Detecting land cover change using an extended Kalman filter on MODIS NDVI time-series data

摘要

A method for detecting land cover change usingNDVI time series data derived from 500m MODIS satellitedata is proposed. The algorithm acts as a per pixel changealarm and takes as input the NDVI time series of a 3x3 gridof MODIS pixels. The NDVI time series for each of thesepixels was modeled as a triply (mean, phase and amplitude)modulated cosine function, and an extended Kalman Filter wasused to estimate the parameters of the modulated cosine functionthrough time. A spatial comparison between the center pixelof the the 3x3 grid and each of its neighboring pixel’s meanand amplitude parameter sequence was done to calculate achange metric which yields a change or no-change decision afterthresholding. Although the development of new settlements isthe most prevalent form of land cover change in South Africa,it is rarely mapped and known examples amounts to a limitednumber of changed MODIS pixels. Therefore simulated changedata was generated and used for preliminary optimization ofthe change detection method. After optimization the method wasevaluated on examples of known land cover change in the studyarea and experimental results indicate a 89% change detectionaccuracy, while a traditional annual NDVI differencing methodcould only achieve a 63% change detection accuracy.