Snow mechanical properties variation at slope scale, implication for snowpack stability assessment and snow cover models

Snow avalanches represent a natural hazard for infrastructures and backcountry recreationists. Risk assessment of avalanche danger is difficult due to sparse nature of available observations informing on snowpack mechanical and geophysical properties. Spatial variability of these properties also add complexity to the decision-making and route finding in avalanche terrain for backcountry recreationists. Snow cover models simulate snow mechanical properties at fairly good resolution (around 100 m). However, small-scale variability such at the slope scale (5-50 m) remains critical to monitor given that slope stability and the possible size of an avalanche are governed by such scale. In order to better understand and predict the spatial variability at the slope scale, this work explores linkages between snow mechanical properties and microtopographic indicators. First, we compare their covariance models and scaling properties. Then, we predict snow mechanical properties, including point snow stability, using GAM spatial models (Generalized additives models) with microtopographic indicators as covariates. Snow mechanical properties such as snow density, elastic modulus, shear modulus and snow microstructural strength were measured at multiple locations over several studied slopes (20-40 m) using a high-resolution penetrometer (SMP), in Rogers Pass, British-Columbia, and Mt Albert, Québec. Point snow stability such as the skier crack length, critical propagation crack length and a skier stability index were derived using the snow mechanical properties from SMP measurements. Microtopographic indicators such as the topographic position index (TPI), vegetation height and proximity, Winstral index (wind-exposed/sheltered area) and potential radiation index were derived from UAV surveys with sub-meter resolution. We computed the variogram and log-log variogram of snow mechanical properties and microtopographic indicators. The comparison shows some similarities in autocorrelation distances for snow depth, snow density, snow microstructural strength, TPI, vegetation height and the Winstral index. GAM models suggest several significant covariates such as snow depth and snow surface slope, but also TPI, Winstral index, vegetation height and distance to vegetation. The percentage of variance explained is around 50% ranging from 20% to 80%. Models predictions were better for the slab depth and slab density with higher variance explained (around 60/70%) with lower RMSE than point snow stability indicator (around 40%) with higher RMSE. At the slope scale, snow surface slope and snow depth remain the most important spatial indicators of point snow stability for backcountry recreationists in their route-finding decision making. The point snow stability map generated represents a good teaching material in avalanche skill training and awareness course. In future work, assuming that snow cover models simulate the mean snow mechanical properties of a simulation cell, the covariance function of microtropographic indicators could be used to infer the covariance function of snow mechanical properties using a gaussian process/Bayesian framework as a sub-grid parametrization scheme.