Detection of Logging Infrastructure in the State of Rondônia Using Remotely Sensed Data

ABSTRACTLogging lands (forest roads and log decks) are an underlying issue during selective logging activities, but they are responsible for most impacts on the forest. This study aimed to apply and assess the performance of five geoprocessing techniques on remotely sensed data using three different spatial resolutions to detect logging lands under forest management at the Jamari National Forest, state of Rondônia, Brazil. The research results showed that Normalized Difference Vegetation Index (NDVI) and Principal Components Analysis (PCA) presented the best overall accuracy using spatial resolutions of 5 and 10 meters, and 30 meters, respectively. Generally, the overall accuracy and Kappa statistics for the selectively logged forest classifications were not good (39.2% or lower, and 0.38 or lower, respectively). The low performance of the geoprocessing techniques is related to the subtle changes on the forest canopy cover under selective logging activities.