中国是滑坡灾害影响最严重的国家之一，受气候变化等因素的影响，未来滑坡灾害的威胁还将持续存在。但是，气候变化对滑坡影响的评估研究，一直是滑坡研究领域的难点，目前学界也还未提出统一适用的评估方法。在全国尺度上如何评估气候变化对中国滑坡易发性的时空影响，从滑坡灾害数据库编制、滑坡易发性评估模型构建和全球气候模式数据的应用模拟及集成等方面，都是目前面临研究难题。
为应对以上挑战，本文首先基于多种数据来源建立长时间序列中国灾难性滑坡事件数据库，分析中国历史灾难性滑坡事件的时空分布规律和影响因素。其次，基于滑坡发生影响因素的条件密度分析，应用非线性混合效应统计模型方法，探索如何在有完整性偏差分布的滑坡编目数据的基础上，建立全国尺度统计滑坡易发性评估模型，识别滑坡高易发性区域。在此基础上，系统集成历史滑坡灾害数据、滑坡易发性评估模型和全球气候多模式数据，提出一个面向国家尺度的未来气候变化情景降雨滑坡易发性评估模型，在全国尺度上从滑坡易发性空间格局变化、高易发区滑坡频率变化，评估了气候变化对中国滑坡发生的影响。论文主要结论如下：
1. 1950-2016年全国记录的灾难性滑坡事件1949起，共造成28,815人死亡。灾难性滑坡事件的发生频次及其造成的死亡人数总体呈增加趋势，空间上呈南多北少的分布格局，多年平均降水和地形因素的交互作用对其空间分布影响最大。
不同等级的灾难性滑坡事件在时间变化趋势上有明显差异，特大型灾难性滑坡在1950-1999年间呈现上升趋势，2000年之后有一定的减少趋势；中小型灾难性滑坡事件在1950-2016年间呈显著上升趋势，2000年之后其上升趋势更为显著。
灾难性滑坡事件的年内分布集中于4-9月，占全年的82.1%。空间分布上，大部分灾难性滑坡在青藏高原东缘的一级阶梯和二级阶梯交界处，二级阶梯和三级阶梯交界处的黄土高原和长江中游地区，四川盆地及其周围山区，云贵高原地区，东南山地丘陵地区；少部分分布于西北天山地区，昆仑山西段以及东北长白山地区。
基于地理探测器方法，分析中国灾难性滑坡事件的空间分布影响因素，结果表明，单因素上，多年平均降水与其空间分布格局关系最密切；交互因素上，降水与地形交互作用对滑坡灾难事件空间格局的影响最大，显著大于单一影响因素。
2. 建立了基于非线性混合效应方法的全国尺度滑坡易发性评估模型，可以减小滑坡编目数据库偏差造成的影响，提高了影响因素关系模拟和滑坡易发性分布预测的精度。
滑坡编目完整性分布偏差是指由于滑坡数据获取方式或研究区内的地域特征差异等原因导致的滑坡在特定属性区域存在明显的缺失，从而使得整个滑坡编目数据存在完整性分布偏差。例如，全国尺度的滑坡编目数据因青藏高原地区调查条件限制，在青藏高原周缘地区会存在一定的数据不完备问题。在全国尺度上，这种滑坡编目数据完整性分布偏差的影响往往不可避免但却容易被忽略。
本文将与滑坡编目数据完整性偏差有关的影响因素作为随机效应项，引入非线性混合效应统计方法建立的全国尺度滑坡易发性模型，一方面可以有效解决滑坡编目分布偏差对影响因素造成混杂影响，在本研究中，避免了由于未考虑与滑坡编目分布偏差有关的影响因素导致的与实际地貌过程规律相悖的现象，即地形起伏度与滑坡发生的关系由正向影响变为负向影响。同时，也能得到较好的滑坡易发性空间分布结果，避免了明显与地貌特征不符的滑坡易发性图空间分布。模型结果与基准模型相比，交叉验证AUROC接近0.8，模拟表现较好，并且模型得到的滑坡易发性空间格局与地貌分布特征相符，局部细节模拟更优。滑坡极高和高易发性区域主要分布于东南沿海山地丘陵区、西南云贵高原山区、四川盆地周围山区、青藏高原东部山区以及昆仑山和祁连山区、西北天山山区、东北大兴安岭和长白山区等，这些高易发区是未来更需要重视并加强滑坡灾害风险评估和管理的区域。本研究采用的非线性混合效应模型，可以在未来应用于其他研究区以解决类似问题。
3. 全国尺度上，未来气候变化情景下各滑坡易发性等级的面积呈现“高增低减”的变化趋势，空间分布上存在显著东西分异性，青藏高原东部和北部滑坡易发性等级变高、西南部易发性等级变低；二、三级阶梯上东部大部分地区滑坡易发性等级变高，西南和西北地区易发性等级不变。
未来气候变化情景下中国滑坡易发性的变化在各易发性等级上存在差异。与历史时期的滑坡易发性相比，极高滑坡易发性区域的面积呈增加趋势，而极低和低滑坡易发性呈减少趋势，中等和高易发性区域的面积呈小幅波动变化。
未来气候变化情景下中国滑坡易发性的变化在不同情景和不同时期存在差异。21世纪末期(2066-2095年)比21世纪中期(2031-2060年)变化更大；典型浓度路径(Representative Concentration Pathway, RCP) RCP8.5情景比RCP4.5情景发生变化更大。21世纪末期RCP8.5情景下多模式中位数平均结果表明，极高易发性区域面积增多53.61%；极低易发性区域面积减少35.99%。
