The file ‘GlobalCoconutLayer_2020_v1-2.zip’ contains 878 raster tiles of 100x100 km in geotiff format. The raster files are the result of a convolutional neural network that classified Sentinel-1 and Sentinel-2 annual composites into a coconut palm layer for the year 2020. The images have a spatial resolution of 20 meters and contain two classes: [0] Other land covers that are not coconut palm. [1] Coconut palm. The file ‘GlobalCoconutLayer_2020_densityMap_1km_v1-2.zip’ contains the 20-meter coconut palm classification aggregated to 1 km. The value of each pixel represents the coconut palm area (in squared meters) within the 1-km pixel. The file ‘Validation_points_GlobalCoconutLayer_2020_v1-2.shp’ includes the 10,200 points that were used to validate the product. Each point includes the attribute ‘Class’, which is the class assigned by visual interpretation of sub-meter resolution images, and the attribute ‘predClass’, which reflects the predicted class by the convolutional neural network. The ‘predClass’ values are the same as the raster files: [0] Other land covers that are not coconut palm. [1] Coconut palm. The attribute ‘Class’ contains the following values: [0] Land cover could not be determined because sub-meter resolution data was not available. [1] Other land covers that are not coconut palm. [2] Sparse coconut palm. Low density of coconut palms; between 1 and 4 coconut palms within the 20-meter pixel. [3] Dense open-canopy coconut palm; more than 4 coconut palms within the 20-meter pixel but coconut trees do not reach the full canopy closure. [4] Closed -canopy coconut palm; more than 4 coconut palms within the 20-meter pixel and coconut palms fully cover the ground. [5] Palm species that are not coconut palm. Changelog v1-2: - Pixels classified as class ‘coconut’ were reclassified to class ‘other’ in West Bengal.