An Empirical Flood Fragility and Vulnerability Modelling Study using Historical Insurance Claims in Jakarta

ABSTRACT Flood vulnerability assessment in rapidly urbanizing cities requires locally calibrated relationships between flood intensity, building occupancy, and observed losses. This study develops an empirical flood fragility and vulnerability modelling framework for Jakarta using 8,721 historical insurance claim records. A Random Forest (RF) classifier was first employed to evaluate the predictive information contained in flood depth and occupancy, followed by ordinal logistic regression to derive statistically consistent fragility curves for residential, commercial, industrial, and warehouse occupancies. The RF model achieved testing accuracy and weighted F1-score of 0.831 and 0.762, respectively, with flood depth identified as the most influential predictor. The resulting fragility curves show monotonically increasing exceedance probabilities with increasing flood depth, while differences among occupancy classes indicate distinct vulnerability responses. The corresponding vulnerability curves generally exhibit lower loss ratios than existing international relationships. At 300 cm flood depth, for example, the proposed curves reach approximately 0.32 for commercial (reinforced concrete) and 0.44 for residential (masonry) buildings, compared with approximately 0.88 and 0.86, respectively for the corresponding vulnerability relationships reported by the Joint Research Centre (JRC) study. For industrial (steel) buildings, the proposed curve reaches approximately 0.13 at 200 cm, compared with approximately 0.76 for JRC. These differences highlight the importance of locally calibrated occupancy-based vulnerability functions rather than direct transfer of international curves. The resulting framework is implemented in a web-based application, providing a practical basis for empirical flood loss estimation and supporting flood risk analysis, insurance applications, as well as disaster risk management.