Integrating spatial database for predicting soil salinity using machine learning methods in Syrdarya Province, Uzbekistan

Abstract

Soil salinization of irrigated lands is a global problem in providing the necessary food and feed to meet the needs of a growing world population. Salinization in arid and semiarid areas can occur when the water table is three and more meters above the soil surface. Nowadays, innovative technologies are widely implemented in agriculture to increase yields and monitor changes in any area timely. Advanced technologies such as remote sensing (R.S.) data have become an economically efficient tool for assessing, detecting, mapping, and monitoring saline areas. This study aims to develop a spatial database for evaluating salinization using R.S. and GIS. This research employs various soil salinity indices based on Landsat 8 OLI images and other related geospatial datasets of the study areas. It aims to predict soil salinity using four machine learning methods (Gaussian Mixture Model (GMM), Random Forest (R.F.), Support Vector Machines (SVM), and K-Nearest Neighbors (KNN)). Results showed that R.F. is the most suitable for predicting the soil salinity in the study area with 93 percent overall accuracy. This research contributes to improving the quality of monitoring and improvement of the state of irrigated lands. Also, it develops a preliminary step toward decision-making tools for agricultural policies, such as managing saline areas related to crop production.