QFGeo-LULC Dataset

<p>The <strong>QFGeo-LULC Dataset</strong> is a geospatial dataset for Land Use and Land Cover (LULC) classification in the Quadrilátero Ferrífero region (Minas Gerais, Brazil), a landscape under intense anthropogenic pressure from mining activity. It was developed as part of a doctoral thesis at the Universidade Federal de Viçosa (UFV).</p> <p>The dataset is built from Sentinel-2 multispectral imagery (dry and rainy seasons), topographic attributes derived from the NASADEM Digital Elevation Model, spectral indices (NDVI, NDSI, NDBI, and EBBI), and geospatial embeddings obtained through Google Earth Engine. Samples were extracted from objects delimited by SLIC (Simple Linear Iterative Clustering) segmentation and produced using two complementary strategies: (i) manual collection validated by experts, with 1,396 samples across seven level-1 classes (Kappa agreement of 0.90 with MapBiomas), and (ii) automated collection from MapBiomas for the refinement of subclasses, with removal of inconsistencies (duplicate and edge samples).</p> <p><strong>Associated code repository:</strong></p><ul><li>Hierarchical classification and sample purity analysis: <a href="https://github.com/marianaschaefer/hierarch-classification">https://github.com/marianaschaefer/hierarch-classification</a></li> <p>If you use this dataset, please cite the associated doctoral thesis (Schaefer, 2026).</p>