Using snowboards and lysimeters to constrain snow model choices in a rain-snow transitional environment

Abstract:

Physically based models of the hydrological cycle are critical for testing our understanding of the natural world and enabling forecasting of extreme events. Previous intercomparison studies (i.e. SNOWMIP I & II, PILPS) of existing snow models that vary in complexity have been hampered by multiple differences in model structure. Recent efforts to encompass multiple model hypothesizes into a single framework (i.e. the Structure for Understanding Multiple Modeling Alternatives [SUMMA] model), have provided the tools necessary for a more rigorous validation of process representation. However, there exist few snow observatories that measure sufficient physical states and fluxes to fully constrain the possible combinations within these multiple model frameworks. In practice, observations of bulk snow states, such as the snow water equivalent (SWE) or snow depth, are most commonly available. The downfall of calibrating a snow model using such single bulk variables can lead to parameter equanimity and compensatory errors, which ultimately impacts the skill of a model as a predictive tool.

This study provides two examples of diagnosing modeled snow processes through novel error source identification. Simulations were performed at a recently upgraded (Oct. 2012) snow study site located at Snoqualmie Pass (917 m), in the Washington Cascades, USA. We focused on two physical processes, new snow accumulation and snowpack outflow during mid-winter rain-on-snow events, for their importance towards controlling runoff and flooding in this rain-snow transitional basin. Main results were: 1) modifying the snow model structure to match what was actually observed (i.e. a snow board), allowed the attribution of daily errors in model new snow accumulation to either partitioning, new snow density, or compaction. 2) Observed snow pit temperature profiles from infrared cameras and manual thermometers found that cold biases in the model snowpack temperature prior to rain-on-snow events could delay model outflow multiple days compared to lysimeter observations. These two examples illustrate a commonly discussed, but rarely implemented, method of diagnosing model errors using multiple observations of internal snowpack states.