Mapping eastern hemlock: Comparing classification techniques to evaluate susceptibility of a fragmented and valued resource to an exotic invader, the hemlock woolly adelgid

Eastern hemlock (Tsuga canadensis Carriére), an ecologically important foundation species in forests of eastern North America, is currently threatened by the hemlock woolly adelgid (HWA, Adelges tsugae Annand, Hemiptera: Adelgidae), an aggressive invasive insect herbivore. HWA colonization of eastern hemlock results in rapid tree mortality. There is a pressing need to accurately determine eastern hemlock distribution in the face of expanding HWA populations to preserve this important forest species. However, efficient modeling of large geographic extents of eastern hemlock habitats to facilitate state-wide HWA management is lacking. We employ two modeling approaches, decision tree classification (based on presence–absence data) and maximum entropy (MaxEnt, based on presence-only data) method, to map eastern hemlock distribution in eastern Kentucky using a comprehensive suite of environmental parameters as predictor variables. Results demonstrate moderate model accuracies around 70%, supporting the practicality of mapping hemlock distribution over extensive regions. Comparison of the two modeling techniques suggests that decision tree classification has higher overall accuracies, while MaxEnt method was more efficient in model construction. In comparison to the decision tree method, MaxEnt suffered from possibly over-fitting as indicated by increased producer’s accuracies yet lower user’s accuracies. Our study provides useful references for selecting optimized approaches in accordance with study region characteristics and end user’s preferences.

► We model eastern hemlock distribution at a large spatial scale (>20,000 km2).
► We examine models with presence–absence and presence-only data.
► Presence–absence model is more accurate but presence-only model is more efficient.