A high-resolution global flood hazard model

TLDR

Floods are a global natural hazard, yet most hazard research has been conducted by wealthy nations, creating a data gap that growing demand in developing regions is trying to fill. The study identifies six key challenges to global flood hazard modeling and proposes a framework methodology that leverages cross‑disciplinary advances to address them. Using this framework, the authors generate ∼90 m resolution return‑period flood hazard maps for the terrestrial land surface between 56° S and 60° N, validated against high‑resolution UK and Canada datasets. The model captures 66–75 % of benchmark risk areas with few false positives, achieves ≈5 % mean absolute error when aggregated to 1 km, and demonstrates that incorporating an automatically parameterized subgrid channel network is essential for performance, with next‑generation terrain data promising further gains.

Abstract

Floods are a natural hazard that affect communities worldwide, but to date the vast majority of flood hazard research and mapping has been undertaken by wealthy developed nations. As populations and economies have grown across the developing world, so too has demand from governments, businesses, and NGOs for modeled flood hazard data in these data-scarce regions. We identify six key challenges faced when developing a flood hazard model that can be applied globally and present a framework methodology that leverages recent cross-disciplinary advances to tackle each challenge. The model produces return period flood hazard maps at ∼90 m resolution for the whole terrestrial land surface between 56°S and 60°N, and results are validated against high-resolution government flood hazard data sets from the UK and Canada. The global model is shown to capture between two thirds and three quarters of the area determined to be at risk in the benchmark data without generating excessive false positive predictions. When aggregated to ∼1 km, mean absolute error in flooded fraction falls to ∼5%. The full complexity global model contains an automatically parameterized subgrid channel network, and comparison to both a simplified 2-D only variant and an independently developed pan-European model shows the explicit inclusion of channels to be a critical contributor to improved model performance. While careful processing of existing global terrain data sets enables reasonable model performance in urban areas, adoption of forthcoming next-generation global terrain data sets will offer the best prospect for a step-change improvement in model performance.