Linking in-situ and simulated soil moisture data for flood prediction: the advantage of joint probabilities of initial soil moisture and rainfall characteristic

Local heavy precipitation regularly causes great damage resulting from flash floods in smaller catchments. Appropriate discharge records are usually unavailable to derive an extreme value statistics and regionalization approaches predicting peak discharge from discharge records of larger basins cannot include the small-scale effects and local processes. In addition, forecasting flash floods from rainfall forecast requires to identify the initial soil moisture conditions under which a catchment is most prone to trigger flash floods. In this respect, soil moisture affects runoff at the local scale during runoff generation (infiltration), but also at the catchment scale during runoff concentration with possible infiltration of overland flow (run-on infiltration) along the flow path.

Our proposed framework to study the role of soil moisture on flash floods includes three steps: (1) to validate long-term hydrological simulations with in-situ soil moisture data to derive typical probability distributions of initial soil moisture depending on soil properties, vegetation and land cover, groundwater influences, etc; (2) to derive the sensitivity of runoff generation to soil moisture at the local and catchment scale by combining different probabilities for rainfall amount, duration and initial soil moisture resulting in the same joint probability and (3) to include the effect of soil moisture on run-on infiltration by linking a distributed hydrological and 2D-hydraulic model to simulate runoff hydrographs with and without run-on infiltration. The final set of simulations with the distributed, process-based rainfall-runoff model RoGeR for different temporal (event to long-term) and spatial scale (plots to submeter scale) allows us for a given catchment to derive the role of soil moisture on different hydrological processes (runoff generation and runoff concentration). We developed a spatial explicit method, which combines the joint probability of soil moisture and rainfall for runoff formation with hydraulic assumptions to determine runoff concentration and thus the corresponding hydrographs and the specific conditions in a catchment that can trigger flash floods. These simulations are compared in different test catchments with discharge records to validate out model chain. Finally, a comparison among different catchments with different characteristics (soil, geology, land-use, geomorphology, etc) enables us to derive a flood generation soil moisture sensitivity which should help to improve hydrological models to include all relevant processes and to focus our future in-situ soil moisture observations in the sensitive catchments to allow for a better prediction of flash floods by including observed soil moisture instead of simulated values.