未来气候变化情景下中国滑坡易发性的变化有明显的空间分异。中国东部大部分地区、青藏高原北部和东部、天山山区东部滑坡易发性等级变高；西南地区和西北地区滑坡易发性等级基本不变；青藏高原西南部和西北边境区域滑坡易发性等级变低。
4. 全国尺度上，未来气候变化情景下中国高易发区的滑坡发生频率总体呈现多模式一致增多特征。空间上有显著区域差异，东南沿海丘陵地区呈轻微增加，其他区域呈显著增加。季节分布上呈双峰型分布，春季和秋季滑坡增幅较大。
21个全球气候模式在不同情景和年代下的预测结果均表明，气候变化影响下中国未来滑坡发生频率与基准历史时期相比会增多，21世纪末期2066-2095年多模式中位数平均结果表明，RCP4.5和RCP8.5情景下中国滑坡发生频率与基准时期相比分别增多19.9%和33.2%。
气候变化对中国滑坡发生影响存在明显的区域分异，西北天山山区，青藏高原北部昆仑山区和祁连山区，青藏高原东南部，川西高原，云贵高原，四川盆地山区和太行山区，这些区域未来滑坡发生呈明显增加趋势且多模式预测结果高度一致。东南沿海丘陵地区则主要呈减少或微弱增加的趋势，且多模型一致性评估呈不确定性。
气候变化对未来滑坡发生的影响有明显的季节分异性，各月份的滑坡发生变化呈双峰型特征，在春季和秋季滑坡增加幅度最大，其次是夏季，冬季变化不明显。本研究构建基于多气候模式和相对变化的评估研究框架可以在未来应用于其他区域气候变化对滑坡影响的研究。
综上，本文基于地质灾害成灾机理，构建面向国家尺度的未来气候变化情景降雨滑坡易发性评估模型，通过系统集成全国历史滑坡灾害数据、滑坡易发性影响因素、多气候模式降雨数据，完成了中国降雨滑坡高易发性区域识别、基于21个气候模式的未来气候变化情景下中国滑坡易发性空间格局分布预估、高易发区滑坡频率的变化预估，以及相应的不确定分析。未来气候变化情景下我国青藏高原北部和东部、天山山区东部滑坡易发性等级变高，且滑坡发生频率呈多模式一致性增加趋势；西南地区和西北地区滑坡易发性等级基本不变；青藏高原西南部和西北边境区域滑坡易发性等级变低。论文的主要创新点有：(1) 构建了基于非线性混合效应方法的全国尺度滑坡易发性评估模型，显著减小滑坡编目数据库偏差造成的影响，相比传统的直接引入各影响因素的基准模型，该模型可以改进对滑坡影响因素关系的模拟和滑坡易发性分布的预测。(2) 完成基于NEX-GDDP 21个气候模式数据的21世纪中期、远期中国降雨滑坡易发性空间格局分布预估、高易发区滑坡频率变化预估(空间分辨率0.25o)，以及相应的不确定分析研究，为减轻未来滑坡灾害风险提供决策支持。(3) 整理建设了多来源长时间的中国灾难性滑坡事件数据库，与历史降雨数据匹配，挖掘中国历史灾难性滑坡的时空分布规律和影响因素。

China is one of the countries that experience a severe loss of life as a result of landslides. There is a high confidence that landslides might continue to cause substantial human loss and damage to property in the future due to climate change. However, research on the climate change impact on landslides has always been one of the most challenging issues in the field of landslides studies. There is no consensus on the assessment framework among scientists. How to assess the spatiotemporal change of landslide susceptibility under climate change scenarios at the national scale? The main challenges included compilation of long-term historical landslide inventory, modelling of national landslide susceptibility, coupling of GCMs with historical landslide inventory and landslide susceptibility model. 
