Invisible people: Exploring how well remote-sensed datasets reveal the distribution of forest-proximate populations

Accurate information on the location and density of people living at the forest frontier is vital for effective and equitable forest conservation. We compare the location of settlements and estimated population density from three global-scale, remote-sensed datasets (World Settlement Footprint 2015, Open Buildings, WorldPop) with a fine-scale, manually-derived dataset of 3,136 human settlements, of which 95% had fewer than 150 households. The study region is located in north-eastern Madagascar, contains three protected areas and the largest unprotected block of humid forest of the island. The Open Buildings dataset detected a much higher proportion (94%) of settlements than did World Settlement Footprint (15%). Population density from WorldPop matches poorly with that estimated from our manually-derived dataset. The accuracy of all three datasets is worse in more remote, forested areas, further away from basic infrastructure. Open Buildings appears to best reveal the distribution of low density scattered populations in forested areas. However, further testing in other climatic regions is still needed. Making good use of appropriate remote-sensed data could revolutionize the inclusion of local communities in conservation policy and practice, improve the quality of inference in conservation research, particularly in times of a planned expansion of the global protected area network.