The infiltration capacity of soils is one of the most important parameters of the water cycle as it directly influences the amount of rainfall runoff on the surface. Soil moisture precondition is critical to the potential and severity of flooding, especially in larger catchment areas where the infiltration potential is high. In modelling this component is often accounted for through empirical numbers and different scenarios as well as the worst-case scenario where the soil is considered a sealed surface. To account for this parameter in a flood prediction model, we use real-time soil moisture data from sensor probes of the most relevant soil types in the respective catchments. The model itself is data driven and based on multiple water level gauges, rain gauges and radar rainfall data and is trained on the relation between soil moisture, rain and water level data to predict gauge levels up to eight hours ahead of an event. The model type allows for a continuous flood monitoring and prediction to run with high temporal resolution and thereby improves risk assessment and reaction opportunities to potential flood events. A fully deployed system in Kreuzberg (Ahr) shows promising results for reliability and scalability.
