Modelling damage occurrence by snow and wind in forest ecosystems

•Analysis of damage occurrence by snow and wind based on machine learning (BRT).•Prediction maps, shape-free relations and interactions between main variables.•Main site variables to predict damage occurrence are latitude, altitude and slope.•Main forest variables to predict damage occurrence are density, diameter and height.•More heterogeneous forest structures make birch dominated stands more resistant. Snow and wind damages are one of the major abiotic disturbances playing a major role in forest ecosystems and affecting both stand dynamics and forest management decisions. This study analyses the occurrence of wind and snow damage on Norwegian forests, based on data from four consecutive forest inventories (1995–2014). The methodological approach is based on boosted regression trees, a machine learning method aiming to demonstrate the effects of different variables on damage probability and their interactions as well as to spatialize damage occurrence to make predictions. In total, 313 models are fitted to detect trends, interactions and effects among the variables. The main variables associated with damage occurrence are consistent across all the models and include: latitude, altitude and slope (related to site and location), and tree density, mean diameter and height (related to forest characteristics). The results show that stand dominant height is a key variable in explaining damage probability, whereas stand slenderness has a limited effect. More heterogeneous forest structures make birch dominated stands more resistant to damage. Finally, the models are translated into occurrence maps, to provide landscape-level information on snow and wind damage hazard. Further application of the models can be oriented towards assessing the probability of damage for alternate stand management scenarios.