Empirical flood depth damage functions based on insurance data from multiple flood events in Japan

Abstract

Empirical flood depth damage functions for detached residential buildings were developed using 340 insurance data, whose contracts were in 14 inundation areas under the three major flood events between 2018 and 2020 in Japan. The developed flood depth damage functions have two components: the probability of damage occurrence with a sigmoid function and the conditional damage ratio with a cumulative distribution function of the standard normal distribution. The developed functions accurately reproduced the observed aggregate losses. Two-fold cross-validation confirmed that the proposed method shows a 13% error to the observed loss in NRMSE, which is 3% larger than estimated by the Ministry of Land, Infrastructure, Transport and Tourism (MLIT). Focusing on the 0.45–1.0 m inundation depth, the MLIT function, which does not consider the probability of no damage, overestimates the aggregated damage loss, but our method considering the probability of no damage has good estimation accuracy. At shallow inundation depths with some probability of no damage, the accuracy of damage estimation may be improved by considering the probability of damage. The proposed method is particularly suitable for estimating damage at shallow inundation depths because it considers the probability of no damage and is extensible to existing damage functions.