A global evaluation of harmonic analysis of time series under distinct gap conditions

Reconstruction of time series of satellite image data to obtain continuous, consistent and accurate data for downstream applications is playing a crucial role in remote sensing applications such as vegetation dynamics, land cover changes, land-atmosphere interactions and climate changes. Among the numerous methods and models developed to reconstruct time series of satellite observations in recent decades, Harmonic ANalysis of Time Series (HANTS) is one of the most widely used. Many studies based on time series reconstructed with HANTS documented the excellent performance of this method.
In the view of this study, the HANTS algorithm can be divided into two sub-processes, i.e., contaminated data identification and series reconstruction based on valid data. This study was dedicated to the evaluation of the performance of the latter sub-process. A simulated reference series dataset was constructed first, and then random gaps were introduced to these reference series. We built a look up table for distinct gap conditions by doing statistics on the deviation between the reference series and series reconstructed from gapped reference series. The look up table was used to evaluate the performance of a global NDVI time series dataset processed by HANTS.
The results show that the size of maximum gap (MGS), the number of loss (NL) and the number of gaps (NG) were significant factors in the reconstruction. When NDVI time series were rebuilt by HANTS, most of the region north than 40 degree N and mountainous areas of earth show bad reconstruction performance, that is, the root mean square deviation (RMSD) could exceed 0.25. This can be attributed to the periodical snow cover in these regions.