Muhammad Isra, Laode Muhamad Irsan, Sakinah Ahyani Dahlan, Widya Rahmawaty Saman, Yoyanda Bait
Accurate monitoring of nitrogen status is essential for improving maize productivity while reducing the environmental impacts of excessive fertilization. This study evaluated the potential of Sentinel-2 satellite imagery combined with machine learning techniques to estimate leaf nitrogen content in a tropical maize agroecosystem in Gorontalo, Indonesia. Field-measured leaf nitrogen data were obtained through laboratory analysis and linked with spectral information derived from Sentinel-2 vegetation indices, including the Normalized Difference Vegetation Index (NDVI), Soil-Adjusted Vegetation Index (SAVI), and Visible Atmospherically Resistant Index (VARI). A total of 50 leaf samples were collected across five georeferenced sampling sites during the vegetative stage of maize growth and used for model development and validation. Regression analysis and machine learning approaches were applied to model the relationship between vegetation indices and leaf nitrogen content. NDVI showed the strongest individual relationship with nitrogen concentration (R² = 0.82), while multiple linear regression improved predictive performance (R² = 0.87). Among the tested models, Random Forest achieved the best overall performance, with R² = 0.88, RMSE = 0.18, and MAE = 0.12, outperforming both linear regression and Support Vector Regression. Spatial mapping derived from the best-performing model revealed substantial variability in nitrogen content across the maize field, ranging from 2.60% to 3.40%. These findings demonstrate the potential of Sentinel-2-based monitoring to support site-specific nitrogen assessment and more targeted fertilizer management under tropical field conditions. Overall, the integration of Sentinel-2 vegetation indices and Random Forest modeling provides a reliable and non-destructive approach for field-scale nitrogen estimation in maize. Further work incorporating broader spatial and temporal observations, as well as multi-sensor data integration, would likely improve model robustness and wider applicability. © 2026 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license. http://creativecommons.org/licenses/by/4.0/
Food Technology Study Program, Faculty of Agriculture, Universitas Negeri Gorontalo, Gorontalo, 96554, Indonesia; Agribusiness Study Program of Agriculture, Universitas Negeri Gorontalo, Gorontalo, 96554, Indonesia
Research at a Glance
Register to unlock
Topics & SDG Alignment
Register to unlock
Collaboration
Register to unlock
Author Profile (Selected)
Register to unlock
References Overview
Register to unlock
Journal & Source
Register to unlock
Metadata & Integrity
Register to unlock