2.3. Time series MODIS NDVI dataMOD13Q1 products (vegetation indices 16-Day L3 Global 250 mversion 5) spanning the vegetation growing season from Marchto November were downloaded from the National Aeronauticsand Space Administration (NASA) of the United States (US) Ware-house Inventory Search Tool (WIST). These data were distributedby the Land Processes Distributed Active Archive Center (LPDAAC), located at the US Geological Survey (USGS) Earth ResourcesObservation and Science (EROS) Center (https://lpdaac.usgs.gov).MOD13Q1 data were provided every 16 days at a spatial resolu-tion of 250 m in the sinusoidal projection. It included red, NIR,blue and SWIR reflectance bands, NDVI and Enhanced Vegeta-tion Index (EVI) from the MODIS onboard of Terra platform. TheNDVI data was extracted for forest cover classification in thisstudy.The Savitzky–Golay (S-G) filter was used to smooth out noisein the time series MODIS NDVI data, specifically that caused pri-marily by cloud contamination and atmospheric variability (Chenet al., 2004; Savitzky and Golay, 1964). The algorithm made dataapproach the upper NDVI envelope and to reflect the NDVI pat-tern of change. It used a moving window, and noisy values wereapproximated by polynomial regression within the moving win-dows. An original NDVI profile and the S-G filtered NDVI profileat a randomly selected pixel was presented in Fig. 2. It was seenthat S-G filter could effectively eliminate data noise and improvethe quality of time series MODIS NDVI data. Projection of thesmoothed MODIS NDVI data was converted to the same projec-tion with Landsat ETM+ data. The spatial resolution of MODIS NDVIdata was resampled to 30 m, and the same columns and lineswere extracted to keep consist with Landsat ETM+ data for furtheranalysis