A prototype multi-hazard early warning system integrating flood and landslide forecasting: application in the upper Garonne River Basin

Multi-Hazard Early Warning Systems (MHEWS) are essential for disaster risk management, yet operational implementation remains limited by challenges in hazard integration, forecast uncertainty, and computational requirements. This study develops and evaluates a fast and lightweight prototype MHEWS integrating fluvial flooding and shallow landslide forecasting. The MHEWS couples snow-hydrological modeling (SNOW-17 and CREST) with a slope stability approach (infinite slope and Random Forest models) driven by bias-corrected Global Ensemble Forecast System (GEFS) meteorological forecasts. Quantile mapping bias correction substantially improves precipitation and temperature forecast performance, reducing variability errors and enhancing distributional agreement. Applied in the Upper Garonne River Basin (Spanish Pyrenees) using 24-hour lead time forecasts, multi-hazard analysis for 2010–2020 reveals asymmetric coupling between hazards: flood warnings respond rapidly to rainfall intensity, while landslide warnings depend on cumulative subsurface conditioning and antecedent soil moisture, with high-level warnings rarely coinciding. Evaluation of multiple forecast lead times (24, 48, 72 h) for June 2013 demonstrates systematic performance degradation with increasing lead time, reflecting amplified uncertainty propagation. This degradation is further compounded by the limitations of the quantile mapping approach; while it effectively aligns marginal statistics, it fails to preserve temporal autocorrelation or event persistence. This introduces residual biases in cumulative dynamics that disproportionately affect landslide prediction compared to flood estimation. The fully automated prototype, designed to be scalable and transferable, and built on open-source software and cloud infrastructure, demonstrates operational feasibility for real-time early warning applications across hydro-meteorologically complex environments.