Machine Learning-Oriented Forecasting Of Soil Degradation Due To Agricultural Land Use Patterns

Abstract

The degradation of soils is a significant threat to agricultural sustainability in the world, and it has been significantly contributed by the lack of sustainability in the land-use practices, including monocropping, excesses in the use of fertilisers and irrigation overuse. The study proposes a machine learning-based solution to predict soil degradation using agricultural land-use, soil health, and climatic conditions. The predictive capabilities of four machine learning algorithms were tested namely: The Random Forest (RF), Support Vectors Machine (SVM), Artificial Neural Networks (ANN) and Gradient Boosting (GB). The pre-processed and modeled data of the soil surveys, GIS dataset, and remote sensing images was used to predict the soil degradation indices in various agricultural areas. It was found that RF was the most accurate (93.4) and robust among diverse data sets with greater strength than GB (91.7), ANN (89. 2), and SVM (86.5). The models also showed superiority over similar studies with maximum prediction accuracy of 7 percent contrary to the regression-based models. Comparison reveals the relevance of ensemble approaches to the processing of noisy high-dimensional data, whereas ANN was found useful in the representation of complicated and non-linear soil-land interactions. These findings verify the position of machine learning as a credible tool of sustainable land administration. The paper is interdisciplinary research based on computational intelligence and agricultural sustainability that offers predictive information which can lead the policy makers, farmers and environmental planners to adopt proactive mechanisms in conserving their soils.