Comparison on Three Algorithms of Reconstructing Time-series MODIS EVI

With the rapid development of remote sensing techniques, higher precisions of the vegetation remote sensing are required. Therefore, before using the time-series data, how to select the optimal algorithms to reconstruct it has been a hot research topic. Based on the five main land cover types in Northeast China, the reconstruction quality of three commonly used algorithms that included in TIMESAT tools has been qualitatively analyzed.Then, the fidelity performance and the capability to keep main characteristics of the three algorithms on EVI with respect to different land cover types were compared. The result shows that the S- G algorithm has a better performance in reconstructing the peak and the width of the EVI curves in the growing seasons, but it is prone to keep the noise data due to excessive fittings, especially common in land cover types of steppe and shrub. AG and DL algorithms generally present similar performances and the results are much closer to the true values for land cover types of steppe, shrub and arable land. But AG algorithm is easily influenced by noises for fitting the peak of the cures, which reduces the maximum EVI and causes the decline of vegetation growth. Spatial patterns of the fidelity performance and the capability to keep main characteristics of the three algorithms are all related to the distribution of vegetation types. Finally, we found that AG is a better algorithm to be used for the land cover types of steppe and shrub, DL is better for arable land, while S-G is better for the broadleaved deciduous forest and coniferous deciduous forest.