Method Development to Process Hyper-Temporal Remote Sensing (RS) Images for Change Mapping

This research showed that by using a modified version of the adaptive Savitzky- Golay filter to force an upper envelope to a noisy NDVI time-series, it is possible to reduce noise, and obtain images that are representative of the good original NDVI values. Following the removal of periods of high variability (due to snow), from an NDVI noise-reduced time-series, it is possible to detect individual years within a class that are significantly different from yearly averaged values of the class. This is the first indication of classes with tendency of changes. By using a formulated Key Index in this research, based on the averaged years within a class, pairs of classes surpassing a Key Index threshold can be merged. This method takes into account of inter-annual variability. Furthermore, the method produced a temporal reference library, with a unique legend of signatures that could be used for change mapping studies.