Can macro- or meso-scale coping capacity variables improve the classification of building flood losses?

This study proposes a novel approach to improve the classification of severe building losses caused by river floods (i.e., identification of buildings with high flood damages). In addition to traditional variables reflecting flood hazard and building vulnerability, we investigate the impact of coping capacity variables (i.e., variables accounting for the preparedness and disaster response of the population and management authorities). These coping capacity variables are evaluated at three different scales: the building level (micro-scale), the census tract level (meso-scale), and the municipality level (macro-scale). Specifically, at the macro- and meso-scale these include: (i) the surprise effect (the ratio of the number of flooded buildings to the number of flooded buildings located in an official flood hazard area), (ii) the overwhelming effect (the fraction of flooded buildings compared to the total number of buildings within each census tract or municipalities), and (iii) flood rarity (the ratio of the peak discharge of the considered event to the 100-year flood peak). A binomial logistic regression model is used to classify flood losses based on field survey data from the extreme 2021 flood in eastern Belgium. Each variable is assessed for statistical significance, physical relevance, and multicollinearity. The results show that macro- and meso-scale coping capacity variables are insignificant in classifying building losses using the current dataset, suggesting that data on the building level are needed to reliably estimate building losses. Instead, the variables that contribute most to the classification are water depth, building footprint area, building finishing level and the heating system location. The performance of the classifier, measured by the AUC value, achieves an accuracy of 83%.