A Comparison of Machine Learning Models for Soil Salinity Estimation Using Multi-Spectral Earth Observation Data

Abstract

Abstract. Soil salinity, a significant environmental indicator, is considered one of the leading causes of land degradation, especially in arid and semi-arid regions. In many cases, this major threat leads to loss of arable land, reduces crop productivity, groundwater resources loss, increases economic costs for soil management, and ultimately increases the probability of soil erosion. Monitoring soil salinity distribution and degree of salinity and mapping the electrical conductivity (EC) using remote sensing techniques are crucial for land use management. Salt-effected soil is a predominant phenomenon in the Eshtehard Salt Lake located in Alborz, Iran. In this study, the potential of Sentinel-2 imagery was investigated for mapping and monitoring soil salinity. According to the satellite's pass, different salt properties were measured for 197 soil samples in the field data study. Therefore several spectral features, such as satellite band reflectance, salinity indices, and vegetation indices, were extracted from Sentinel-2 imagery. To build an optimum machine learning regression model for soil salinity estimation, three different regression models, including Gradient Boost Machine (GBM), Extreme Gradient Boost (XGBoost), and Random Forest (RF), were used. The XGBoostmethod outperformed GBM and RF with the coefficient of determination (R2) more than 76%, Root Mean Square Error (RMSE) about 0.84 dS m−1, and Normalized Root Mean Square Error (NRMSE) about 0.33 dS m−1. The results demonstrated that the integration of remote sensing data, field data, and using an appropriate machine learning model could provide high-precision salinity maps to monitor soil salinity as an environmental problem.

Impacts of the Urmia Lake Drought on Soil Salinity and Degradation Risk: An Integrated Geoinformatics Analysis and Monitoring Approach

“…Earth observation satellite imagery is integral for soil salinity monitoring. The spectral reflectance characteristics of salt at the soil surface can be used to identify and monitor soil salinity [8]. Furthermore, well-organized soil management using the latest remote sensing methods and soil salinity monitoring plays an important role, especially in the preservation of arable lands [40].…”

“…We employed the eight best-known soil salinity indices to identify saline regions and compare their efficiency (Table 3). Soil salinity indices derived from the visible and near-infrared (NIR) bands of satellite images have been used in several studies to create soil salinity maps [5,8,19,[45][46][47][48]. However, comparing the efficiency of soil salinity indices remains a field of interest and further studies are required to introduce, compare and apply the efficient indices.…”

“…Different salt thresholds may have different reflectivity curves based on the physiochemical and environmental characteristics (e.g., texture, climate), which serve as the basis for selecting the most suitable estimation model. We applied eight indices recommended by earlier studies [1,8,17,19,50] and correlated the modeled results with the wavelengths obtained from sampling points to determine the soil salinity.…”

“…In arid and semiarid areas, the term 'soil degradation' refers to the process(es) by which soil quality declines, often due to improper use or poor management, and thus becomes less fit for a specific purpose such as crop production [6]. Soil salinization affects the physical, chemical, and biological soil processes through the salt-induced water deficits, ion toxicity, and nutritional imbalances, which ultimately lead to soil degradation [7,8]. Various environmental stressors (e.g., poor land use and management, pesticides, erosion), often impacted by the effects of climate change, drought, heavy and harmful metals, soil salinity, and flooding, can endanger agricultural lands and food production [9,10].…”

“…A review of the research literature indicated that earlier studies applied several methods for SMDM, such as the geostatistical visible and near-infrared spectroscopy index [7], multiple linear and random forest regression models [7], cubist and partial least square regression and electrical conductivity [34], ordinary kriging combined with back-propagation network [35], machine learning and particularly deep techniques [8,[35][36][37][38], the geographic weighted regression technique [39], deep neural network regression [38], and integrated fuzzy object-based image analysis [22]. In light of the global issues of soil salinization and soil degradation and under consideration of the recent advancements in earth observation technologies, including satellite images with improved spatial, spectral, and temporal resolution, and GIS methods (e.g., machine learning and advanced decision rules), there is a need to examine the efficiency of the different data-driven methods and decision rules for soil studies and, particularly, SMDM.…”