Estimating Soil Quality Indicators Using Remote Sensing Data: An Application of Machine Learning Regression Models

Intensive agriculture has resulted in a decline in soil quality, presenting significant challenges in terms of increasing agricultural productivity while maintaining environmental sustainability. Consequently, farmers require methodologies that integrate low-cost technologies for data collection and processing with traditional soil quality assessment tools to estimate soil quality indicators (SQIs). This study presents an application of machine learning regression models to estimate soil quality in local-scale agricultural systems through the processing of a georeferenced-multidimensional database. Additionally, the impact of eight standardization methods on the performance of the machine learning models is investigated. The regression analysis results for the analyzed SQIs demonstrate satisfactory performance based on two metrics: negative mean square error and r2. These findings contribute to the establishment of an estimation model for SQIs using machine learning algorithms. Notably, indicators directly linked to chemical fertilizers, such as nitrogen, potassium, and phosphorus, exhibit performance levels of 70%, 71%, and 92%, respectively.