Validation and reference datasets for 2 m WorldView-CNN land-cover mapping, product evaluation, and change analysis in Amhara Region, Ethiopia

This data release provides validation and reference datasets used to support three related studies on land-cover and land-use (LCLU) mapping, product evaluation, and change analysis in Amhara Region, Ethiopia. The datasets were developed for assessment of 2 m WorldView-based convolutional neural network LCLU products and associated regional analyses. The release supports: (1) regional 2 m WorldView-CNN LCLU classification and accuracy assessment; (2) benchmarking of public medium-resolution LCLU products against validated WorldView-derived reference information; and (3) two-epoch LCLU change analysis using multiyear WorldView-CNN composites representing 2009–2016 and 2017–2024. The datasets include validation/reference point locations and labels, class definitions, field documentation, and supporting tabular outputs used for region-wide, product-comparison, terrain-stratified, and change-analysis summaries. The authoritative LCLU class label is val_class: 0 = Crop, 1 = Tree/Shrub, 2 = Grass, 3 = Built-up, and 4 = Water. Commercial WorldView and GeoEye source imagery is not included because it is restricted under the NGA NextView license and cannot be redistributed by the authors. Derived WorldView-based LCLU products are made available through the project Google Earth Engine application as described in the associated manuscripts.