High-resolution global map of closed-canopy coconut

Abstract. Vegetable oil crops cover over half of global agricultural land and have varying environmental and socioeconomic impacts. Demand for coconut oil is expected to rise, but the global distribution of coconut is understudied, which hinders the discussion of its impacts. Here, we present the first 20-meter global coconut layer, produced using deep learning for semantic segmentation, specifically a U-Net model that was trained using annual Sentinel-1 and Sentinel-2 composites from 2020. Results confirmed the feasibility of using Sentinel-1 for mapping palm species that present full canopy closure. The overall accuracy was 99.10 ± 0.20 %, which was significantly higher than the no-information rate. The producer’s accuracy was 72.07 ± 22.83 % when only closed-canopy coconut was considered in the validation, but decreased to 12.34 ± 2.60 % when sparse and open-canopy coconut areas were considered, indicating that this planting context remains difficult to map with accuracy. We report a global coconut area of 12.31 ± 3.83 x 106 ha for dense open- and closed-canopy coconut, but the estimate is three times larger (36.72 ± 7.62 x 106 ha) when sparse coconut is included in the area estimation. This large area of sparse and dense open-canopy coconut is important as it indicates that production increases can likely be achieved on the existing lands allocated to coconut. The Philippines, Indonesia, and India account for most of the global coconut area, or about 82 % of the total mapped area. Our study provides the high-resolution, quantitative, and precise data necessary for assessing the relationships between vegetable oil production and the synergies and trade-offs between various sustainable development goal indicators. The global coconut layer is available at https://doi.org/10.5281/zenodo.7453178 (Descals, 2022).