Modelling and prediction of wind damage in forest ecosystems of the Sudety Mountains, SW Poland

Windstorms are one of the most important disturbance factors in European forest ecosystems. An understanding of the major drivers causing observed changes in forests is essential to improve prediction models and as a basis for forest management. In the present study, we use machine learning techniques in combination with data sets on tree properties, bioclimatic and geomorphic conditions, to analyse the level of forest damage by windstorms in the Sudety Mountains over the period 2004-2010. We tested four scenarios under five classification model frameworks: logistic regression, random forest, support vector machines, neural networks, and gradient boosted modelling. Gradient boosted modelling and random forest have the best predictive power. Tree volume and age are the most important predictors of windstorm damage; climate and geomorphic variables are less important. Forest damage maps based on forest data from 2020 show lower probabilities of damage compared to the end of 20th and the beginning of 21st century.

Automated summary

This study utilized machine learning techniques to analyze the forest damage caused by windstorms in the Sudety Mountains from 2004 to 2010. The results showed that tree volume and age were the most important predictors of windstorm damage, while climate and geomorphic variables were less important. Gradient boosted modelling and random forest performed the best in terms of predictive power.