Using river flood event drivers for model-intercomparison – a process-based analysis of global water models

Global water models are valuable tools for predicting river flood hazard in data‑scarce regions and under future climate scenarios. Their ability to produce spatially coherent projections means that their results can be used broadly for global or large‑scale scientific analysis and policy planning. However, the complexity of these models, together with the large volume of data they generate, creates challenges for evaluating how well they represent key processes. At long‑term climatic timescales, global water models show marked differences in the controlling processes for water‑balance components, as previous studies revealed. At event timescales, accurate representation of hydro‑meteorological dynamics requires consideration of multiple flood drivers, such as precipitation, soil moisture, or snowmelt.

In this analysis we compare simulations from six global water models (CWatM, H08, LPJmL, JULES‑W2, MIROC, and WaterGAP2) that were run within the international model‑comparison framework ISIMIP3a. We evaluate event drivers at the spatial scale of individual model cells (0.5°). We define high‑flow events as annual runoff maxima that exceed the 2‑year flood threshold. We classify potential drivers (short extreme rainfall, long extreme rainfall, soil moisture and snowmelt proxies). Drivers can be identified either individually or in combination with others (e.g., snowmelt + rainfall, soil moisture + rainfall, etc.).

We find that, in some models, extreme rainfall (short rain or short + long rain) often dominates high‑flow events, while other models show more influence from combinations of drivers such as snow or soil moisture. Except in snow‑dominated regions, all models share one feature: short extreme rainfall, either alone or combined with other factors, is part of the dominant flood driver almost everywhere. This has potentially significant consequences for future estimates of flood frequency under changing conditions. Still, the importance of antecedent soil moisture in flood generation remains ambiguous among the models, which contrasts with current process understanding and observation‑based analyses. This and other results demonstrate that process‑based model intercomparison provides valuable guidance for model development.