Abstract:
Time series Landsat data have been used to track ecosystem disturbances using an algorithm such as the vegetation change tracker. However, efficiently identifying and separating types of disturbances (e.g., wildfires and harvests) still remain a technical challenge. In this letter, we tested the support vector machine algorithm in separating forest disturbance types, including wildfires, harvests, and other disturbance types (a generalized disturbance class, including insect disease outbreak, tornado, snow damage, and drought-induced mortality) in the Greater Yellowstone Ecosystem (GYE) using annual Landsat images from 1984 to 2010. The algorithm has been proven to be highly reliable, with overall accuracy about 87% for the study region. Average producer's and user's accuracy for wildfires and harvests were 85% and 90%, respectively. Based on the mapped forest disturbance type results, fire was the most dominant disturbance in GYE National Parks (NPs) from 1984 to 2010, affecting over 37% of the forested area in GYE NPs, whereas other disturbances such as insect and disease outbreaks were more frequent in national forests of the region during this time interval. With the free public access of the Landsat data and careful selection of training samples, this method can be useful in other ecosystems with similar disturbance dynamics as GYE.
