Monitoring sub-pixel percent forest cover using multi-temporal NDVI

Normalized difference vegetation index (NDVI) time series have the potential to delineate the vegetation phenology change. For a mixed pixel, time series fluctuations representing phenological change in a year indicate percent cover changes. In the present study, MODIS NDVI time series were decomposed by discrete wavelet transformation (DWT) to remove noises. Principal component analysis (PCA) was implemented to generate new variables characterizing the variations of percent cover. The first PC and percent forestland cover are significantly correlated, with R²=0.5679. MODIS percent forestland cover can be roughly estimated by multi-temporal NDVI data in cloud-prone and rainy area.