Time series of NASA HLS for coffee and land use mapping

This dataset was developed to assess the performance of dense time series derived from spectral bands, vegetation indices, and texture metrics of HLS (Harmonized Landsat and Sentinel-2) imagery, combined with Random Forest and XGBoost algorithms under an ensemble learning framework, for land-use and coffee production stage mapping. The data are structured across a hierarchical classification approach with increasing thematic detail: Level 1: Discrimination between natural vegetation and anthropogenic land use. Level 2: Classification of anthropic vegetation into pasture, annual crops, and perennial crops. Level 3: Differentiation between coffee plantations and planted forestry (silviculture) within perennial crops. Level 4: Mapping of four coffee production stages — planted (PL), producing (PR), skeletonized (ES), and stumped-renewed (RN) plantations. The dataset includes linear gap-filled multiband time series (Blue, Green, Red, NIR, SWIR, GNDVI, NDVI, NDWI, and SAVI) and GLCM-based texture metrics from 2023, organized into georeferenced samples collected during a field campaign in Caconde, São Paulo, Brazil, in October 2023. Each subset is labeled according to its respective classification level. These data support the study “Dense Time Series of Harmonized Landsat Sentinel and Ensemble Machine Learning to Map Coffee Production Stages”, accepted for publication in Remote Sensing - MDPI.