基于机器学习方法的浙江省台风灾害风险评估和动态风险预报

Typhoon disaster risk assessment and dynamic risk forecasts in Zhejiang Province based on machine learning methods

利用机器学习方法建立以区县为基本研究单元的浙江省台风灾害风险评估模型,并进一步结合气象预报与实测数据形成覆盖全省、时空连续的台风过程动态风险预报,为科学应急减灾提供决策支持。首先,本研究以浙江省各区县为研究对象,考虑危险性、孕灾环境、暴露性和脆弱性等风险要素选择台风灾害风险评估模型的预测变量;其次,基于10个重大历史灾害的灾损数据(直接经济损失)划分风险等级作为输出变量;最后,采用机器学习模型XGBoost建立台风灾害风险评估模型。同时,以利奇马为例,进一步探索该模型的实战应用潜力,即以气象预报和实测数据为模型驱动,实现浙江省全域各区县台风灾害风险的实时更新预报。

A typhoon disaster risk assessment model is developed based on a machine learning method for Zhejiang Province with counties as the basic research unit.This model is further combined with the real-time meteorological observation and forecast data to generate spatiotemporal risk forecasts across the whole province continuously and dynamically,which can further facilitate decision makers to decide more scientific disaster prevention and mitigation policies.First,this model chooses a set of predictor variables as input by considering risk elements such as hazard,inducing environment,exposure and vulnerability.Secondly,the model takes the risk levels of counties as outputs classified based on the disaster damage(more specifically,direct economic loss)in 10 major historical typhoon events.At last,the model is established using the machine learning model XGBoost.Taking typhoon Lekima as an example,this study also explores the potential of the proposed model to real-time dynamic risk forecasting during the evolution process of a typhoon disaster for all counties in Zhejiang Province.