Smart sensing-enabled risk-aware nitrogen prescriptions via conformal profit bounds for precision agriculture

Introduction: Smart sensing is becoming central to plant science because it supports crop management decisions that reflect dynamic plant-environment interactions rather than field-level averages. Nitrogen fertilization is one of the most important decisions in precision agriculture, but site-specific prescriptions are often difficult to trust because spatial and seasonal variability can make point recommendations unstable. Methods: We propose an uncertainty-aware nitrogen prescription framework that combines yield-response modeling with conformal prediction intervals and propagates uncertainty into profit bounds under an environmental proxy penalty. Nitrogen rates are selected by maximizing the lower confidence bound of profit, and the framework allows abstention when competing rates are indistinguishable or uncertainty is excessive. Evaluation used a leakage-free group split of 10,000 samples from a Kaggle yield dataset. Results: The selected tree-ensemble model achieved RMSE 1.531 and MAE 1.214. Conformal intervals reached empirical coverage of 0.912 at the 0.90 target and 0.963 at the 0.95 target. Expected-profit optimization gave mean profit 4.68 with mean N 121.34, whereas the risk-aware strategy gave mean profit 4.54 with mean N 112.07 and lower variability. Abstention withheld recommendations for 18.4% and 27.9% of cases at 0.90 and 0.95 coverage. Discussion: The framework supports conservative, trustworthy nitrogen decisions and promotes practical nutrient stewardship in variable agricultural fields.