This research deals with an application of Support Vector Machines to the exploratory analysis and spatio-temporal prediction of avalanches based on historical data. The problem is considered as a two-class classification problem where the purpose is to discriminate the conditions leading to avalanche releases from the safe situations using empirical data. Statistical binary classification was performed in a high dimensional geo-feature space with both spatially and temporally varying features describing local avalanching conditions. The set of features included topographic indices computed from the digital elevation model, meteorological information and snowpack conditions. We applied Support Vector Machines (SVM), a classifier based on structural risk minimization principle, i.e. on the minimization of training error and model complexity, giving rise to low validation error. The main objectives considered in the present research are the following: modeling of spatio-temporal patterns in avalanche data and forecasting of avalanche activity at the scale of the avalanche paths for 2 validation winter seasons; selection of relevant features using dimensionality reduction techniques; spatio-temporal mapping of the avalanche susceptibility indicators; enhancing the interpretability of the produced forecast by supplying conventional outputs such as reference events from the past, aspect/elevation diagrams of the avalanche danger, and sensitivity surfaces. The results are based on the dataset collected over 18 years of observations at an avalanche-prone site of Ben Nevis, Lochaber region, Scotland.

Keywords: Avalanche forecasting; Spatio-temporal prediction; Statistical classification; Support vector machines

Biography: Giona Matasci was born in Locarno, Switzerland, in 1985. He received the M.Sc. degree in environmental sciences from the University of Lausanne, Lausanne, in 2009. He is currently working toward the Ph.D. degree at the Institute of Geomatics and Analysis of Risk (University of Lausanne) in the field of machine learning and its applications to remote sensing and environmental data analysis.