Downscaling Surface Water Depth to Simulate Small-Scale Ponding: A Case Study in the Prairie Pothole Region of North America

Abstract:

Water ponding in small surface depressions plays an important role in hydrology such as flood mitigation and habitat provisions. Physically based integrated hydrologic models have the ability to explicitly model such small-scale ponding but typically require high resolution to resolve individual ponds and hence significant computational resources. We propose to overcome this challenge with a downscaling technique. This involves simulating surface water depth at a relatively coarse resolution compared to the pond scale, and then resolving areas of ponding through a regression-based downscaling technique. We examined the application of the above approach for the Cottonwood Lake Study Area in North Dakota within the Prairie Pothole Region (PPR). Using ParFlow-CLM, we simulated the surface water depth at 10m resolution to adequately capture the filling of the potholes from a dry initial condition. The results were then aggregated to a coarser resolution from which a regression function was applied to establish the relationship between the coarse grid value and the corresponding 10m fine grid values. This relationship forms the basis of the downscaling that enables the model to be run at a coarser resolution, thereby reducing computational cost significantly. Preliminary validation results show downscaling of the simulated surface water depth from resolutions of 30m, 60m, 90m and 180m generally matches that of 10m resolution well, with a decreasing Nash-Sutcliffe Coefficient of 0.94, 0.88, 0.80 and 0.72 respectively. It was also found that the approach was able to reproduce the distribution of the surface water depth fairly accurately nearer to the center of the depression but the performance reduced towards the ponding edge due to co-existence of saturated and unsaturated grid cells. For partially filled depressions in undulating terrain, the surface water depth tends to be underestimated at the deeper locations and overestimated at the shallower locations.