Assessment of Salinization Through ANN Learned with Remote Sensing and DEM Data in Soils of the Lower Cheliff Plain (Algeria)

One of the constraints for the assessment of salinization is the scarcity of reliable methodologies for accurately quantifying and mapping soil salinity expressed in values of electrical conductivity (EC). In this work, we have estimated soil salinity and assessed the role of topography by applying artificial neural networks (ANN). For that, remote sensing data (Landsat-8 OLI), topography data (ASTER GDEM) and soil salinity values (n = 796) were utilized to perform some ANN capable of quantifying soil salinity whose 80% of these data were intended for ANN learning and 20% for results' validation. Soil salinity was estimated through an ANN based on remote sensing data of three bands (B3, B4, B5) with a high reliability (R = 0.89). The established soil salinity map basing those results showed 23.8% of soil was affected by salinization in the last decade, and confirmed the advanced state of land degradation. The application of an ANN allowed to estimate the correlation between soil salinity values and elevation and slope data (R = 0.74). This correlation improved with the addition of bands data in another ANN with R = 0.77.

Automated summary

In this study, soil salinity was estimated using artificial neural networks and the role of topography was assessed. A reliable soil salinity map based on remote sensing data and soil salinity values was generated, showing the extent of land degradation.