Simulating Maize Response to Split-Nitrogen Fertilization Using Easy-to-Collect Local Features

Abstract

Maize (Zea mays) is a high-nitrogen (N)-demanding crop potentially contributing to nitrate contamination and emissions of nitrous oxide. The N fertilization is generally split between sowing time and the V6 stage. The right split N rate to apply at V6 and minimize environmental damage is challenging. Our objectives were to (1) predict maize response to added N at V6 using machine learning (ML) models; and (2) cross-check model outcomes by independent on-farm trials. We assembled 461 N trials conducted in Eastern Canada between 1992 and 2022. The dataset to predict grain yield comprised N dosage, weekly precipitations and corn heat units, seeding date, previous crop, tillage practice, soil series, soil texture, organic matter content, and pH. Random forest and XGBoost predicted grain yield accurately at the V6 stage (R2 = 0.78–0.80; RSME and MAE = 1.22–1.29 and 0.96–0.98 Mg ha−1, respectively). Model accuracy up to the V6 stage was comparable to that of the full-season prediction. The response patterns simulated by varying the N doses showed that grain yield started to plateau at 125–150 kg total N ha−1 in eight out of ten on-farm trials conducted independently. There was great potential for economic and environmental gains from ML-assisted N fertilization.

Feature-specific nutrient management of onion (Allium cepa) using machine learning and compositional methods

“…Although the N dosage can vary widely under different growing conditions the number of N trials was limited (25) in the present study compared the 93 and 461 multi-environmental N fertilizer trials to run ML models on potato ( Solanum tuberosum ) 35 and maize ( Zea mays ) 36 , respectively. More trials and universality tests should be conducted to validate model outcomes in growers’ fields.…”

“…Kyveryga et al 33 stated that the development of new nutrient calibration procedures has been limited by the inability in the past to collect a sufficient number of yield responses to enable calculating reliable economic optimum rates. To follow-up on model predictions, universality tests are needed to verify the reliability of model outcomes in growers’ fields 36 , 61 . The prediction of N dosage can be conducted as shown in S4 by providing the site-specific feature and drawing a response curve predicted from those features.…”

“…Efforts to develop technological tools of precision agriculture for site-specific fertilization have been limited by non-specific state-based fertilizer recommendations. For some high-valued crops like maize, the nitrogen dosage can be adjusted to local factors using ML methods 36 . Observational and experimental data sets could be further combined and processed by machine learning to customize nutrient management for a given set of controllable and uncontrollable features 62 .…”

Vegetable Response to Added Nitrogen and Phosphorus Using Machine Learning Decryption and the N/P Ratio

“…Indeed, features not documented in the database could contribute to model variation. In comparison, the accuracy of ML models for multi-environmental trials reached 0.80 for maize (Zea mays) [13] in Quebec, Canada. Given the substantial strength of the RF model to predict the marketable yields of vegetables, optimum growing conditions could be defined for a given species.…”

“…Such assumption fails at the step of assembling results from several MEFTs because climatic, managerial, and soil factors vary widely and simultaneously among the experimental sites [12]. Using large and diversified datasets, machine learning (ML) models can combine relevant site-specific features to predict yield response to added nutrients [13]. Although complex models such as ML models may show high accuracy, universality tests are still required to verify the model's generalization capacity to real cases in growers' fields unseen by the model [14,15].…”