Following Burgess et al., 2014, Alesina et al., 2014 and Lubowski et al., 2014, we used the new high-resolution data Hansen et al. (2013) that identifies 1 arc second (about 30m at the equator) cells as forested or cleared for every year from 2001 to 2012. It also provide, for this 1 arc-second grid, with an estimate of the forest land cover in 2000 (coded as the percentage of the pixel size with vegetation is taller than 5meters in height), allowing to compute the deforestation rates and cover for each year. We first averaged the 2000 forest cover of every pixel of 1 arc-second on every country area. Contrarily to Lubowski et al. (2014), it is thus not a count model but a finer estimation averaging the cover of every 1 arc-second / 30 meters at the equator pixels. Given the way the data was computed from satellite imagery (Landsat 7 ETM+), we consider (1 arc-second) grid cells with more than 25% of forest cover as forests, weight the average of pixel deforested by their relative forest cover for each 30 arc-second resolution grid. Then we averaged the forest land cover on a reduced resolution grid of 30 arc-seconds (about 900m at the equator) in order to meet computing time constraints. Using this new resolution grid of 30 arc-second, we multiply the forest cover and deforestation ratios by the surface of each cell and sum every cells that have their barycenters inside a given country border, for a given year.
