Toward benchmarking land surface models for large scale snow predictions using in-situ station observations

Given the high spatial variability of snowpack distribution, it is always challenging to effectively assess large-scale snow prediction results from spatially distributed land surface models (LSMs) with point-based observations from in-situ snow measurement stations, due to the incommensurability between the model predictions and station observations such as scale differences. This renders it difficult to assess the benefit from improved model forcing/physics or snow data assimilation. In this study, we present two approaches for overcoming this issue, including 1) evaluating the area mean of model snow predictions against that of station observations in a sufficiently large region with reasonable station density, and 2) evaluating the anomaly departure of model snow predictions from the long-term climatology against that of station observations. The underlying assumptions for these approaches are that large-scale area means and long-term anomalies of snow variables (e.g., snow water equivalent (SWE) and snow depth) are relatively scale independent. We test these approaches for evaluating the SWE and snow depth simulations from the Noah LSM over the contiguous US at 12.5 km resolution for a period of 32 years (1980-2011), using observations from Snowpack Telemetry (SNOTEL) stations and the Global Historical Climatology Network (GHCN). Two statistical models based on the long-term daily and monthly climatologies of the station observations are used as benchmarks for evaluating the Noah LSM. We will present results showing how these approaches (as opposed to directly comparing model predictions against station observations) can help to properly assess the improvements (if any) in snow predictions from enhanced model physics and forcing, as well as snow data assimilation.