Spectral Response Analysis and Land Cover Classification of Soil, Vegetation, and Water in a Semi-Arid Region Using Sentinel-2 MSI Surface Reflectance Data

The soil, vegetation and surface water of a landscape are in constant flux, and tracking that flux is central to ecological well-being and to the sustainable management of resources. This study presents a spectral analysis and land cover classification of those three surfaces from Sentinel-2 MSI Level-2A surface reflectance (BOA) imagery, in which radiometric calibration and atmospheric correction are already applied within the product itself. Spectral indices — NDVI, EVI, NDWI and related measures — were computed and passed to a KMeans clustering that resolved the scene into water, vegetation and soil; spectral thresholds then subdivided each category by soil moisture, vegetation health and water quality. Principal Component Analysis (PCA) isolated the bands and sources of variance that mattered most, PC1 and PC2 together accounting for 94.0% of the total variability. Validation was pursued internally through the Silhouette (0.612) and Calinski-Harabasz (132422.14) scores and externally through precision, recall, F1-score, Kappa and MCC. The mapping was dominated by Vegetation (53.63%), followed by Soil (42.48%) and Water (3.89%); within the sub-categories Moist Soil (52.33%) and Clay Soil (25.45%) prevailed, alongside distinct vegetation and water-quality signatures. External validation returned strong agreement between the final classification and the initial masks, with balanced performance across classes — Water Accuracy 0.983, Vegetation Accuracy 0.842 and Soil Accuracy 0.827 — confirming the strength of Sentinel-2 surface reflectance data for environmental monitoring and detailed condition assessment.