Quantifying Uncertainty of Semi-distributed Hydrologic Model Simulations for Catchment Scale Applications

Semi-distributed hydrologic modeling frameworks are viable alternatives to fully distributed hydrologic models, which can resolve fine-scale spatial structure of hydrologic fluxes and states with improved computational efficiency. Recently, a new GIS based semi-distributed Soil Moisture and Runoff Simulation Toolkit (SMART) was developed to simulate catchment water balance using a 2-dimensional, Richards' equation based hydrologic model. SMART simulation elements are representative hillslopes or equivalent cross sections (ECS) that consist of topologically connected Hydrologic Response Units (HRUs). HRU parameters within an ECS are obtained by averaging slope and soil depth across multiple cross sections, and assigning soil hydraulic and vegetation parameters for the dominant soil type and vegetation class, respectively. While earlier investigations have shown that ECSs formulation reduces computational time significantly without compromising simulation accuracy, in some cases defining representative soil hydraulic parameters are impacted by the quality of soil texture data. To overcome this limitation, we coupled a Hybrid Markov Chain Monte Carlo (MCMC) algorithm with SMART for quantifying uncertainty in runoff simulations at catchment scale. We set-up the model over the Little Washita watershed in Oklahoma for which high resolution soil moisture and runoff observations are available. Our results illustrate that the coupled framework is able to improve runoff simulations by identifying representative soil hydraulic parameters for each soil texture class in the catchment. Future work will be focused on developing alternate ways to reduce the number of model evaluations for uncertainty quantification.