Mapping soil salinity using spectral mixture analysis of landsat 8 OLI images to identify factors influencing salinization in an arid region

Soil salinization is one of the most serious environmental issues degrading land resources globally, particularly in arid and semi-arid regions. Therefore, regional and precise monitoring of soil salinity is required to prevent and mitigate salinization. This study aimed to specify an effective monitoring method with remote sensing techniques using the Dakhla Oasis, central Western Desert of Egypt as a case study area. For ground-truthing, electrical conductivity, pH, reflectance spectra, and mineral compositions were measured for top soil samples from 31 points. Spectral data from a Landsat8 OLI image of one scene acquired close to the ground sampling time was used to estimate soil salinity using a variety of methods, including single band, band ratio and combination, spectral index, linear spectral unmixing (LSU), and mixture tuned matched filtering (MTMF). After estimating the salinity over the study area through the best regression model between the spectral data and measured salinity data, the image was classified into five salinity classes. The classified salinized zones were verified by the resistivity and thickness of the near-surface layers and depth to the groundwater table, using vertical electrical sounding (VES) at 46 profiles. The most salinized zones in the southern area were congruent with the lowest VES resistivity. The surface layer thickness and clay content were specified as the main cause of the salinity difference between the southern and northern areas. The land surface temperature (LST) retrieved from the thermal band data of the OLI image and another Landsat ETM + image in 2001 was identified as increasing salinization. Finally, urban and vegetation land covers along with the five soil salinity classes were characterized by the influencing factors of elevation, slope, LST, soil pH, top layer resistivity and thickness, and depth to the groundwater table. LSU proved to predict salinity more accurately with 76% correctness than the MTMF model (67%) and the band combination and spectral indices (55% at most). The proposed methods will be useful for soil salinity mapping from satellite imagery in similar environments to this study.