Trees outside forests in global drylands

Mitigating climate change while securing livelihoods is a major societal challenge in the 21st century. In this context, trees and shrubs (hereafter collectively referred to as trees) play a key role, as trees constitute a stable carbon sink while providing ecosystem services to the local population. Quantifying and monitoring the worlds trees is thus of highest importance, however, current assessments and statistics focus on trees within forests, overlooking trees outside forests. These non-forest trees are mainly located in drylands and include savannas and farmlands, but also deserts and other land with sparse vegetation. Most scientific and non-scientific attention is devoted to forests, largely ignoring trees that do not form a closed canopy, in spite several studies suggesting that trees outside forests exceed the resources provided by forests for numerous countries. The extent of non-forest trees has thus never been systematically assessed at large scale, so it has never been possible to study their impact on biochemical cycles as well as their economic or ecological importance at large scales. This work builds on two revolutionary developments impacting on the field of remote sensing which allow the mapping and monitoring of individual trees outside forests in great details at a continental scale, introducing variables that go beyond woody cover and opening new perspectives for research in the context of ecosystem services. First, the massive increase in available data with petabytes of high resolution images recorded by satellites every year. Second, deep learning has been the main driver of the tremendous progress in artificial intelligence over the last decade and has been shown to deliver excellent results in the field of object segmentation in high resolution satellite images. Here we introduce our ongoing work aiming at mapping trees in global drylands, as well as their carbon stocks using deep learning, very high resolution satellite/aerial imagery and field data. We show examples and results from different data sources (DigitalGlobe, PlanetScope, Gaofen, aerial photos) from different continents.