태풍 경로 군집화와 행정구역별 손실예측모형: 기후변화 시나리오와 기계학습 응용

This study proposes a two-step machine learning approach to predicting economic damage caused by tropical cyclones (TCs) in the Korean Peninsula. We first analyze track clusters of TCs to classify regional physical hazards. We then predict regional TC losses by considering geographic and socio-economic information. Comparing traditional regression model(multiple linear regression), we show better performance of our machine learning-based model than the traditional linear model. We also determine that some economic and geographic factors (i.e. embankment, local income per capita, artificial water) are statistically significant in explaining TC losses. We further predict the economic loss per TC in the year 2100 under a climate change scenario of the Intergovernmental Panel on Climate Change(IPCC) and provide a TC risk map on a regional basis. We find in this prediction that the projected losses per TC can be 6.4 to 14.8 times larger than those at present. Our findings can provide (re)insurers with a stepwise process for predicting climate change-driven TC losses can help them for better financial management and proactive responses to climate changes.