What Controls Snow Depth at the Watershed Scale?: Measuring Spatial and Temporal Variability in the Effects of Topography and Vegetation

Mountain hydrology is often controlled by small-scale variation in topography and vegetation, yet our knowledge of the effect of specific features relating to these aspects of a given basin is limited. With respect to snowpack, in particular, similar weather patterns can yield dramatically different snow depths in adjacent areas yet our ability to predict snow depth at a given location is poor. In the Western U.S. and many semi-arid regions near mountain ranges worldwide, over half of annual water supply is derived from snowmelt; thus, estimating the distribution of snow in a watershed and understanding the implications of a changing climate on this distribution are of utmost importance for water supply planning and ecosystem management in these regions.

As airborne and satellite remote sensing capabilities increase, we will continue to gain an increasingly detailed knowledge of physical characteristics of individual watersheds, and we will better understand how both natural climate variability and climate change are affecting vegetative patterns through observation. To make the link between these changes and hydrologic changes, we need improved methods for determining the complicated, nonlinear interactions between terrain and vegetation that control hydrology. In this study, we focus on empirical relationships between numerous topographic/vegetation features and snow depth in the Tuolumne River Basin in California.

In this study, we use Geographically Weighted Regression (GWR) to identify the spatial variability of the effect of these features and to characterize spatial patterns. The presented findings focus on an exploratory analyses of the change in spatial variability of specific topography-snow or vegetation-snow relationships (i.e. regression coefficients) observed during the snow accumulation and ablation periods. We hypothesize about potential physical processes driving these differences. Additionally, we compare the predictive skill of a model that accounts for local nonstationarity in these relationships with a watershed-scale global model. Applying nonstationarity in relationships calculated from 2013 and 2014 snow depth observations to 2015 observations, we obtain a 3cm improvement in RMSE of a simple multiple regression (p < .001), suggesting inter-seasonal persistence of these spatial patterns and the potential for improving the predictive power of statistical snow distribution models.