Annual Land Use and Urban Land Cover: Ethiopia, Nigeria, and South Africa, 2016-2020

This dataset provides a two-tier annual Land Use (LU) and Urban Land Cover (LC) product suite over three African countries, Ethiopia, Nigeria, and South Africa, across a 5-year period of 2016-2020. Remote sensing data sources were used to create 30-m resolution LU maps (Tier-1), which were then utilized to delineate urban boundaries for 10-m resolution LC classes (Tier-2). Random Forest machine learning classifier models were trained on reference data for each tier and country (but one model was trained across all years); models were validated using a separate reference data set for each tier and country. Tier-1 LU maps were based on the 30-m Landsat time series, and Tier-2 urban LC maps were based on the 10-m Sentinel-2 time series. Additional data sources included climate, topography, night-time light, and soils. The overall map accuracy was 65-80% for Tier-1 maps and 60-80% for Tier-2 maps, depending on the year and country. The data are provided in cloud optimized GeoTIFF (COG) format.

本数据集提供了涵盖三个非洲国家——埃塞俄比亚、尼日利亚和南非——在2016至2020年五年期间的分层土地利用（LU）和城市土地覆盖（LC）产品套件。通过遥感数据源，构建了分辨率为30米的土地利用地图（一级），并以此为基础，对10米分辨率的土地覆盖类别（二级）进行城市边界划定。针对每一层级和每个国家（但有一模型跨越所有年份）的参考数据，对随机森林机器学习分类器模型进行了训练；模型验证则使用了每一层级和每个国家的独立参考数据集。一级土地利用地图基于30米分辨率的Landsat时间序列，二级城市土地覆盖地图基于10米分辨率的Sentinel-2时间序列。此外，还包括气候、地形、夜间灯光和土壤等附加数据源。整体地图精度为65-80%，适用于一级地图，而对于二级地图，精度为60-80%，具体取决于年份和国家。数据以云优化GeoTIFF（COG）格式提供。