Dealing with sub-pixel landscape elements in distributed rainfall-runoff modelling in agricultural catchments

Abstract

Vegetated landscape elements (vLEs) (e.g. hedges and grass buffers) are increasingly recognised for their ability to retain more water longer and mitigate downstream flood risk. To optimise positioning of these –typically small– vLEs, the impact of possible configurations needs quantifying, often requiring numerous hydrological model runs. To limit computational time, models must be run at lower spatial resolution leading to sub-pixel vLEs. The performance of a distributed rainfall-runoff model at 5 m resolution was assessed for 15 historical rainfall events in a 191 ha agricultural watershed in the Belgian loess belt. The model was then upscaled to 20 m resolution using four scaling approaches for saturated hydraulic conductivity (K_s) and Manning's coefficient, and three methods to set the hydro-physical parameters of subpixel vLEs in the upscaled model. The high-resolution model performed best for K_s equaling 0.72 mm h⁻¹. The upscaled model performed best when applying a flow length-based scaling factor for the Manning's coefficient, decreasing the RMSE by 25% and 10% for discharge volume and peak discharge rate respectively. Adjusting K_s and Manning's coefficient of vLE pixels using upslope area-based weighting was most effective for discharge volume, achieving an RMSE of 10.80% and R² of 0.64. Peak discharge rate could not be modelled accurately with sub-pixel vLEs at 20 m resolution. Our research can support scenario analysis in which accounting for the reduction of discharge volume caused by the presence of vLEs and their spatial configurations matters and therefore can support landscape design studies in the context of flood risk mitigation.