Evaluating Multi-sensor Combination of Normalized Difference Vegetation Index (NDVI) Time Series Data over Southeast Asia

Normalized Difference Vegetation Index (NDVI) data is the most commonly used vegetation proxy from remote sensing data to model the vegetation biophysical properties. The longest time-series data of NDVI from the earlier era of remote sensing satellites is available from AVHRR GIMMS employing the red and near-infrared bands in NOAA sensors from 1981 to 2015 in 8-km spatial resolution in the monthly interval. This study aims to evaluate the compatibility of NDVI data from the newer sensors such as MODIS Terra (MOD13C2), Proba-V and Visible Infrared Imaging Radiometer Suite (VIIRS) data when combined with GIMMS data. Calibration between two time-series data from different sensors was constructed by using image-matching Pseudo Invariant Features (PIF) method and the fitness levels using all pixels and at different land-cover classes were assessed. In addition, structural change analysis was conducted to identify the sensor-shift problems at the best data combination. Our results suggested the best fit of GIMMS when being paired with VIIRS data with the R2 of 0.91 (n = 3132) and 0.89 (n = 1044) for model and validation analysis. Although the fitness level from the linear regression showed a good fit, an artifact as a result of sensor-shift problems still can be detected from structural change analysis, revealing the imperfection of linear calibration method. Future works should aim to explore the performance of non-linear methods to calibrate the different time-series data and explore the combination with other sensors.