The Application of Machine Learning Technique to Soil Salinity Mapping in South of Kazakhstan

In this paper, we consider the problem of assessing the salinity of the lands of the Turkestan region using remote sensing data. We aim to analyze the applicability of machine learning methods to evaluate the salinity of agricultural lands in southern Kazakhstan based on remote sensing. The machine learning algorithm uses Sentinel 1 radar data as features and Model A results expert assessment of soil salinity as output or target values for Gaussian Process training. The Gaussian Process model demonstrates a high degree of agreement with an expert estimation on the test subset of data (the recall, precision, and f1 metrics have a value of 0.89). The results allow us to recommend this approach for further testing based on ground-based measurement data and other machine learning methods for mapping the salinity of agricultural lands. The examined process of categorizing salinity has the potential to enhance the effectiveness of addressing challenges related to automating the digital mapping of salinity across extensive regions.