Human-induced and natural disturbances are an important feature of forest ecosystems. Disturbances influence forest structure and composition and can impact crucial ecosystem services. However, deriving spatially explicit estimates of past forest disturbance across a large region can prove challenging. Researchers have recognized that remote sensing is an important tool for monitoring forest ecosystems and mapping land use and land cover change. One of the most important sources of remotely sensed imagery is the United States Geologic Survey’s Landsat program which has continuously acquired earth observations since 1972. This repository of imagery has the spatial, spectral, and temporal resolution necessary to produce maps of disturbance which are meaningful for the analysis of forested ecosystems.
In this analysis, we utilize the imagery from the Landsat archive to produce maps of forest disturbance from 1985 to 2017 for the New England states and the Canadian Maritime provinces. The change detection maps were developed using stacked generalization, a modeling technique that fuses the outputs of an ensemble of individual change-detection algorithms through the use of a secondary classifier. To better understand the error associated with these classifications, we quantified the spectral characteristics associated with different harvesting practices. Using two case studies, the 1998 ice storm and the 2016 gypsy moth outbreak in southern New England, we performed experiments to examine how the stacked generalization framework can be utilized to increase the accuracy of disturbance maps following large-scale natural disturbances. The change detection maps developed in this analysis possessed a 98.7% overall accuracy and a 27.5% balance of the errors of omission and commission. Our results indicated that adjusting the probability threshold associated with the secondary classifier in the stacked generalization framework increase the spatial coherence of disturbance patches and better capture the low- to moderate-severity disturbances.