Machine Learning Models for Estimating Soil Salinity Using Sentinel-1 SAR and Landsat-8 OLI Data

Abstract

As Hungary has the largest expanse of naturally salt-affected soils in Europe with a continuous decrease in groundwater levels due to climate change, the expansion of saline soils to the detriment of arable lands has become a potential risk that requires continuous monitoring to sustain agricultural productivity and ensure food security. The study aims to estimate soil salinity in the Great Hungarian Plain, Eastern Hungary, using Sentinel-1 Synthetic Aperture Radar (SAR) C-band and Landsat-8 OLI data combined with three state-of-the-art machine learning models, i.e., Artificial Neural Network with feature extraction (PCANNET), Random Forest (RF) and Support Vector Machine (SVM). For this purpose, seventy-four soil samples provided by the Research Institute of Soil Sciences and Agricultural Chemistry (RISSAC) were collected in the Hungarian Soil Information and Monitoring System framework between September and October 2016. We compared the predictive performance of machine-learning-based models using the root mean square error (RMSE) and the correlation coefficient (r). The results revealed that the SVM-based model outperformed the other machine learning models with an RMSE equal to 0.24 g/kg and a correlation coefficient of 0.73. The study demonstrates the efficiency of machine learning techniques as valuable alternatives to estimate soil salinity and assist in land management planning with affordable costs.