Soil depth classification using synchronous soil descriptors and machine learning for steep coffee plantation applications

Soil depth variability is a key factor influencing soil-machine interactions in steep agricultural environments. However, systematic approaches for soil depth classification based on soil physical-mechanical predictors remain limited, particularly for supporting mechanization in steep Arabica coffee plantations. This study aims to classify soil depth (5 cm and 10 cm) in Arabica coffee plantations in Aceh Province, Indonesia, by applying synchronous transformation of soil physical-mechanical predictors combined with machine learning algorithms. A total of 150 soil samples were collected from 25 sampling points, and 11 original predictors were derived from direct measurement and calculated parameters. These predictors were transformed synchronously into 66 features and processed using standard normal variate (SNV) before model development. Three machine learning algorithms, including principal component analysis (PCA), linear discriminant analysis (LDA), and k-nearest neighbor (KNN), were applied and optimized using 5-fold GridSearchCV. The results demonstrated that both LDA and KNN, applied to synchronous predictors, achieved perfect classification (accuracy = 1.00), outperforming PCA. This finding highlights the potential of synchronous transformation combined with machine learning as a reliable approach to soil depth classification, which can support the design of adaptive tractor wheel lugs for steep land coffee farming.