Comparison of Semi-Distributed and Fully Distributed Hydrological Models in Complex Terrain

This study examines the performance of semi-distributed and fully distributed hydrological models for flood runoff prediction in the Napa River basin, California. ModClark is an event-oriented semi-distributed rainfall-runoff model, using a soil curve number approach to compute effective rainfall; it uses two parameters (time of concentration and storage coefficient) to represent the watershed response to input precipitation. The Research Distributed Hydrological Model (RDHM) is a fully distributed gridded version of the Sacramento Soil Moisture Accounting (SAC-SMA) continuous simulation model used operationally by the NWS for river forecasts nationwide. The model has six soil moisture states and 16 parameters, not counting the 12 monthly adjustment factors of potential evaporation. The gridded version has overland flow and channel routing algorithms using kinematic wave procedures. The comparison involves several criteria including simulation accuracy, ease and appropriateness of parameter estimations, sensitivity to space and time scales, and sensitivity to precipitation magnitude and timing. The MRMS (Multi-Radar/Multi-Sensor) data is used as the spatially distributed rainfall data and MPE (Mean Precipitation Estimation) data is used to provide spatially lumped rainfall data to force both models. Data for two storm periods are used to evaluate a single peak event (Dec. 2014) versus a multi-peak event (Feb. 2015). The river gage for the Napa River at St Helena (USGS 11456000) gage is used for comparison purposes. Preliminary conclusions include that the calibration process ModClark is more straightforward and easier than RDHM as it has fewer parameters. As expected, calibrated versions of the both models outperformed the uncalibrated models. In terms of the double peak, RDHM is superior to ModClark for the second peak as it is a continuous simulation which updates the model states as it progresses to the next event. The quality of the precipitation input data affects the results from both models. The wide range of accuracies among model results suggest that factors such as model formulation, parameterization, and calibration, and the skill of the modeler can have a bigger impact on simulation accuracy than simply whether or not the model is semi-distributed or distributed.