Automatic Mapping of 10 m Tropical Evergreen Forest Cover in Central African Republic with Sentinel-2 Dynamic World Dataset

This dataset provides annual 10-m resolution maps of tropical evergreen forest cover in the Central African Republic (CAR) for the period 2017–2023. The Central African Republic hosts some of the most biodiversity-rich and conservation-critical tropical evergreen forests on Earth, making fine-scale, up-to-date mapping essential for ecological research, biodiversity conservation, and land management. The maps were generated using a novel automated method that leverages the Sentinel-2 Dynamic World dataset. Stable evergreen forest samples were first identified using the Copernicus Global Land Cover Layers (CGLC) and Global Forest Change (GFC) products. Initial evergreen forest cover maps were then produced by applying year-specific thresholds to the annual median forest cover probability layers derived from Dynamic World. Final refined maps were obtained by integrating a Non-Evergreen Forest Index (NEFI) with the thresholded initial maps. Independent accuracy assessment shows that the maps achieve an overall accuracy >94.10%, Cohen’s Kappa >87.63%, and F1-score >94.05% across all years, substantially outperforming existing products such as the CGLC evergreen forest layers and simple median probability maps from Dynamic World. These annual 10-m maps offer detailed spatial representation of evergreen forest extent and consistent temporal tracking of changes from 2017 to 2023 in the Central African Republic, providing a valuable resource for studying tropical forest dynamics, monitoring deforestation and degradation, and supporting conservation planning in this ecologically sensitive region.本数据集提供了2017至2023年间，中非共和国（Central African Republic，CAR）境内热带常绿森林覆盖范围的年度10米分辨率地图。中非共和国拥有全球生物多样性最为丰富、生态保护优先级最高的热带常绿森林之一，因此高精度、现势性强的森林覆盖制图对于生态学研究、生物多样性保护以及土地管理而言至关重要。该套地图采用一种创新性自动化方法生成，该方法依托哨兵-2（Sentinel-2）动态世界（Dynamic World）数据集构建。研究人员首先借助哥白尼全球土地覆盖图层（Copernicus Global Land Cover Layers，CGLC）与全球森林变化（Global Forest Change，GFC）产品，识别出稳定常绿森林样本。随后，针对由动态世界数据集导出的年度森林覆盖概率中值图层，通过应用针对各年份设定的阈值，生成初始常绿森林覆盖地图。最终通过将非常绿森林指数（Non-Evergreen Forest Index，NEFI）与经过阈值处理的初始地图进行融合，得到经优化的最终地图。独立精度评估结果显示，该套地图在所有研究年份中总体精度均高于94.10%，科恩卡帕系数（Cohen’s Kappa）高于87.63%，F1分数（F1-score）高于94.05%，其性能显著优于现有产品，如哥白尼全球土地覆盖图层的常绿森林图层，以及动态世界数据集生成的简单中值概率地图。这套年度10米分辨率地图能够细致展现中非共和国境内常绿森林的空间分布范围，并实现2017至2023年间森林变化的连贯时序追踪，可为该生态敏感区域的热带森林动态研究、森林砍伐与退化监测以及保护规划制定提供宝贵的数据资源。