Establishing Transferable Sub-Pixel Relationships for Estimating Snow Depth from Remotely-Sensed Snow Covered Area and Terrain Variability

Snowmelt is the primary water source in the Western United States and mountainous regions globally. Forecasts of streamflow and water supply rely heavily on snow measurements from sparse observation networks that may not provide adequate information during abnormal climatic conditions. In this regard, remote sensing can be used to monitor snow covered area (SCA), which we hypothesize can be used in conjunction with terrain information to estimate spatially explicit snow depth (SD). Small-scale terrain variability can be considered a proxy for the snow holding capacity of the ground. SCA should be more sensitive to changes in snow depth for smooth, or low variable terrain, and less sensitive to rougher terrain. To this end, we have developed a method that is not expected to depend on repeated climatic conditions because it accounts for the static accumulation capacity rather than dynamic processes. In preliminary investigations, a LiDaR dataset from 2010 from Green Lakes Valley, Colorado, USA (Harpold et al. 2014) was used to relate snow depth with fSCA and the sub-pixel terrain variability. Snow depth (dependent variable) and fSCA (independent variable) were aggregated from 1 meter to 30 meters from the LiDaR snow depth product while terrain variability metrics such as the coefficient of variation of elevation were calculated using the 900 1-meter elevation pixels inside each 30 meter pixel. Single linear regression of SD fit with fSCA explains 38% of the variability with a mean absolute error (MAE) of 0.36 m, but the goodness of fit increases to an average of 53% with MAE of 0.25 m as the data is binned by elev-cv; this may indicate that SD-SCA relationships vary by terrain type (slope, aspect, etc.). Further analysis of the scales at which these relationships are applicable and the viability with off-the-shelf DEM and fSCA products is needed. The utility of these relationships is such that snow depth could be estimated above treeline for any set of climatic conditions and could have far-reaching implications for understanding snow distribution and forecasting water supply.