Improving empirical storm damage models by coupling with high-resolution gust speed data

Abstract

Empirical forest storm damage models can assist in identifying the key factors of the occurrence of storm damage in order to develop locally adapted measures to minimize damage in forests. Yet, there is a significant lack of knowledge in these models concerning the correlation between storm damage and high-impact near-surface airflow. To improve our understanding in this field, we built Random Forests (RF) and Generalized Linear Models (GLM) for evaluating the association between high resolution gust speed data and long-term, multi-event forest storm damage data from long-term permanent forest growth and yield plots. The tested gust speed data were derived from two different gust speed models: a numerical non-hydrostatic mesoscale model and a statistical model.In all RF and GLM models gust speed was a statistically significant predictor. The performance of the evaluated empirical models was very high (area under the receiver operating characteristic curve values AUC = 0.86–0.99). Depending on the type of model, the relative importance of gust speed was moderate to very high (up to 35%). However, starting from models using all significant predictors and excluding gust speed, the performance loss was almost negligible in all models. Furthermore, modeling long-term storm damage for each storm event individually performed better compared to modeling average long-term, event-unspecific storm damage.Our results demonstrate that empirical storm damage models using only gust speed as a predictor can reach moderate (GLM) to very high (RF) performance, even without any other information on terrain and forest attributes. However, if detailed terrain and forest data are available, empirical storm damage models may have such a high performance that adding gust speed data improves them very little. The correlation between gust speed and storm damage in the coupled modeling system is a fundamental first step in being able to evaluate potential changes of forest storm damage in a changing climate with potentially changing wind regimes. Additionally, further improvements could be achieved by improved representation of airflow in complex forest.