Plot-Based Classification of Macronutrient Levels in Oil Palm Trees with Landsat-8 Images and Machine Learning

Abstract

Oil palm crops are essential for ensuring sustainable edible oil production, in which production is highly dependent on fertilizer applications. Using Landsat-8 imageries, the feasibility of macronutrient level classification with Machine Learning (ML) was studied. Variable rates of compost and inorganic fertilizer were applied to experimental plots and the following nutrients were studied: nitrogen (N), phosphorus (P), potassium (K), magnesium (Mg) and calcium (Ca). By applying image filters, separability metrics, vegetation indices (VI) and feature selection, spectral features for each plot were acquired and used with ML models to classify macronutrient levels of palm stands from chemical foliar analysis of their 17th frond. The models were calibrated and validated with 30 repetitions, with the best mean overall accuracy reported for N and K at 79.7 ± 4.3% and 76.6 ± 4.1% respectively, while accuracies for P, Mg and Ca could not be accurately classified due to the limitations of the dataset used. The study highlighted the effectiveness of separability metrics in quantifying class separability, the importance of indices for N and K level classification, and the effects of filter and feature selection on model performance, as well as concluding RF or SVM models for excessive N and K level detection. Future improvements should focus on further model validation and the use of higher-resolution imaging.