Empirical Snow Modeling

Spatial Analog Models of April 1 SWE and Snow Residence Time

Building on the work of Luce et al., (2014), Dr. Charlie Luce and I explored spatial analog models of April 1 SWE and snow residence time (Lute and Luce, 2017).  We used winter temperature and precipitation at SNOTEL sites to build the spatial analog models.  These models captured the nonlinear relationship between temperature, precipitation, and snow metrics and provided very strong fits to the data. One interesting finding of this work was the overwhelming importance of precipitation (relative to temperature) to April 1 SWE at most SNOTEL sites. Simple spatial analog models can be used to quickly evaluate output from global climate models, providing a computationally efficient way to compare models and scenarios in terms of their snow projections.


Spatial analog model of April 1 SWE shown with mean winter temperature (x-axis) and cumulative winter precipitation (y-axis). Each point represents a SNOTEL site. Contour lines and colors indicate April 1 SWE.


Historical and future (2080s, RCP8.5) snow residence time calculated by forcing the spatial analog model of snow residence time with MACAv2 downscaled global climate model projections.

We evaluated potential tradeoffs between model complexity and spatiotemporal transferability of these spatiotemporal analog models using a series of non-random cross validation tests. We found that low to moderate complexity models transferred best. We also concluded that temporal cross validation tests suffer from pseudoreplication and don’t provide a strong test of the model’s transferability. A more rigorous test of model transferability would use truly distinct data for calibration and validation, such as one set of years in one region for calibration and a different set of years in another region for validation.

Spatiotemporal SWE Variability of the Columbia River Basin

Understanding the spatial and interannual variability of snowpack in montane regions such as the western U.S. is a key component of water management. Studies in this vein have typically focused on west wide analyses and found correlations with the El Nino Southern Oscillation and the Pacific Decadal Oscillation. However, smaller scale, subregional analyses may offer different insights into the spatiotemporal variability of snowpack which may be useful for water management at local scales as well as ecological assessments and planning. We considered March 1, April 1, and May 1 snow water equivalent at 243 SNOTEL sites within the Columbia River Basin. Using a rotated principal component analysis, we found that the first and second components of April SWE were strongly linked to latitudinal shifts in the jet stream. The second component was also correlated with the occurrence of Pineapple Express events. The third component was most strongly correlated with interannual temperature and precipitation variability.

This analysis was presented at the 2015 Western Snow Conference (poster, paper).


Spatial loadings of principal components 1, 2, and 3 for March 1, April 1,and May 1 SWE. The percent of variance explained by each principal component is indicated in the upper right of each plot.