Reconstruction of global MODIS NDVI time series: Performance of Harmonic ANalysis of Time Series (HANTS)

The harmonic analysis of time series (HANTS) algorithm has been widely used to reconstruct time series of normalized difference vegetation index (NDVI), Leaf area index (LAI), and land surface temperature (LST) as well as the polarization difference brightness temperature (PDBT) during the past 20years to remove random noise or eliminate cloud/snow contamination. So far no systematic study on the accuracy of such reconstruction has been done. This study aims at taking the global MODIS vegetation index as an example to develop a generic method to evaluate the reconstruction performance of HANTS. The overall reconstruction error was divided into gap related error and fitting-method related error. Firstly, ten annual NDVI time series for a pixel were used to extract reference series and gap statistics. Then the gap and fitting-method related errors were quantified independently. The results suggest that the gap related error for most of the high latitude forest area (between 50°N and 70°N) was rather large (the mean root mean squared deviation (RMSD) reached to 0.15), which may be attributed to the fact that large gaps appear in the NDVI profiles between snow melting and vegetation regreening season. The gap related error was found negligible for the other areas of the globe except the North China Plain, the North India and several mountainous areas where the mean RMSD is around 0.1. The inadequate capability of low frequency harmonics to capture the rapid transition during snowmelt in spring at the high latitude region of the North Hemisphere makes the fitting-method related error in this region rather large (RMSD can reach 0.1). The method developed in this study was applied to map globally the spatial pattern of HANTS performance in the reconstruction of NDVI time series and it can also be applied to evaluate the reconstruction performance of time series of other land surface variables or the performance of other time series reconstruction algorithms.