Detection of contaminated pixels based on the short-term continuity of NDVI and correction using spatio-temporal continuity

The present study developed and assessed a correction technique (CSaTC: Correction based on Spatial and Temporal Continuity) for the detection and correction of contaminated Normalized Difference Vegetation Index (NDVI) time series data. Global Inventory Modeling and Mapping Studies (GIMMS) NDVI data from 1982 to 2006 with a 15-day period and an 8-km spatial resolution was used. CSaTC utilizes short-term continuity of vegetation to detect contaminated pixels, and then, corrects the detected pixels using the spatio-temporal continuity of vegetation. CSaTC was applied to the NDVI data over the East Asian region, which exhibits diverse seasonal and interannual variations in vegetation activities. The correction skill of CSaTC was compared to two previously applied methods, IDR (iterative Interpolation for Data Reconstruction) and Park et al. (2011) using GIMMS NDVI data. CSaTC reasonably resolved the overcorrection and spreading phenomenon caused by excessive correction of Park et al. (2011). The validation using the simulated NDVI time series data showed that CSaTC shows a systematically better correction skill in bias and RMSE irrespective of phenology types of vegetation and noise levels. In general, CSaTC showed a good recovery of the contaminated data appearing over the short-term period on a level similar to that obtained using the IDR technique. In addition, it captured the multi-peak of NDVI, and the germination and defoliating patterns more accurately than that by IDR, which overly compensates for seasons with a high temporal variation and where NDVI data exhibit multi-peaks.