International interest in restoring trees to landscapes resulted in the 2011 Bonn Challenge and the creation of voluntary national restoration targets by many countries. The Bonn Challenge seeks to bring 150 million hectares into restoration by 2020, and 350 million hectares by 2030, and has thus far had 172 million hectares pledged. Such ambitious commitments to increase tree cover need to be assessed in light of prior progress towards increasing global tree cover. One such dataset, the Hansen global forest change product (HGFC) quantifies gain in tree cover from the year 2000 to 2012. However, the HGFC gain product maps tree cover, not natural forest cover, and thus includes agricultural tree crop monocultures (e.g., banana, oil palm, rubber) and timber plantations (e.g., pine, eucalyptus monocultures) in its estimates of gain. In this study, we ask three related questions. First, is it possible to use satellite imagery to distinguish tree plantations from natural regrowth at a global scale? Second, do tree plantation expansion and natural regrowth have distinct spatial patterns across continents and biomes? Third, how does the persistence of tree plantations over time compare to that of natural regrowth? To reclassify the HGFC gain dataset into two new land use classes (plantation or natural regrowth), we integrated data from three types of satellite imagery (Landsat, Alos PalSAR, and Sentinel 1) across a large, pan-tropical training dataset (n>800,000) with labeled land uses in 2015. We then used an ensemble machine learning approach to classify all HGFC gain footprints between 25° N and 25° S. We assessed accuracy by randomly sampling 3000 points across the tropics, and further sampling 2000 random gain polygons. At each selected polygon, we further assessed land cover change between 2015 and 2017 using high-resolution satellite imagery.
Results/Conclusions
We distinguished tree plantations from natural regowth with >90% overall accuracy. Tree plantations had larger mean patch sizes and were significantly more abundant near coastlines and rivers. Natural regrowth was most abundant in biomes with high remnant natural forest cover. Southeast Asia and South America were centers of plantation expansion. Finally, we found that, with the exception of short-rotation eucalyptus plantations, tree plantations had much lower disturbance rates than natural regrowth patches. The high observed rate of turnover of natural regrowth and the widespread expansion of tree plantations implies that lasting gains from restoration efforts may be largely limited to the establishment of tree plantations.