Quantitative Assessment of Population Risk to Tropical Cyclones Using Hybrid Modeling Combining GAM and XGBoost: A Case Study of Hainan Province

Chenna Meng,Wei Xu,Peng Su,Lianjie Qin,Xinli Liao,Junlin Zhang
DOI: https://doi.org/10.1016/j.ijdrr.2024.104650
IF: 4.842
2024-01-01
International Journal of Disaster Risk Reduction
Abstract:Global climate change is expected to increase the proportion of intense tropical cyclones (TCs) in the Northwest Pacific. This study focused on how factors, especially extreme events, may affect disaster losses. To address this issue, an event-based multivariate TC risk assessment model, which employs copulas, a generalized additive model, and undersampling extreme gradient boosting decision tree techniques, is developed to enhance the accuracy of disaster loss prediction. The results suggest that on Hainan Island, the rate of the affected population is positively correlated with the maximum wind speed and maximum daily rainfall while negatively correlated with the gross domestic product and elevation. The study also shows that the TC risk in the cities of Hainan Island increases as the return period expands, and each return period scenario shows a unique geospatial distribution of TC risk on Hainan Island, with higher risks in the coastal and eastern regions. These results highlight the importance of implementing effective disaster management strategies to mitigate the impact of severe TCs in the region.
What problem does this paper attempt to address?