Applications of Digital Agriculture for Rice Nutrient Management and Yield Prediction Using UAV-Based Remote Sensing and Machine Learning Models

Author

• Ritik Pokharel, Louisiana State University and Agricultural and Mechanical CollegeFollow

Semester of Graduation

Spring 2026

Degree

Master of Science (MS)

Department

Plant, Environmental, and Soil Sciences

Document Type

Thesis

Abstract

Efficient nitrogen (N) management and accurate yield prediction in rice (Oryza sativa L.) are constrained by strong interactions among genotype, environment, and crop N demand. This study integrates multi-year field experiments (2023-2025), unmanned aerial vehicle (UAV) multispectral remote sensing, and machine learning to quantify varietal N response, assess nitrogen use efficiency (NUE), and develop operational frameworks for yield prediction and nitrogen-efficient variety characterization. Nitrogen × variety trials at the LSU AgCenter Rice Research Station included six N rates (0-235 kg N ha⁻¹) across multiple cultivars. Quadratic-plateau modeling revealed economic optimum nitrogen rate (EONR) ranging from 96 to 229 kg N ha⁻¹, with most varieties achieving optima between 130 and 180 kg N ha⁻¹. NUE declined consistently beyond 150-170 kg N ha⁻¹. A classification framework distinguished nitrogen-efficient (Class A) from inefficient (Class D) varieties based on yield-EONR performance. Among 19 vegetation indices, the normalized difference red-edge index (NDRE) showed the strongest relationship with grain yield (r = 0.73-0.85 at 90 DAP). Temporal analysis revealed Class A varieties maintained 9.6-27.8% higher NDRE than Class D across all N rates, with efficiency originating from 27.8% superior baseline N acquisition and 14.6% higher plateau capacity. Binary logistic regression achieved 80% accuracy and perfect specificity (100%) in classifying varieties using NDRE at 168 kg N ha⁻¹ (threshold = 0.530). Machine-learning models achieved high grain yield predictive accuracy (R² = 0.89 internal validation), with XGBoost maintaining robust independent validation in 2025 (R² = 0.63; NRMSE < 9%). This integrated framework provides scalable decision support for variety specific nitrogen management and climate-smart rice production.

Date

3-27-2026

Recommended Citation

Pokharel, Ritik, "Applications of Digital Agriculture for Rice Nutrient Management and Yield Prediction Using UAV-Based Remote Sensing and Machine Learning Models" (2026). LSU Master's Theses. 6375.
https://repository.lsu.edu/gradschool_theses/6375

Committee Chair

Setiyono, Tri

LSU Acknowledgement

1

LSU Accessibility Acknowledgment

1