, 1102–1111. DOI:10.6038/cjg20130406