Comparison of three time-series NDVI reconstruction methods based on TIMESAT

Abstract

This paper has presented the reconstructing performances of AG, D-L and S-G methods for different types of vegetation cover and quantitatively comparing residuals and RMSE between raw and reconstructing NDVI. Time-series NDVI curves of six vegetation cover types of desert, desert steppe, meadow, shrub, crop, and forest. The results show that AG and D-L have better performances in keeping overall shape of curves. However, D-L may generate slightly more deviations in fitting the peak points of growing season curve than AG in general. The highest correlation coefficients of reconstructing curve with raw curve are of AG, followed by D-L, S-G. The AG method reconstructed the NDVI curves slightly better than D-L and better than S-G, particularly for greenup and dormancy period of growing season (RMSE (AG) < RMSE (D-L) < RMSE(S-G)). However, in peak of curves, the RMSE value of S-G was lower than that of other two methods. © 2012 IEEE.