High-resolution global map of smallholder and industrial closed-canopy oil palm plantations

Abstract. Oil seed crops, especially oil palm, are among the most rapidly expanding agricultural land uses, and their expansion is known to cause significant environmental damage. Accordingly, these crops often feature in public policy debates which hampered or biased by a lack of accurate information on impacts. In particular, global crop maps remains concern. Recent advances deep-learning remotely sensed data access make it possible address this gap. We present map closed-canopy palm (Elaeis guineensis) plantations typology (industrial versus smallholder plantations) at scale with unprecedented detail (10 m resolution) for year 2019. The DeepLabv3+ model, convolutional neural network (CNN) semantic segmentation, was trained classify Sentinel-1 Sentinel-2 images onto an cover map. characteristic backscatter response stands ability CNN learn spatial patterns, such as harvest road networks, allowed distinction between industrial globally (overall accuracy =98.52±0.20 %), outperforming existing regional datasets that used conventional machine-learning algorithms. user's accuracy, reflecting commission error, smallholders 88.22 ± 2.73 % 76.56 4.53 %, producer's omission 75.78 3.55 86.92 5.12 respectively. layer reveals found 49 countries, covering mapped area 19.60 Mha; estimate 21.00 0.42 Mha (72.7 27.3 plantations). Southeast Asia ranks main producing region 18.69 0.33 89 plantations. Our analysis confirms variation ratio growers, but also that, from typical development perspective, large areas legally defined resemble industrial-scale plantings. Since our study identified only stands, lower than harvested reported Food Agriculture Organization (FAO), particularly West Africa, due young sparse nonhomogeneous settings, semi-wild An planted can help shape ongoing debate about impacts expansion, if other be same level accuracy. As model regularly rerun new become available, monitor monocultural settings. second half 2019 resolution 10 https://doi.org/10.5281/zenodo.4473715 (Descals et al., 2021).