MONITORING AND PREDICTING SOIL SALINITY LEVEL USING REMOTE SENSING SENTINEL-2A IMAGE AND MACHINE LEARNING MODELS: A CASE STUDY OF RWANDA

Abstract

Soil salinity significantly threatens agricultural productivity and food security, particularly in regions with limited agricultural land areas like Rwanda. Induced activities can increase soil salinity due to continuous intensive agricultural activities dominated by irrigation practices and increased agricultural input use. In Rwanda, where agriculture is a vital sector of the economy, monitoring and predicting soil salinity levels are crucial for sustainable land management and crop production. This study explores the feasibility of using remote sensing Sentinel-2A satellite images and machine learning algorithms to monitor and predict soil salinity in Rwanda. We utilised Sentinel-2A multispectral satellite imagery, which offers high spatial resolution and frequent revisits, allowing us to detect subtle changes in land surface characteristics. The Normalized Difference Salinity Index (NDSI), known for its good correlation with soil salinity levels, was calculated from the satellite images to quantify soil salinity across different regions of Rwanda by comparing with available observed EC soil salinity levels from 500 soil samples from agricultural areas taken across the country. Machine learning algorithms, including Random Forest, Support Vector Regression, and Multiple Linear Regression, were employed to develop predictive models based on Satellite Salinity Indices. These models were trained using historical soil salinity data from field measurements and laboratory analyses. The random forest regression model was the best predictor and accurate correlation. Among salinity indices, NDVI and SR are good in correlation and relative importance. The machine learning models demonstrate promising performance in predicting soil salinity across various landscapes and land use types in Rwanda. Integrating remote sensing data and machine learning techniques provides a cost-effective and scalable approach for monitoring soil salinity over large spatial extents and temporal scales. The findings of this study contribute to advancing our understanding of soil salinity dynamics in Rwanda. They can inform targeted interventions for soil salinity management and agricultural development in the region.