Using temporal averaging to decouple annual and nonannual information in avhrr ndvi time series

Abstract

As regularly spaced time series imagery becomes more prevalent in the remote sensing community, monitoring these data for temporal consistency will become an increasingly important problem. Long-term trends must be identified, and it must be determined if such trends correspond to true changes in reflectance characteristics of the study area (natural), or if their source is a signal collection and/or processing artifact that can be identified and corrected in the data (artificial). Spectrally invariant targets (SITs) are typically used for sensor calibration and data consistency checks. Unfortunately, such targets are not always available in study regions. The temporal averaging technique described in this research can be used to determine the presence of artificial interannual value drift in any region possessing multiyear regularly sampled time series remotely sensed imagery. Further, this approach is objective and does not require the prior identification of a SIT within the region of study. Using biweekly Advanced Very High Resolution Radiometer (AVHRR) normalized difference vegetation index (NDVI) data from 1990 to 2001 covering the conterminous United States, an interannual trend present in the entire scene was identified using the proposed technique and found to correspond extremely well with interannual trends identified using two SITs within the region.