近年来，全球极端灾害（例如：台风、地震等）逐渐增多，我国处于太平洋的西北侧，是全球遭受热带气旋灾害影响最严重的国家和地区之一，以2018年的登陆我国的台风“山竹”为例，造成了重大的经济损失，严重的影响了我国东南沿海地区的人民生命财产安全和社会稳定。因此通过科学有效的方法进行台风风险评估、提高台风灾害防御能力有重要意义。 本研究选择我国东南沿海地区的福建省、浙江省两个省份作为研究区，基于西北太平洋热带气旋数据集分析了登陆福建省和浙江省台风的时空分布特征，统计研究区在台风期的平均风速和降水特征，以及路过研究区且在研究区境内因台风而产生的风速与降水规律，运用多种指标、从不同角度分析研究区历史降水和风速与台风造成损失的空间相关关系，使用经验正交函数分析方法（EOF, Empirical Orthogonal Function）、多元经验正交函数分析方法（MVEOF, MultivariateEmpirical Orthogonal Function）探索研究区降水和风速的空间模态分布和主成分变化特征，根据主成分和直接经济损失的相关性确定影响研究区台风灾害损失的主要致灾因子。 根据1949-2016年西北太平洋产生的热带气旋历史纪录，运用蒙特卡罗仿真、时间序列向量自回归模型（VAR, Vector Autoregressive Model）两种方法，实现对未来台风随机路径进行模拟。在蒙特卡罗仿真方法中，随机模拟台风的方位角和移动距离实现台风随机路径的生成；而VAR方法是将历史台风路径的位置（经纬度）、风速强度和降水统一到四维时间序列向量上，通过估计时间序列模型的参数，模拟产生新的台风时间序列。另外，本研究根据登陆点位置分布和强度大小对比了两种台风路径模拟方法的结果，VAR模型因考虑到了经纬度、风速强度和降水之间的自相关性更加符合历史路径。 利用全球海温数据的分析，筛选出对研究区台风登陆频次具有显著相关关系的特征海域以及变量，建立逐步回归模型和支持向量机模型，实现在台风期来临前对登陆研究区的台风频次进行年度预测。最后，利用台风模拟数据集评估主要致灾因子强度，通过构建的主要致灾因子与损失间的定量关系对台风致损进行年度预测。 本研究得出如下结论：1）福建省的主要致灾因子是路过福建且在福建境内最大风速，浙江省台的主要致灾因子是路过浙江且在浙江境内总降水、总风速；2）在构建台风频次预测模型时，机器学习方法利用前期海温等气象数据可实现对年度台风登陆研究区 的频次进行预测更加准确，这对开展年度台风风险评估具有积极意义；3）运用VAR时间序列模型建立台风路径随机模拟，不但考虑了地理相似性，还考虑了台风路径前后时间点特征的自相关性，因而其模拟结果更接近于历史台风路径，更能反映出登陆我国沿海地区台风的宏观特征；4）通过构建全球海温信号与区域致灾因子定量关系，结合台风频次预测和随机事件仿真，可实现区域台风年度风险预测。

These days, extreme disasters are occurring repeatedly all over the world. China is located in the western coast of the Pacific Ocean, and is one of the countries that are influenced by natural disasters seriously. The number of typhoons in China are more than 20 even up to 40 in typical years, which cause serious death and loss. In 2018, Super Typhoon Mangkhut influenced nearly 3 million people and caused 5.2 billion loss. Therefore, efficient and scientific methods should be used to assess typhoons risk, and it is of vital important to improve the ability of disasters prevention and risk reduction. In this research, I choose Fujian and Zhejiang as research fields. Based on tropical cyclone database of the Pacific Ocean, I analyse the temporal and spatial characteristics of typhoons landed on Fujian and Zhejiang. And then I get the average wind speed and rainfall of typhoon season, maximum and sum of wind speed and rainfall when typhoons pass away these two provinces. From different indexes and angles, I try to find the spatial correlation between rainfall or wind speed with loss caused by typhoons. EOF and MVEOF methods are used to explore changes of wind speed and rainfall over time to decide the main hazards quangtitatively. Next, Monte Carlo and VAR time series methods are used to simulate typhoon paths based on cyclones of Northwest Pacific occurred from 1949 to 2016. By Monte Carlo methods, I generate distances and azimuths between two steps. VAR methods look longitude, latitude, wind speed and rainfall as time series vectors and estimate parameters of VAR methods. And then I use historical VAR models to generate new time series. In addition, I compare the results of these two methods by the distribution and intensity of landing points, and find that VAR method is better for considering the autocorrelation of typhoon paths. Global sea surface temperature are used to find region or variables, which have significant correlation with frequency of typhoons. By these significant variables, I build stepwise regression and SVM models to predict frequency of typhoons passing away the research fields. Finally, I assess the intensity of main hazards by simulated typhoon paths, and predict the loss by the quantified relationship between main hazards and loss. In this research, 1) Main hazard in Fujian is max wind speed when typhoon passing away Fujian, and main hazard in Zhejiang is whole wind speed and rainfall when typhoon passing away Zhejiang. 2) When I build prediction model of typhoon frequency, I find that the result of machine learning is more accurate than stepwise regression. It is meaningfull to annual typhoon risk assessment. 3) VAR time series consider not only the geographic similarity but also the autocorrelation of the time series before and after the typhoon track, which reflect the macroscopic characteristics of typhoons landing in coastal areas of China. 4) I build quantitive model of global sea surface temperatureand main hazard. I combine the quantitive model with simulation data to predict annual typhoon risk in the typical region.