Statistical Modeling of Monthly Snow Depth Loss in Southern Canada

Abstract

Quantifying the dynamics of snow depth is essential for understanding freshwater availability, mitigating flood and drought hazards, and monitoring the effects of climate change in cold regions. Here, a statistical approach for describing the dynamics of monthly snow depth loss (SDL) is developed and tested in 67 climate stations throughout southern Canada. The framework fuses an input selection scheme with multiple linear regression to approximate the SDL using a set of climate proxies, either explicitly or implicitly through modeling snow depth. Our findings suggest that statistical models—if properly developed and used—have the potential to form effective tools for describing the dynamics of SDL. In particular, the implicit statistical model, in which climate proxies are selected globally among all stations, provides an accurate model (expected R2=0.75), which can outperform a frequently-used temperature-index model in a majority of stations. In addition, parameters of the statistical model can be regionalized efficiently (expected R2=0.71 for the generalized model) using latitude, longitude, and altitude. This ability can provide a basis to extend the model application into ungauged sites.