Long-term monthly NDVI calculated through daily images of AVHRR (NOAA/CDR/AVHRR/NDVI/V5) and MOD09GA.006 (MODIS/006/MOD09GA), which were cloud screened and aggregated by percentile 90th on Google Earth Engine-GEE. All the images were exported from GEE to R environment (MOD09GA data were downsampled to 5km using the average), where the follow procedures were executed on a pixel-basis: Outlier removal based on a temporal moving window, used to identify and remove NDVI outlier values (i.e. greater or less than 1.5 standard deviation calculated on 60-months time window) Smoothing and gap filling approach implemented by the greenbrown package (TSGFssa method) Temporal harmonization using MOD09GA NDVI as target variable to train a model [modis.ndvi ~ poly(avhrr.ndvi, 2) + max.temp + min.temp] able to predict the MODIS NDVI before 2000 The final product is composed by 456 NDVI global images without gaps, which were harmonized, from 1982 to 1999, to match with the expected MODIS NDVI values from 2000 onwards. For each month, an auxiliary image indicates which pixels were: Disregarded: value 0 (nodata) Gapfilled: value 1 smoothed: value 2 To access and visualize maps use: OpenLandMap.org Publication in preparation. All files were internally compressed using "COMPRESS=DEFLATE" creation option in GDAL. File naming convention (veg_ndvi_avhrr.mod09ga_p90_5km_s0..0cm_20000101..20000131_v1.0.tif): veg = Vegetation theme ndvi = Normalized Difference Vegetation Index avhrr.mod09ga = NOAA CDR AVHRR V5 and MOD09GA version 6 products p90 = Percentile 90th to aggregate the daily images 5km = 5 km of spatial resolution s0..0cm = land surface as vertical reference 20000101..20000131 = from 2000-01-01 to 2000-01-31 (time reference) v1.0 = version number