A primary activity required to support sustainable forest management is the detection and mitigation of forest disturbances. These disturbances can be planned, through urbanization and harvesting, or unplanned, through insect infestations or fire. Detection and characterization of disturbance types are important, as different disturbances have different ecological effects and may require unique managerial responses. As such, it is necessary for forest managers to have as complete and current information as possible to support decision making. In this study, we developed a framework to automatically detect and label disturbances derived from remotely sensed images. Disturbances were detected through traditional image differencing of medium-resolution imagery (Landsat-7 Enhanced Thematic Mapper Plus (ETM+), resampled to 30 m) but were refined and augmented through comparison with edge features extracted from high spatial resolution satellite imagery (Indian Remote Sensing (IRS) satellite 1C panchromatic imagery, resampled to 5 m). By incorporating spectral information, derived composite band values (tasselled cap transformations), spatial and contextual information, and secondary datasets, we were able to capture and label disturbance features with a high level of overall agreement (91%). Areal features, such as harvest areas, are captured and labelled more reliably than linear features such as roads, with 92% and 72% agreement when compared with control data, respectively. By incorporating rule-based disturbance attribution with remote sensing change detection, we envision the update of land cover databases with reduced human intervention, aiding more rapid data integration and opportunities for timely managerial responses.
ASJC Scopus subject areas
- Earth and Planetary Sciences(all)