Assessment of economic losses from tropical cyclone disasters based on PCA-BP

Abstract

Tropical cyclones are the most common natural disasters in coastal regions and are the most costly in terms of economic losses. Economic loss assessment is the basis for disaster prevention and alleviation and for insurance indemnification. We use data from 1970 to 2008 for Zhejiang Province, China, in this study evaluating economic losses. We convert direct economic losses from tropical cyclone disasters in Zhejiang Province into indices of direct economic losses. To establish our assessment model, we process disaster-inducing assessment factors, disaster-formative environments and disaster-affected bodies using the principal component analysis method, and we abstract the principal component as the input of a BP neural network model. We found in the actual assessments of five tropical cyclones affecting Zhejiang Province in 2007 and 2008 that the post-disaster loss assessment values of tropical cyclones were higher than the actual losses, but that for more severe storms, the gap was smaller. This reflects the beneficial effect of efforts toward disaster prevention and alleviation for severe tropical cyclones. Pre-assessments based on relatively accurate forecast values of wind and precipitation at the start of a tropical cyclone have been in accordance with the post-disaster assessment values, while the pre-assessment results using less accurate forecast values have been unsatisfactory. Therefore, this model can be applied in the actual assessment of direct economic loss from tropical cyclone damage, but increasingly accurate forecasting of wind and precipitation remains crucial to improving the accuracy of pre-assessments.