Annual tree cover data from 2001-2020 across the Eastern Guinean Forest in Ghana and Nigerian Lowlands Forest in Nigeria

We combined machine learning and LandTrendr temporal segmentation to generate annual tree cover from 2001-2020 from the Landsat archive for Eastern Guinean Forest in Ghana and Nigerian Lowlands Forest in Nigeria. The continuous tree canopy cover generated using only the machine‑learning model exhibited larger errors than the temporally segmented tree canopy cover estimates derived from LandTrendr. The LandTrendr‑based approach produced higher predicted–observed, lower mean absolute error, and lower root mean squared error, indicating that the processed time‑series segmentation more effectively estimates areas with high versus low tree cover across Eastern Guinean Forest and Nigerian Lowlands Forest. Using a 30% threshold, the classified tree cover achieved a producer accuracy of 82% and a user accuracy of 83%. The data is available as continuous tree canopy cover (CanopyCover_Continuous.zip) and binary tree cover (TreeCover_Classified.zip) in GeoTIFF format.我们将机器学习与LandTrendr时序分割算法相结合，基于陆地卫星（Landsat）存档数据，为加纳东部几内亚森林与尼日利亚低地森林生成了2001至2020年的年度林木覆盖数据集。仅通过机器学习模型生成的连续林冠覆盖数据，其误差高于基于LandTrendr时序分割得到的林冠覆盖估算结果。基于LandTrendr的方法可获得更高的预测-观测匹配度、更低的平均绝对误差（mean absolute error, MAE）与更低的均方根误差（root mean squared error, RMSE），这表明经处理的时序分割方法能够更有效地估算加纳东部几内亚森林与尼日利亚低地森林内林木覆盖度高低各异的区域。采用30%阈值进行分类后，林木覆盖分类数据的生产者精度（producer accuracy）达82%，使用者精度（user accuracy）达83%。本数据集以GeoTIFF格式提供，包含连续林冠覆盖数据（压缩包名称：CanopyCover_Continuous.zip）与二值化林木覆盖分类数据（压缩包名称：TreeCover_Classified.zip）。