Temporal Land Cover Classification and Change Deteciton Samples

This dataset provides a collection of high-quality training and validation samples used for long-term land cover classification and change detection in different area (mainly in Hunan Province, China (1990–2023)). The sample set was developed for the Multi-Task UNet for Temporal Classification and Change Detection (MTUTC) model, which integrates annual land cover classification and structured change detection from multi-decadal Landsat time series. The dataset contains manually interpreted and quality-controlled sample points across twelve representative regions covering six major land cover types: forest, cropland, grassland, water, bare land, and impervious surface. Each sample includes multi-temporal spectral features (Landsat reflectance bands), annotated land cover labels, and change/non-change indicators to support temporal modeling. The samples were generated using a combination of public reference data, high-resolution imagery, and expert visual interpretation to ensure temporal consistency and spatial representativeness. This dataset can be freely used for training deep learning models, benchmarking temporal land cover classification, validating change detection methods, or supporting studies on landscape dynamics, ecological monitoring, and carbon storage assessment.