C41D-04:
Correcting Inadequate Model Snow Process Descriptions Dramatically Improves Mountain Hydrology Simulations

Thursday, 18 December 2014: 8:45 AM
John W Pomeroy and Xing Fang, University of Saskatchewan, Saskatoon, SK, Canada
Abstract:
The vast effort in hydrology devoted to parameter calibration as a means to improve model performance assumes that the models concerned are not fundamentally wrong. By focussing on finding optimal parameter sets and ascribing poor model performance to parameter or data uncertainty, these efforts may fail to consider the need to improve models with more intelligent descriptions of hydrological processes. To test this hypothesis, a flexible physically based hydrological model including a full suite of snow hydrology processes as well as warm season, hillslope and groundwater hydrology was applied to Marmot Creek Research Basin, Canadian Rocky Mountains where excellent driving meteorology and basin biophysical descriptions exist. Model parameters were set from values found in the basin or from similar environments; no parameters were calibrated. The model was tested against snow surveys and streamflow observations. The model used algorithms that describe snow redistribution, sublimation and forest canopy effects on snowmelt and evaporative processes that are rarely implemented in hydrological models. To investigate the contribution of these processes to model predictive capability, the model was “falsified” by deleting parameterisations for forest canopy snow mass and energy, blowing snow, intercepted rain evaporation, and sublimation. Model falsification by ignoring forest canopy processes contributed to a large increase in SWE errors for forested portions of the research basin with RMSE increasing from 19 to 55 mm and mean bias (MB) increasing from 0.004 to 0.62. In the alpine tundra portion, removing blowing processes resulted in an increase in model SWE MB from 0.04 to 2.55 on north-facing slopes and -0.006 to -0.48 on south-facing slopes. Eliminating these algorithms degraded streamflow prediction with the Nash Sutcliffe efficiency dropping from 0.58 to 0.22 and MB increasing from 0.01 to 0.09. These results show dramatic model improvements by including snow redistribution and melt processes associated with wind transport and forest canopies. As most hydrological models do not currently include these processes, it is suggested that modellers first improve the realism of model structures before trying to optimise what are inherently inadequate simulations of hydrology.