Modeling the susceptibility of an uneven-aged broad-leaved forest to snowstorm damage using spatially explicit machine learning.

Snowstorms are disturbance agents that have received relatively little research attention rather than significant disturbances that they pose to forest ecosystems. In this study, we modeled the interactions between snowstorms and different characteristics of a forest stand in northern Iran and spatially visualized the susceptibility of the stand to damage caused by snowstorms using the random forest (RF) and logistic regression (LR) methods. After a severe snowstorm in November 2021 that caused stem breakage and uprooting of individual trees, the location of 185 damaged trees was identified via field surveys and used for generating an inventory map of snowstorm damage. The thematic maps of fourteen explanatory variables representing the characteristics of damaged trees and the study forest were produced. The models were trained with 70% of the damaged trees and validated with the remaining 30% based on the area under the receiver operating characteristic curve (AUC). The results indicated the better performance of RF compared to LR in both training (0.934 vs. 0.896) and validation (0.925 vs. 0.894) phases. The RF model identified slope, aspect, and wind effect as the variables with the greatest impacts on the forest stand sustainability to snowstorm damage. Approximately 30% of the study area was categorized as high and very high susceptible to snowstorms. Our results can enable forest managers to tailor more informed adaptive forest management plans in readiness for snowstorm seasons and recovery from their damage.