Spatial Dependence in Extreme-Rainfall Risk and Its Effect on Portfolio-Level Flood Loss
Year: 2026
Type: report
Abstract: Flood risk models used by insurers and banks conventionally treat losses at different locations as statistically independent, an assumption known to distort both the probability of system-wide failure and the tail of aggregate loss distributions in delta regions such as the Netherlands. This note presents a small, fully reproducible simulation study that quantifies one specific consequence of this assumption for financial risk management: the effect of spatial dependence in extreme-rainfall footprints on the tail of an insurer’s aggregate annual loss distribution. A Gaussian-copula field with generalized Pareto (GPD) margins is used to generate synthetic but extreme-value-consistent rainfall fields over a stylized delta-region exposure grid; a synthetic vulnerability function converts rainfall depth into portfolio loss. Across correlation ranges of 0–60 km, the mean annual loss is essentially unchanged (as expected under the same marginal distributions), but the 99.5% Value-at-Risk almost doubles relative to the mean once realistic spatial correlation is introduced, and the ratio of 99.5% VaR to expected loss rises from 1.10 to 1.99. A second experiment shows that this dependence effect compounds with, rather than substitutes for, climate-driven intensification of extreme precipitation. A small empirical check against real KNMI station data for June 2020 estimates a spatial correlation range of roughly 106 km, larger than the illustrative range studied, suggesting the reported dependence effect is, if anything, conservative. The paper closes by relating both results to two specific research pillars of the FIRM programme, namely the BREACH-type modelling of dependent flood-defence failures and the construction of a probability distribution for rainfall-event size from radar data, and by outlining how the synthetic pipeline here would be replaced with real KNMI radar composites, dike-ring geometries and the BREACH+ framework in a PhD project. All code is provided.
Author: Ameya, Damle
Institution: University of Reading
Language: English
Subfield: Global and Planetary Change
Field: Environmental Science
Domain: Physical Sciences
SDG: Climate action
Open Access status: green
OpenAlex ID: W7208832104
Cites:
Cited by:
Related to: