Development and comparison of various stand- and tree-level modeling approaches to predict harvest occurrence and intensity across the mixed forests in Maine, northeastern US

To overcome existing knowledge gaps regarding altering harvest activities in the mixed species and heavily forested USA state of Maine, we used continuous forest inventory data to develop statistical models that predict harvest probability of occurrence and intensity. Among the three modeling approaches examined, the first one directly predicted stand-level basal area removal as a percentage of initial total stand basal area. The second approach first predicted stand-level harvest probability and then the stand-level basal area removal of the harvested plots only. The third approach used the stand-level harvest probability equation of the second approach, while subsequently predicting individual tree harvest probability for only the harvested plots. Among the most influential stand-level attributes were quadratic mean diameter, stand density, elevation, and ownership type, while the most influential tree-level attributes were diameter at breast height, basal area in larger trees, and species. Differences in prediction accuracy between modeling approaches were small with the third approach performing slightly better. Our findings suggest that harvesting in Maine might be less opportunistic and short-term driven than generally perceived. Overall, the analysis highlights the complex array of factors that influence harvesting patterns and provides a framework for better representing contrasting harvest behavior in future wood supply projections.