A rules-based approach for predicting the eastern hemlock component of forests in the northeastern United States

Abstract

The expanding threat of hemlock woolly adelgid (Adelges tsugae Annand) infestation has generated interest in locating eastern hemlock ( Tsuga canadensis (L.) Carr.). Prior studies have incorporated remotely sensed imagery to detect eastern hemlock presence or absence. The goal of this study was to develop methodology to quantify hemlock abundance using software and data accessible to forest managers. Three seasons of Landsat ETM+ scenes served as the imagery basis, whereas simple (slope, aspect, and curvature) and detailed (heat and wetness) environmental indices were extracted from a digital elevation model. Three hundred and forty-nine forest plots representing the typical forest cover found in the Catskill Mountain Region, New York, served as ground reference; model input used the percentage of hemlock basal area for each plot. The models generally underpredicted in plots with substantial hemlock composition, whereas overpredictions mainly occurred in mixed forests that lacked hemlock. Underpredictions negated overpredictions in mixed hemlock deciduous forests resulting in a neutral model. Correlation coefficients ranged from a high of 0.67 for the model created from three Landsat images to a low of 0.01 for the heat and wetness indices model. Although the models were typically within 10% of field measurements, there was no overall benefit in including topographic indices for mapping hemlock abundance.