Hierarchical forest classification with thermal and multispectral imagery using modified regression tree techniques

This study utilizes imagery from the Advanced Thermal Land Applications Sensor across the visible, near infrared and thermal portions of the electromagnetic spectrum to perform hierarchical forest classifications. Use of thermal data for forest classification is largely ignored, even when using visible and near infrared imagery collected from the same sensor. This is primarily due to the difficulty in separating surface emissivity from surface temperature. To evaluate the effect that thermal data may have, classifications were done at several levels, from simple forest delineation to individual species. A regression tree technique was employed to make full use of the thermal data and other ancillary data sets such as soil types, LiDAR intensity and various filters included in the classification. Lower order classifications such as forest vs. non-forest and hardwood-softwood responded with slightly higher accuracies to the addition of thermal imagery. However, the higher order classifications showed less response to the addition of thermal imagery.