Abstract: MODIS NDVI products from maximum value composite (MVC) still contain noise pixels which may affect application of the data for extraction of vegetation information. In order to efficiently remove the noise, it is suggested that high quality pixels representing the same ground object as the noise pixels do be located within a sliding window and means of the high quality pixels be used to replace values of the noise pixels in reconstruction of Savitzky-Golay(S-G) filter, and then the high quality pixels be kept in storage. Both this method and self-adaptive S-G filter can be used effectively to reconstruct 2001 – 2003 MODIS-NDVI chronological data of Jiangxi Province. Compared to the self-adaptive S-G filter reconstruction. the new method improves the correlation of the reconstruction with the high quality portion of the original data, and lowers the correlation of the reconstruction with the noise in the original data; and the reconstructed noise data get much closer to the means of high quality data and standard deviation. Besides, the new method can enhance the fidelity and stability of high quality pixels. Based on pixel quality analysis, S-G filter can reconstruct and yield better MODIS-NDVI dataset and extract more accurate vegetation coverage.  

Key words: MODIS, NDVI, S-G filter, time series