Henan Benchmark-Year Built-Up Area Rasters for 2015, 2018, 2020, and 2024, with Point-Level Random-Forest Scores, Validation Data, and Processing Scripts

This archive contains benchmark-year built-up area raster products for Henan Province, China, for 2015, 2018, 2020, and 2024; point-level random-forest scores used in the controlled threshold experiment; annual M2 fixed-threshold diagnostic results; CLCD-based external-reference and minimum-mapping-unit assessments; a stratified 150-point manual-reference assessment for 2020; selected training and validation samples; and the principal Google Earth Engine and Python scripts associated with the study. The released M4 rasters are archived operational products generated using year-specific configurations. The verified workflows differ in Sentinel-2 data level and preprocessing, random-forest output mode, probability or hard-class decision, nighttime-light candidate gating, minimum patch size, and morphological processing. The released rasters should therefore not be interpreted as a strictly homogeneous time series generated by one unchanged frozen classifier. Raster products are provided for the four benchmark years 2015, 2018, 2020, and 2024. Ten-year M2 diagnostic results and annual summary statistics are provided as tabular data where available. The 2020 stratified manual-reference assessment contains 150 interpreted reference points. Area-adjusted estimates are supplied together with sampling-stratum areas, the adjusted confusion matrix, built-up precision, recall, F1 score, adjusted built-up area, standard errors, and confidence intervals. Original Sentinel-2, VIIRS, CLCD, GHSL, and other third-party source products are not redistributed. Users should obtain those source datasets from their original providers under the applicable terms. Version 1.0.0. Zenodo DOI: 10.5281/zenodo.21835137.