Global evaluation of gap-filling approaches for seasonal NDVI with considering vegetation growth trajectory, protection of key point, noise resistance and curve stability

A variety of approaches are available to fill the gaps in the time series of vegetation parameters estimated from satellite observations. In this paper, a scheme considering vegetation growth trajectory, protection of key point, noise resistance and curve stability was proposed to evaluate the gap-filling approaches. Six approaches for gap filling were globally evaluated pixel-by-pixel based on a reference NDVI generated from MODIS observations during the past 15 years. The evaluated approaches include the Fourier-based approach (Fourier), the double logistic model (DL), the iterative interpolation for data reconstruction (IDR), the Whittaker smoother (Whit), the Savitzky-Golay filter (SG) and the locally adjusted cubic spline capping approach (LACC). Considering the five aspects, the ranks of the overall performance are LACC > Fourier > IDR > DL > SG > Whit. The six approaches are similar in filling the gaps and remaining the curve stability but there are large difference in protection of key points and noise resistance. The SG is sensitive to noises and the Whit is poor in protection of key points. In the monsoon regions of India, all evaluated approaches don't work well. This paper provides some new views for evaluating the gap filling approaches that will be helpful in selecting the optimal approach to reconstruct the time series of parameters for data applications.