Inferring spatial distribution from sparse 
		observations: snow depth
			   James McCreight
			Abstract

Estimating the spatial distribution of environmental variables can be important but challenging. Small-scale (~10m) spatial variability of snow depth can be important for estimating water balances, stream flow timing, and atmospheric energy balances which have applications to hydropower and reservoir management, weather forecasting, ecological processes modeling, and validation of satellite observations.

Because observations are sparse we traditionally have had only a very limited view of our ability to accurately infer the spatial distribution of snow depth. Significant questions about the estimation have persisted. The development of airborne Light Detection and Ranging (LiDAR) technology has provided a new technique for measurement of snow depth which can provide detailed, non-sparse observations. In this study, LiDAR provides over a half million observations six 1.17km2 study sites, with a nominal spatial resolution of 1.5m.

This talk will highlight the following issues surrounding estimation of spatial snow depth distribution from sparse observations: regression model choice, observation count (value of observations), sampling design, cross-validation bias, predictor quality, and model spatial resolution.

Our results demonstrate how observation and modeling choices can compensate for each other and highlights the value of airborne LiDAR observations for mapping snow depths at large spatial scales.