在全球变暖的背景下，台风灾害对我国沿海地区的影响越发严重，加之城市化带来的人口、财产高度密集，使我国成为世界上受到台风影响最频繁且损失最严重的国家之一。因此，台风灾害风险评估作为灾害预测、防治减缓、提高社会适应性和恢复力以及灾害补偿机制研究的重要基础，日益得到国内外各领域学者的广泛关注。但是，目前尚没有形成一套标准的台风灾害风险评估方法体系，并且台风灾害风险是自然环境因素与社会经济因素相互影响的综合产物，其影响因素及风险本身的评估结果都仍存在较大的不确定性。

因此，本文首先基于广义线性模型理论，利用中国2000?2015年的历史台风灾害事件构建台风灾害直接经济损失评估模型；然后，基于模型的“弹性系数”，评估在全球变化背景下，21世纪后期（2081?2100年）我国的台风灾害风险。研究得到的主要结论如下：

（1）基于台风最大风速（W）、资产存量暴露度（K）、人均GDP（I）和非钢混结构建筑比例（B）构建的WKIB模型为最优台风灾害直接经济损失评估模型。本文在模拟台风风场和雨场的基础上，重建了我国2000?2015年的台风灾害影响范围，选取台风最大风速、中心最低气压（P）为致灾因子危险性的候选解释变量，资产存量为暴露度的候选解释变量，人均GDP和非钢混结构建筑比例为脆弱性的候选解释变量。通过共线性检验、相关性分析和相对权重等进行解释变量优选，构建了WKIB、PKIB、WKI、WKB、PKI和PKB六个不同解释变量组合的直接经济损失评估模型；并基于AIC准则、R²、误差比、方差分析等对比了六个直接经济损失评估模型。对比结果显示，WKIB模型的AIC值（5287.1）最小，（0.65）最大，说明该模型的拟合优度最高；WKIB模型的误差比（3.45）最小，说明该模型对直接经济损失的预估效果最好。

（2）考虑到全球变暖导致的台风强度变化，以及社会经济发展背景下资产存量的增加，21世纪后期我国台风灾害风险将大大增加：

①对于致灾因子强度，WKIB模型的最大风速弹性系数为1.43% [0.69%, 2.20%]，表示当损失的其他影响因素固定不变的情况下，台风最大风速每增加1%，将导致直接经济损失风险增加1.43% [0.69%, 2.20%]。本文利用CNRM、ECHAM、GFDL和MIROC四个全球气候模式，预估在SRES A1B情景下21世纪后期影响我国的台风强度和频次的变化。21世纪后期，CNRM、ECHAM、GFDL和MIROC模式预估的台风平均最大风速相比于基准期（1950?2015年）分别增加了11.27%、1.00%、5.80%和15.36%，这将导致我国的台风灾害风险分别增加16.12% [7.78%, 24.79%]、1.42% [0.69%, 2.19%]、8.30% [4.00%, 12.77%]和21.96% [10.60%, 33.78%]。

②WKIB模型的资产存量弹性系数为0.97% [0.78%, 1.16%]，表示当其他解释变量不变时，暴露于台风灾害影响范围内的资产存量价值每增加1%，会导致直接经济损失风险增加0.97% [0.78%, 1.16%]。本文基于共享社会经济路径（SSP1、SSP2和SSP3）预测了2020?2100年我国台风灾害影响范围内资产存量价值的时空发展趋势。21世纪后期，在SSP1、SSP2和SSP3情景下，我国台风灾害影响范围内的平均资产存量价值相比于基准期（1990?2010年）的增长率分别为18.9倍、16.6倍和10.3倍，这将导致我国台风灾害风险分别增加1834.3% [1475.0%, 2193.6%]、1612.5% [1296.7%, 1928.4%]和996.8% [801.6%, 1192.1%]。

③人均GDP弹性系数为1.15% [0.66%, 1.62%]，即当其他解释变量不变时，人均GDP每增加1%，直接经济损失风险将减少1.15% [0.66%, 1.62%]。非钢混结构建筑比例弹性系数为2.45% [0.21%, 4.61%]，即当其他影响因素不变时，非钢混结构建筑比例每减少1%，直接经济损失风险将降低2.45% [0.21%, 4.61%]。

总体而言，本文收集了我国历史台风灾情数据，建立了历史台风灾害信息库；综合考虑台风致灾因子强风和暴雨的影响重建历史台风灾害影响范围，并在此基础上分析量化台风风险影响要素数据，提高了台风灾害直接经济损失评估模型的模拟精度。同时，本文通过对台风风险的影响要素优选和模型优选，对比分析得到了适用于中国的台风灾害直接经济损失评估模型，结合情景分析结果，更能科学、定量刻画未来全球变化背景下灾害风险要素变化对我国台风灾害风险的影响。本研究对于从风险管理视角认识台风灾害与经济发展、自然环境变化之间的关系具有重要意义，对于管理者理解人为因素在减轻台风灾害风险中的重要作用，以及减灾策略的制定具有重要实践指导价值。

Under the background of global warming, the impact of tropical cyclones (TCs) on the coastal areas of China is becoming increasingly serious, and the highly intensive population and property caused by urbanization makes China one of the countries in the world most frequently affected by TCs and suffered the most serious losses. Therefore, TC disaster risk assessment, as an important basis for disaster prediction, prevention and mitigation, improving social adaptability and resilience, and disaster compensation mechanism research, has been increasingly widely concerned by scholars in various fields. However, there is no standard method system of TC risk assessment, and TC risk is a combination of natural factors and socio-economic factors, the results of risk drivers and itself still contain large uncertainties.

