부산대학교 도서관

Error inherent in satellite remote sensing data due to the imaging process needs to be reduced to allow accurate measurement of vegetation phenology patterns. Compositing - and the selection of images and spectral data - is an important process to this end. Here, the success of cloud and viewing geometry masking algorithms employed to remove imaging error in the MODIS 250m 16-day vegetation index product is investigated. Green-up dates estimated using these data and those estimated by removing cloud-affected pixels manually are compared with much finer resolution Sentinel-210m and Planetscope 3m imagery for 208 sample canopies in four major vegetation types across a diverse 400km 2 savanna landscape in south-eastern Zimbabwe. RMSE was reduced by 10 days and the identification of spurious early green-up largely avoided by removing cloud-affected pixels by eye. Treating data from the Aqua and Terra platforms separately also reduced spurious early green-up dates.