In this thesis, a long time series fatal landslides inventory of China is compiled based on multiple data sources and the spatiotemporal pattern and influencing factors of fatal landslide events are analyzed in detail. Then, based on the heuristic analysis of landslide influencing factors, a more realistic landslide susceptibility model is built and highly susceptible areas are identified by using nonlinear mixed effects models to reduce inventory-based biases. Finally, the historical landslide inventory, the landslide susceptibility model and GCMs are coupled systematically. The framework of investigating the climate change impact on landslides by assessing the change in spatial pattern of landslide susceptibility and projecting the change in landslide occurrence in highly susceptible areas is presented; the impact of climate change on landslides in China is assessed on a national scale. The main conclusions are as follows:
1. During the period from 1950 to 2016, there are 1949 reported fatal landslide events in China, resulting in 28,815 deaths. The occurrence of fatal landslides and the number of deaths show a significant increasing trend. There are more fatal landslides and fatalities in southern than in northern China; the interacting factor between annual average precipitation and topography is dominant in shaping this spatial pattern.
The occurrence of fatal landslides presents significantly different trends for different grades of events. Very large fatal landslide events (fatalities>?=?30) were on the rise during 1950-1999 and declined from 2000 to 2016. The small and medium-sized fatal landslide events (fatalities<?10) show a significant increasing trend between 1950 and 2016, especially during the period of 2000–2016.
The fatal landslides mainly occurred between April and September (82.1%). Spatially, hotspots of fatal landslides is in 14 provinces: five southwestern provinces (Yunnan, Sichuan, Guangxi, Guizhou, and Chongqing), five southeastern provinces (Hunan, Guangdong, Fujian, Jiangxi, and Zhejiang), Shaanxi, Shanxi, Hubei and Gansu.
The spatial association between the fatal landslide density and possible influencing factors is assessed based on a geographical detector method. The results show that annual average precipitation is the most closely-related singe factor and interacting factor between the precipitation and topography is more closely related to the spatial distribution of fatal landslides than the individual factor.
2. The inventory-based biases have a significant impact on landslide susceptibility models. Nonlinear mixed effects models can reduce these biases and produce a more realistic national-scale landslide susceptibility model.
Inventory-based biases are inevitable when modelling landslide susceptibility on a national scale. The national-scale susceptibility model based on nonlinear mixed effects methods in which bias-describing variables are introduced as random effect items can reduce mixed effects of influencing factors caused by inventory-based biases. Using these methods, the geomorphically implausible susceptibility patterns resulting from neglecting inventory-based biases can be avoided, which is the modelled relationship between topography relief and landslide changes from a positive relationship to a negative relationship. In addition to that, our model can produce realistic susceptibility maps which are consistent with geomorphological process patterns. Compared with the baseline model, our model has a relative good performance. The cross-validation AUROC is near 0.8; the susceptibility map is consistent with geomorphological process patterns and has significant improvements in local details. The nonlinear mixed models used in this thesis can reduce inventory-based biases and can be applied to other regions.
3. The area of high susceptibility levels is projected to increase while the area of low levels is projected to decrease under climate change in China. The spatial pattern differs significantly in eastern and western China. The susceptibility levels of eastern and northern Qinghai-Tibet Plateau is projected to increase while decrease in southwestern part. The susceptibility levels of most regions in eastern part of the second and the third terraces are projected to increase while invariant in southeastern and southwestern parts.
