Multi-source geospatial modelling of direct economic losses from rainstorm-related disaster events in Guangxi, South China

Rainstorm-related disasters cause substantial economic losses in Guangxi, China, but event-level loss estimates are complicated by spatially clustered events, changing socioeconomic exposure, and heterogeneous reporting. We compiled 1,811 county-event records from the official disaster archive for 2012–2021 and linked them to event-period precipitation, year-matched gridded population and GDP, and selected vulnerability-related proxies. We fitted log-link generalised linear models and evaluated candidate specifications using storm-grouped, county-held-out, and temporal validation. In the selected IPB specification, higher maximum daily precipitation and exposed population were associated with higher expected direct economic loss conditional on an event being recorded. The estimated elasticities were 2.30 (95% CI 1.55–3.04) for precipitation intensity and 1.21 (95% CI 0.48–1.94) for exposed population. Influence analyses indicated that the precipitation association was retained after excluding upper-tail losses, whereas the population association weakened and its interval crossed zero. A parameterised sensitivity analysis characterises how fitted losses vary under specified changes in precipitation intensity and exposed population. The results provide an interpretable event-level framework for examining recorded rainstorm losses in Guangxi and for conducting transparent, model-based comparisons under alternative hazard and exposure conditions.