Using Sentinel-2 Images and Unsupervised Classification to Characterize Saline Soil Areas in the Banat Plain

Abstract

Soils affected by salinization represent an important limitation for the sustainable agricultural use of land in lowland areas, where reduced natural drainage and a water table close to the surface favor the accumulation of soluble salts. The purpose of the research was to identify and spatially delimit areas affected by salinization in the Ciacova–Jebel–Voiteg area, Timiș County, using Sentinel-2 multispectral images, false color compositions, vegetation indices and unsupervised K-Means classification. The analyzed area was 1046584.70 m², or 104.66 ha. The classification separated ten spectral classes, of which class 5 was interpreted as being associated with areas with pronounced salinization, having 25660.76 m², or 2.45% of the total. The values ​​of NDVI, SAVI and MSAVI2 indices differed significantly between classes, according to the analysis of variance ANOVA (p<0.001). The results confirm the usefulness of integrating Sentinel-2 images with statistical analysis and digital classification for rapid mapping of salinities and for supporting agricultural land management decisions.