Comparison of logistic regressions and snowfall intensity–duration threshold as forecasting tools for direct-action snow avalanches in the Presidential Range, New Hampshire, USA

Abstract

Snow avalanches are an important natural hazard for backcountry skiers and alpinists in the Presidential Range (New Hampshire, United States). Over the past 20 years, over 72 people were involved in avalanche accidents. Direct-action avalanches resulting from recent snowfall dominate the avalanche regime of this region. Given this peculiarity, avalanche activity should be influenced by the intensity and duration of snowfalls. This study compares logistic regressions (LR) and a novel approach based on snowfall intensity–duration (ID) thresholds to analyse the weather patterns favourable to the triggering of snow avalanches. LR models suggest that 72-h solid precipitation, 24-h liquid precipitation, 72-h average daily maximum temperature and 72-h average winds peed are the best predictors of avalanche activity. The accuracy of the snowfall ID threshold to predict snow avalanches was similar to the accuracy of rainfall ID thresholds to identify landslide activity. However, a Monte Carlo cross-validation procedure suggests that the LR model is more accurate and more robust in predicting avalanche activity, since they allow the inclusion of more meteorological predictors. Finally, we discuss the relevance of this novel approach and the need for replicate studies in similar, direct-action avalanche-dominated contexts.