Supplementary material to "The global forest above-ground biomass pool for 2010 estimated from high-resolution satellite observations"

The global forest above-ground biomass pool for 2010 estimated from high-resolution satellite observations A.1 Auxiliary datasetsThe European Space Agency (ESA) Climate Change Initiative Land Cover (CCI-LC) dataset consists of annual (1992-2018) maps classifying the world's land cover into 22 classes (Table S6).The overall accuracy of the 2010 land cover dataset was 76% (Defourny et al., 2014), with the most relevant commission and omission errors in mixed classes or in regions of strongly heterogeneous land cover.The land cover maps were provided in equiangular projection with a pixel size of 0.00278888° in both latitude and longitude.In this study, we used the land cover map of 2010, version 2.07.The dataset was reprojected to the map geometry of our AGB dataset.The Global Ecological Zones (GEZ) dataset produced by the Food and Agriculture Organization (FAO, 2001) divides the land surface into 20 zones (Figure S1, Table S2) with "broad yet relatively homogeneous natural vegetation formations, similar (but not necessarily identical) in physiognomy" (FAO, 2001).In this study, the dataset has been rasterized to the geometry of the images requiring stratification by ecological zones.Spatially explicit datasets of GSV representative of dense forests were obtained to support the model calibration described in Sections A.2 and A.3.This dataset was first compiled by assigning a value to the centre of each tile in a regular 2°×2° grid.Where available, in situ measurements from field plots or spatially explicit datasets of GSV were used.The GSV of dense forests was then defined as the 90 th percentile of the histogram within the 2°×2° area (Santoro et al., 2011).Elsewhere, it was estimated with an empirical piece-wise linear function (Santoro et al., 2015a) starting from values of the average biomass reported at provincial or national level.For tiles including several provinces or nations, the average biomass representative for the tile was obtained by weighting the individual averages by the area of each within the tile.In regions where values based on inventory measurements were unavailable but we could gather more than one map of AGB (preferably based on laser scanning observations), we