Wadi Hadhramaut 2021 Agricultural Support-Rule Evaluation: A 250-Cell Reference Dataset and Validation Results (Version 1.0)

Dataset Description This dataset provides reference interpretation records and validation results for a remote-sensing assessment of agricultural land detection in Wadi Hadhramaut, Yemen, for the year 2021. The archive includes: Reference classifications for 250 sampled cells. Repeat interpretation records for 51 cells, including the predefined QA50 sample and one additional low-confidence cell. Researcher-approved adjudication records for 14 cases. A technically consolidated 250-cell reference dataset. Sampling information, evaluation weights, and 1,500 prediction records from six candidate classification rules. Recomputed accuracy statistics and supporting documentation for reproducibility. The evaluation demonstrates that high overall classification accuracy (approximately 95.46%) can coexist with low agricultural detection sensitivity (approximately 15.26%), highlighting the limitations of overall accuracy when agricultural areas are rare. Methodological limitations: The reference dataset underwent retrospective technical consolidation. Independent second-interpreter identity was not documented, and the dataset should not be described as independently verified field ground truth. The dataset is intended to support scientific transparency, methodological evaluation, and further research on agricultural remote sensing in hyper-arid environments.