Dataset of spatial products and supporting tables for the paper: "Mapping Kenya's grassland and rangeland extent using window-consistent machine learning and multi-sensor Earth observation"

Résumé fourni par la source

This dataset contains the final spatial products and supporting documentation for a national 2024 grassland/rangeland mapping study in Kenya. The products were generated using a window-consistent Random Forest classification framework that integrated manually interpreted 500 m reference image chips with multi-sensor Earth observation predictors from Sentinel-1, Sentinel-2, Dynamic World, ESA WorldCover and terrain datasets. The repository includes a 500 m continuous grassland/rangeland probability raster, the final binary grassland/rangeland extent map derived using probability threshold p ≥ 0.40 and a 5 km² minimum mapping unit, and three threshold-sensitivity products representing inclusive, minimally filtered and conservative interpretations of mapped extent. Supporting files include area-summary tables, threshold/MMU sensitivity tables, product metadata, raster inventory, data dictionary, product manifest, README documentation and MD5 checksums. The final selected binary product maps 110,277.5 km² of grassland/rangeland, equivalent to 18.8% of the valid mapped area of Kenya. The probability map is intended as the primary scientific product because it preserves uncertainty and ecological gradients, while the binary map is intended for national and regional reporting, rangeland assessment, restoration planning, biodiversity assessment, land degradation screening and land-resource applications. The dataset is not intended for parcel-level or field-boundary delineation without local validation.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.