Replication archive for "Cross-validation optimism in product-seeded Sentinel-2 change mapping: a map-blind reference audit in semi-arid Ajmer"

Replication archive for the study "Cross-validation optimism in product-seeded Sentinel-2 change mapping: a map-blind reference audit in semi-arid Ajmer" (Journal of the Indian Society of Remote Sensing). It contains the analysis pipeline (Python scripts for feature stacking, product-seeded training-label construction, Random Forest/SVM/XGBoost classification, buffered spatial cross-validation, paired bi-temporal reference sampling, and design-based accuracy and area-change estimation), the classified rasters (Random Forest and XGBoost land-cover maps, built-up change, WorldCover seed and valid masks as NumPy arrays on a 3000x3000 10 m grid, EPSG:32643), the probability samples and blinded reference interpretation, computed statistics, an independent Google Earth Engine cross-check (GHSL, Dynamic World, Open Buildings), and a README with the reproduction order. Raw Sentinel-2 imagery is not redistributed; scene IDs and public asset URLs are listed in data_manifest.json.