Space-time downscaling of extreme rainfall using stochastic simulations, intense runoff susceptibility modeling and remote sensing-based pluvial flood mapping

Accurate rainfall modeling is crucial to understand the way water is intercepted, infiltrates and flows through surfaces and rivers. In particular, it is paramount for the study of the influence of rainfall spatio-temporal distribution on basin hydrologic response and the structure of floods. Current weather radar products allow capturing the variability of rainfall extremes mainly at 1 km spatial resolution. In France, radar measurements are performed at a 5-minute time step, while gauge-based reanalysis are computed at hourly resolutions. During short-duration high-intensity precipitations, pluvial floods (PF, or flash floods) can occur outside the hydrographic network in runoff-prone areas, leading to various types of damages such as soil erosion, mud and debris flows, landslides, vegetation uprooting or sediment load deposits. Contrary to fluvial floods, PF are highly correlated to local rainfall. Depending on generic susceptibility linked to topography, soil texture and land use, specific precipitation patterns can trigger intense overland flow. Hence, after extreme weather events, precise reports on PF locations provide key information for rainfall reanalysis and downscaling at fine spatial resolution.

This work focuses on two extreme Mediterranean events (more than 300 mm of rainfall in 24 hours) that took place in the South of France between 2018 and 2020. Time series of hourly rainfall intensities from Comephore radar reanalysis data at 1 km resolution (Météo-France) are confronted to (i) maps of PF that occurred during the events and (ii) generic susceptibility maps to surface runoff. For (i), runoff-related impact maps of the events are produced using the remote sensing-based FuSVIPR algorithm (Cerbelaud et al., 2023) based on Sentinel-2 temporal change images and Pléiades satellite or airborne post event acquisitions. For (ii), the IRIP© method (Dehotin and Breil, 2011; Cerbelaud et al., 2022) is used to generate PF susceptibility maps. The model is run with the RGE Alti® 5 m DEM, the OSO French land cover dataset, and soil type susceptibility characteristics derived from both climatological information and the ESDAC database.

We primarily show that areas with higher IRIP levels are more likely to be impacted by PFs, and even more so where short-term precipitation was heavier. Additionally, rainfall intensities are negatively correlated with IRIP susceptibility scores in PF impacted areas. This corroborates that somewhat higher rainfall intensities are required for flash floods to occur in less susceptible areas. Similarly, comparatively smaller rainfall amounts can trigger PFs in locations where susceptibility is high. Then, the Comephore products are downscaled at 50 m resolution on both events using the SAMPO stochastic simulator (Leblois and Creutin, 2013). Among multiple scenarios, optimal ones are chosen based on the assumption that the negative correlation with the IRIP susceptibility levels in the affected areas should be equally or even more present in the downscaled rainfall time series. This study hence suggests an original way of selecting disaggregated extreme rainfall scenarios that are consistent with the observed consequences of intense runoff on the land surface using various tools such as a stochastic simulator, a hydrological risk mapping method and earth observation data.