The evaluation of erosion risk in dry areas is challenging because erosion is often an outcome of individual rainstorms and is highly dependent on rainfall spatiotemporal patterns and on local land-use and topography. This study integrates a hybrid erosion model with rainfall data from high-resolution weather radar to simulate soil erosion during 22 high-intensity flash-flood generating rainstorms in a Mediterranean watershed (69 km2). We examine erosion over individual hillslopes and their spatial average over the watershed, representing intra-watershed and watershed-scale erosion, respectively. Our objectives are to: (a) determine how intra-watershed erosion corresponds to various physiographic factors (rainfall, land-use, topography); (b) determine which of these factors contributes to intra-watershed erosion the most; (c) quantify the effect of temporal variations in rainfall intensities on storm-scale erosion in relation to land-use type. We use for the first time a hybrid erosion model (K2-RHEM-DWEPP) based on the watershed-scale KINEROS2 model, that integrates the hillslope-scale Dynamic WEPP (DWEPP) and RHEM models, which were individually developed to represent erosion processes in croplands and rangelands, respectively. Watershed-scale storm erosion is best correlated with spatially-averaged 10-minutes maximum intensities (R2 = 0.58), and the correlation decreases for longer durations (R2 ≤ 0.54). When the spatially-averaged 10-minutes maximum intensity is multiplied by the area that contributes sediment, a better correlation with watershed-scale erosion is observed (R2 = 0.75). Hillslope erosion rates are higher when both rainfall intensities and topographic slopes are high, while land-use has a second-order effect. Higher storms maximal intensities result in higher hillslope erosion rates, especially over croplands. Our conclusions are useful to target locations for conservation practices and to better understand the effects of climate change on soil erosion.

Soil erosion affects agricultural landscapes worldwide, threatening food security and ecosystem viability. In arable environments, soil loss is primarily caused by short, intense rainstorms, typically characterized by high spatiotemporal variability. The complexity of erosive events challenges modeling efforts and explicit inclusion of extreme events in long-term risk assessment is missing. This study is intended to bridge this gap by quantifying the discrete and cumulative impacts of rainstorms on plot-scale soil erosion and providing storm-scale erosion risk analyses for a cropland region in northern Israel. Central to our analyses is the coupling of (1) a stochastic rainfall generator able to reproduce extremes down to 5-minute temporal resolutions; (2) a processes-based event-scale cropland erosion model (Dynamic WEPP, DWEPP); and, (3) a state-of-the-art frequency analysis method that explicitly accounts for rainstorms occurrence and properties. To our knowledge, this is the first study in which DWEPP runoff and soil loss are calibrated at the plot-scale on cropland (NSE is 0.82 and 0.79 for event runoff and sediment, respectively). We generated 300-year stochastic simulations of event runoff and sediment yield based on synthetic precipitation time series. Based on this data, the mean annual soil erosion in the study site is 0.1 kg m−2 [1.1 t ha−1]. Results of the risk analysis indicate that individual extreme rainstorms (>50 return period), characterized by high rainfall intensities (30-minute maximal intensity > $\sim$60 mm h−1) and high rainfall depth (>$\sim$200 mm), can trigger soil losses even one order of magnitude higher than the annual mean. The erosion efficiency of these rainstorms is mainly controlled by the short-duration (≤30 min) maximal intensities. The results demonstrate the importance of incorporating the impact of extreme events into soil conservation and management tools. We expect our methodology to be valuable for investigating future changes in soil erosion with changing climate.