Variations in the landslides susceptibility in China under climate change scenarios vary in different susceptibility levels. Compared with the landslide susceptibility in the historical period, the area of the very high landslide susceptibility level shows an increasing trend, while the very low and low landslide susceptibility level shows a decreasing trend, and the area of the medium and high susceptibility level shows fluctuations.
Variations in the landslides susceptibility in China are different in different emission scenarios and periods, with variations in the late 21st century (2066-2095) larger than the middle 21st century (2031-2060) and variations under RCP8.5 larger than RCP4.5. The result of multi-model median indicates that the area with very high susceptibility level is projected to increase by 53.61% while the area with very low levels is projected to decrease by 35.99%.
Variations in spatial pattern of landslide susceptibility in China vary in different regions. The landslide susceptibility levels in most parts of eastern China, the northern and eastern Qinghai-Tibet Plateau, and the eastern Tianshan Mountains have become higher; the levels in the southwestern and northwestern regions have remained unchanged; the levels in the southwestern and northwestern border areas of the Qinghai-Tibet have become lower.
4. The consensus of multiple GCMs is that occurrence of future landslides in highly susceptible areas in China is projected to increase. The change pattern differs significantly spatially; landslide occurrence in hilly coastal areas in the southeastern China is projected to increase slightly while significant increase in other regions. The monthly variations in landslides are bimodal, with the largest increase in spring and autumn.
The result of 21GCMs indicates that landslide occurrence is projected to increase under both future periods and both emission scenarios. The figures for the late 21st century (2066-2095) under the RCP4.5 and RCP8.5 scenarios are 19.9% and 33.2%, respectively. 
The climate change impact differs significantly in different regions. Regions with pronounced increases in landslides and high model consistency mainly include the Tianshan Mountains, the Kunlun Mountains and Qilian Mountains in NW region, the southeastern Qinghai-Tibet Plateau, the western Sichuan Plateau, the Yunnan-Guizhou Plateau and the Sichuan Basin in Southwest region and the Taihang Mountains. Landslide occurrence in hilly areas of the southeastern China tends to decrease or weakly increase and the consistency of most grids in this region is quite uncertain.
The climate change impact on future landslides has obvious seasonal differentiations. The monthly variations in landslides are bimodal, with the largest increase in spring and autumn and followed by summer while winter has no significant change. The presented method based on multiple GCMs and relative variation can be applied to assess the impact of climate change on landslides in other regions.
In summary, the research framework for investigating the climate change impact on landslide susceptibility on a national scale is presented based on the mechanism of geological disasters. 
The historical landslide inventory, landslide susceptibility influencing factors and multi-GCMs data systematically are coupled. Then, the high susceptibility areas of landslides in China are identified, the change in spatial pattern of landslide susceptibility and projecting the change in landslide occurrence in highly susceptible areas on a national scale are assessed, and the corresponding uncertainty analysis are presented. The landslide susceptibility levels in most parts of eastern China, the northern and eastern Qinghai-Tibet Plateau, and the eastern Tianshan Mountains have become higher; the levels in the southwestern and northwestern regions have remained unchanged; the levels in the southwestern and northwestern border areas of the Qinghai-Tibet have become lower. The main innovations of this thesis are: (1) A national scale landslide susceptibility model based on a nonlinear mixed effect method is established, which significantly reduces the impact caused by the inventory-based biases. Compared with the traditional model of directly including all influencing factors, nonlinear mixed effect method can improve the simulation of the relationship between landslide occurrence and landslide influencing factors, and the prediction of landslide susceptibility. (2) The change in spatial pattern of landslide susceptibility and projecting the change in landslide occurrence in highly susceptible areas with a spatial resolution of 0.25o on a national scale are assessed. The uncertainty analysis are presented. Change risk provides a decision basis for formulating scientific landslide disaster mitigation strategies and effectively reducing landslide disaster risk. (3) A long-term fatal landslide event inventory in China is compiled based on multiple data sources. The spatiotemporal pattern and influencing factors of fatal landslide events are analyzed by combining with historical rainfall data.