This study was broken up into two phases, both with an emphasis on using a wide variety of snow observations. The first portion of the study looked at running the model retrospectively over a five year period using the same configuration as the NWM (1 km LSM, 250m overland routing, and NHDPlus Version 2.1 channel network). The land surface component of WRF-Hydro chosen is also the same version of Noah-MP that is used in the NWM. Forcing from the National Land Data Assimilation System (NLDAS) was downscaled from its native 0.125 degree resolution to the 1 km land modeling grid used in WRF-Hydro. In-situ observations from the SNOTEL network were used to assess model performance, along with gridded analysis from the Snow Data Assimilation System (SNODAS) being ran at the OWP office in Chanhassen, MN. Additionally, several flights from the Airborne Snow Observatory lidar program were used in performance metrics for key basins out west, such as the Upper Rio Grande in Colorado, and the Tuolumne basin in California. This study also compared gridded output from the NWM to snow products from the Joint Polar Satellite System (JPSS) suite. Results show the model has a low bias across portions of the mountain west, while magnitude and timing is better east of the Rocky Mountains. Some events during the retrospective period were missed by the model, but this may be due to the nature of the precipitation being mixed-phase. These analyses will serve as a benchmark for future model upgrades to parameters and physics upgrades pertaining to snow.
The second portion of this study evaluated the model using real-time land surface output from the NWM being ran at the National Center for Environmental Prediction (NCEP) for the first half of the 2016/2017 winter season. Observations used in real-time analysis are similar to the retrospective study, minus ASO data as flights primarily occur during the spring timeframe. Output from the Analysis and Assimilation cycle was evaluated for key select regions, such as the mountain west and where significant snow events occurred. These findings offer a first glimpse into the real-time performance of the snow component within the NWM, which will serve as a benchmark for future upgrades and enhancements. These initial findings will also assist forecasters going into the spring melt season in regions that rely heavily on snowmelt for water resources management. The analysis will also assist in identifying particular regions where the model needs improvement for the next version of the NWM. This, combined, with the benchmark retrospective analysis, will serve as a key component within the research-to-operations cycle of the NWM. Input from the academic community alongside internal physics upgrades will improve model performance going forward.