Multi-Nutrient Mapping in Oil Palm Using Sentinel-2 and Random Forest: A Cost-Efficient Approach for Precision Agriculture Wiratmoko, D. (a*) and Diana, S.R.(b)
a) Indonesian Oil Palm Research Institute, Medan, Indonesia
*dhimaswiratmoko[at]iopri.org and wiratmoko2nd[at]gmail.com
b) Research Organisation for Governance, Economy, and Public Welfare, PRKP, National Research and Innovation Agency (BRIN), Indonesia
Abstract
Efficient and accurate assessment of leaf nutrient content is essential for optimizing fertilizer use and ensuring sustainable oil palm production. Conventional methods that rely on extensive field sampling and laboratory analysis are costly, labor-intensive, and have limited spatial and temporal coverage. Advances in remote sensing and cloud-based analytics have offered opportunities for efficient, scalable, and timely nutrient mapping. However, research integrating multi-nutrient prediction in oil palms using satellite imagery and machine learning, particularly within the Google Earth Engine (GEE) platform, remains scarce. This study developed and evaluated a cost-efficient method to map multiple essential leaf nutrients, including nitrogen (N), phosphorus (P), potassium (K), and magnesium (Mg), using Sentinel-2 multispectral imagery and a Random Forest (RF) algorithm in GEE. Field sampling was conducted in oil palm plantations in North Sumatra, Indonesia, and nutrient concentrations were determined by laboratory analysis. Sentinel-2 spectral features, including vegetation indices and reflectance bands, were extracted as predictor variables for RF classification. Model performance was evaluated using the overall accuracy, kappa coefficient, producer accuracy, and user accuracy. The results showed high classification performance, with overall accuracies of 91.13% (N), 91.60% (P), 91.48% (K), and 92.18% (Mg) and kappa values between 0.873 and 0.917. Producer accuracies exceeded 90% for all nutrients, indicating reliable detection, while user accuracies were consistently above 89%, confirming classification stability. Compared with traditional approaches, this method can reduce operational costs by up to 60% and significantly shorten the processing time, enabling large-scale and frequent nutrient monitoring. By linking nutrient mapping to precision agriculture, the proposed approach supports site-specific fertilizer recommendations, minimizes environmental risks from over-fertilization, and enhances plantation management decision-making. The integration of Sentinel-2 imagery, RF modelling, and cloud-based processing is a practical, scalable, and economically viable solution for sustainable oil palm cultivation.
