Eastern hemlock (Tsuga canadensis), an ecologically important species, has been challenged by the invasive Hemlock woolly adelgid (HWA) (Adelges tsugae), resulting extensive mortality.  The consequences of widespread hemlock mortality for hemlock-associated and hemlock-dependent aquatic and terrestrial wildlife will be devastating. In order to understand the extent of our eastern hemlock resources and the extent to which the invasive HWA will affect these resources, maps must be produced showing accurate areas of hemlock cover.  Ground mapping the patchy distribution of eastern hemlock is impractical, so alternative techniques are needed. In this study, we investigate the feasibility and accuracy of three different approaches to map eastern hemlock: 1) a decision tree analysis using remotely-sensed spectral data and environmental variables (TREE-ETM+), 2) a decision tree analysis using only environmental data (TREE-OMIT), and 3) maximum entropy species distribution modeling using presence-only data (MaxEnt).

Results/Conclusions

Of the three models, the TREE-ETM+ model classified the most area as eastern hemlock coverage with the distribution spread generally throughout the study area. The TREE-OMIT model moderately underestimated eastern hemlock distribution, while the MaxEnt model over-fit the predicted eastern hemlock data.  The TREE-ETM+ model was the most difficult and time consuming technique, followed by the TREE-OMIT model. The MaxEnt program was the least complicated and quickest model to execute, but was also the least accurate. All three methods produced reliable maps of eastern hemlock distribution in the Coal Field and Pine Mountain study areas, with no accuracy value less than 78.7%.