Toward Improving Snowpack Prediction and Snow Cover Fraction Parameterization in Land Surface Models

Accurate estimation of snow cover fraction (SCF) and snow water equivalent (SWE) is essential for improving predictions of snowpack state in land surface models (LSMs). Typically, the mapping relationship from snow state variables to SCF has been restricted to be either simple linear, non-linear with terrain-related covariate, or hyperbolic tangent with conceptual or empirical physical assumptions. Furthermore, the hyperparameters in these schemes have typically been assumed to be the same at all spatial locations, thereby providing no explanatory power to enable a spatially distributed representation of SCF dynamics over different computational units.

Here, we apply spatial regularization to relate terrain static variables derived from a Digital Elevation Model to these hyperparameters in the SCF schemes. This relationship (parameter transfer function) is developed by hypothesizing its functional structure (linear, polynomial, power law etc.) at 4km pixel-scale, using the University of Arizona (UA) 4km ground-based daily snow dataset and the 1km-derived 4km Interactive Multi-sensor Snow and Ice Mapping System (IMS) snow cover extent data. Each candidate scheme is compared to its default scheme, accepted (or rejected) via the likelihood ratio test, and quantified with regards to information gain using an information-theoretic metric.

The proposed SCF schemes (hypotheses) will be evaluated for inclusion in the Noah-MP land surface model, by comparing the long-term climatologic mean and linear trend for ground/grid SCF to UA snow product and monthly MODIS SCF product. Future work will include the use of strategies for simultaneous regularization and automatic estimation of parameter transfer functions.