A MULTI-MODELING APPROACH TO ADDRESS SNOW MODEL UNCERTAINTY FOR HYDROLOGIC PREDICTION

Streamflow forecasting methods are moving towards probabilistic approaches that quantify the uncertainty associated with the various sources of error in the forecasting process. The multi-model averaging methods, which try to address modeling uncertainties by considering more than one model, are gaining much popularity in hydrologic prediction. Multi-modeling also offers a means to advance forecast systems without removing the existing skill of current forecast models. The primary snow accumulation and ablation model in the US National Weather Service streamflow prediction system is the temperature-based SNOW17 model. We have altered the SNOW17 snowpack heat exchange and melt subroutines and applied a simplified energy balance approach (we call this the SNOW17 Energy Balance model (SNOW17-EB)). The effects of wind, water retention and release, and ground surface heat exchange processes are parameterized according to the original SNOW17 structure. Three versions of the SNOW17-EB with varying albedo formulations were created. The SNOW17 and the SNOW17-EB models were calibrated via the Shuffled Complex Evolution (SCE) using three different objective functions, resulting in 12 models with varying parameters or structures. The 12 models were then combined using Bayesian Model Averaging (BMA). The individual models, BMA predictive mean, and BMA predictive variance were evaluated for six SNOw TELemetry (SNOTEL) sites in the western U.S. We found that the snow models performed best at the colder sites with high winter precipitation and little mid-winter melting. Model weights range from 0-58%, and at most sites all models received some weighting. An individual snow model would often out-perform the BMA predictive mean. However, observed snow water equivalent (SWE) was captured within the 95% confidence intervals of the BMA variance at an average rate of 80% at all sites.