Change Detection in Image Time-Series Affected by Directional Reflectance and Phenological Variability: Application to Forest Disturbance Monitoring

Rapid identification of vegetation change over extensive areas requires the use of high temporal resolution data from wide-swath sensors, but analysis of such data is difficult due to the effects of vegetation directional reflectance and phenological variability. The magnitude of these effects can be large relative to the often subtle reflectance changes associated with natural forest disturbance, especially at coarse spatial resolutions. To address these challenges we propose a novel change detection methodology. We first apply a median-based compositing technique to a multi-year surface reflectance data record to create a reliable baseline phenological time-series. We then apply an algorithm - Dynamic Class Delta Resistant-Z Analysis - to reduce the magnitude of the change signal deriving from extraneous sources, allowing the use of composite images containing pixels collected under substantial variability in sensor and solar geometry. We apply these methods to generate an evolving disturbance product at an approximately weekly timestep and 250-meter spatial resolution for recent years in the U.S. Pacific Northwest. Comparison of results with other methodologies indicate the technique provides advantages that should substantially enhance the ability to reliably and rapidly detect natural forest disturbance.