Therefore, this paper firstly developed the direct economic loss (DEL) assessment model of TC disaster based on the historical TC events in China from 2000 to 2015, using the generalized linear model theory. Then, this study estimated the TC disaster risk in China in the late 21st century (2081?2100) under the background of global changes using the elasticity coefficient of the model. The main results of this paper are as follows:

(1) The WKIB model based on TC maximum wind speed (W), asset value exposure (K), per capita GDP (I), and proportion of non-steel-concrete residential buildings (B) is the optimal direct economic loss assessment model. This paper reconstructed the TC-affected areas in China from 2000 to 2015 based on the simulation of TC wind field and rain field, and chose the maximum wind speed and minimum pressure (P) as candidate explanatory variables of hazard, asset value as candidate explanatory variables of exposure, and GDP per capita and proportion of non-steel-concrete residential buildings as candidate explanatory variables of vulnerability. Through collinearity test, correlation analysis and relative weight, etc., the optimization of explanatory variables was carried out. This paper constructed six direct economic loss assessment models WKIB, PKIB, WKI, WKB, PKI and PKB with different explanatory variable combinations, and provided comparative analysis of these models based on AIC criteria, R², error ratio, anova analysis and other methods. The results show that the AIC value (5287.1) of the WKIB model is the smallest and the R² (0.65) is the largest, which indicates that this model has the best fitness; while the error ratio (3.45) of the WKIB model is the least, which means that the model has the best estimated effect on TC direct economic loss.

(2) Considering the changes in TC intensity caused by global warming and the increase in asset value in the context of socio-economic development, the TC risk in China will greatly increase in the late 21st century.

① For the hazard, the maximum wind speed elasticity is 1.43% (0.69%, 2.20%) in the WKIB model, indicating that a 1% increase in maximum wind speed of a TC will lead to a 1.43% (0.69%, 2.20%) increase in the potential DEL risk, holding the other factors constant. This paper used four global climate models (GCMs), CNRM, ECHAM, GFDL and MIROC, to estimate the changes in the intensity and frequency of TCs affecting China under the SRES A1B scenario from 2081?2100. In the late 21st century, the average maximum wind speed of TCs projected by CNRM, ECHAM, GFDL and MIROC will increase by 11.27%, 1.00%, 5.80% and 15.36%, respectively, compared to the baseline period (1950-2015). This will lead to an increase in TC risk of China by 16.12% [7.78%, 24.79%], 1.42% [0.69%, 2.19%], 8.30% [4.00%, 12.77%] and 21.96% [10.60%, 33.78%], respectively.

② The asset value elasticity is 0.97% (0.78%, 1.16%), indicating that a 1% increase in asset value exposure will lead to a 0.97% (0.78%, 1.16%) increase in the potential direct economic loss, holding the other factors constant. This paper estimated the spatial and temporal development trend of asset value within the TC-affected areas in China under Shared Socioeconomic Paths (SSP1, SSP2 and SSP3) from 2020 to 2100. In the late 21st century (2081?2100), under the SSP1, SSP2 and SSP3 scenario, the average asset value in the TC-affected areas will increase by 18.9 times, 16.6 times and 10.3 times, respectively, compared to the baseline period (1990?2010). This will lead to an increase in TC risk of China by 1834.3% [1475.0%, 2193.6%], 1612.5% [1296.7 %, 1928.4%] and 996.8% [801.6%, 1192.1%].

③ The GDP per capita elasticity is 1.15% [0.66%, 1.62%], indicating that a 1% increase in GDP per capita will reduce the direct economic loss by 1.15% [0.66%, 1.62%], holding the other factors constant., direct economic losses will be reduced by. The proportion of non-steel-concrete residential buildings elasticity is 2.45% (0.21%, 4.61%), indicating that a 1% decrease in the proportion of non-steel-concrete residential buildings, the direct economic loss will decrease by 2.45% (0.21%, 4.61%).

Overall, this paper collected the historical TC DELs and established a TC disaster database; and analyzed and quantified the data of TC risk drivers, based on comprehensively considering the impact of TC hazards—strong winds and rainstorms—to reconstruct the historical TC disaster impact range. These contributions have improved the simulation accuracy of the TC disaster direct economic loss assessment model. Meanwhile, this paper developed a TC disaster direct economic loss assessment model, which is suitable for China, through the comparative analysis based on variables and model optimization. More scientifically and quantitatively depict the impact of changes in TC hazards, exposure and vulnerability to TC risk in China in the context of global change. This paper is of great significance to understand the relationship between TC disaster and economic development and natural environment change from the perspective of risk management, and it is of great practical guidance for managers to understand the important role of man-made factors in reducing TC disaster risk, and to the formulation of TC disaster reduction strategy.