Refining the processing of paired time series data to improve velocity estimation in snow flows

For effective avalanche risk mitigation, numerical models with a correct description of snow rheology are needed. Conventionally, velocity in snow flow experiments is inferred by cross-correlating the voltage signals of paired sensors. The intention of this paper is to reconsider this problem to enhance processing of these data, leading to more effective estimates of fluctuating velocity quantities. The algorithm consists of a wavelet decomposition, a denoising step and a weighting method for the reconstituted signal. The resulting velocity time series are both consistent and informative, providing confidence that one can analyse not only the mean velocity profiles, but also the velocity distribution. Our approach is illustrated using a typical chute experiment undertaken at Col du Lac Blanc in the French Alps. Not only has the mean velocity profile a more complex shape than the bilinear one postulated from the results of the standard cross-correlation processing, but the probability distribution functions of the velocity at different heights is much more continuous and dispersed, revealing interesting new patterns of greater dynamical relevance.