Related Experiment Video
Updated: Jun 26, 2026
Calibrated Passive Sampling - Multi-plot Field Measurements of NH3 Emissions with a Combination of Dynamic Tube Method and Passive Samplers
Published on: March 21, 2016
Machine learning for ammonia volatilization prediction and slurry application management
Armand Favrot1, Sophie Génermont2, Céline Décuq2
1Université Paris-Saclay, INRAE, AgroParisTech, UMR EcoSys, 91120 Palaiseau, France; Université Paris-Saclay, INRAE, AgroParisTech, UMR MIA-PS, 91120 Palaiseau, France.
Machine learning models, specifically random forest and gradient boosting, accurately predict agricultural ammonia emissions. These advanced models outperform traditional methods and show promise for evaluating emission reduction strategies in slurry management.
Area of Science:
- Environmental Science
- Agricultural Science
- Data Science
Background:
- Anthropogenic ammonia emissions, mainly from agricultural fertilization, cause nitrogen loss and air pollution.
- Accurate estimation of ammonia emissions is vital for national inventories and environmental policy.
- Machine learning (ML) applications for predicting ammonia emissions are currently limited.
Purpose of the Study:
- To train and compare ML models (random forest, gradient boosting, lasso) for predicting ammonia emissions after manure application.
- To evaluate the performance of ML models against the semi-empirical ALFAM2 model.
- To assess the effectiveness of different slurry management techniques in reducing ammonia emissions.
Main Methods:
- Utilized 5939 ammonia emission data points from 538 trials in the ALFAM2 database.
- Trained random forest, gradient boosting, and lasso ML models to predict cumulative ammonia emissions 72 hours post-manure application.
- Compared ML model performance against the ALFAM2 model using an independent test dataset and evaluated slurry management techniques across 128 scenarios.
Main Results:
- Random forest (RMSE = 4.51, r = 0.94) and gradient boosting (RMSE = 6.19, r = 0.89) demonstrated superior performance in predicting ammonia emissions.
- Both ML models outperformed the semi-empirical ALFAM2 model, which performed comparably to the lasso model.
- Alternative slurry management techniques reduced emissions by a median of -13.6% to -61.7% compared to broadcast application.
Conclusions:
- Machine learning models, particularly random forest and gradient boosting, show significant promise for accurately predicting agricultural ammonia emissions.
- ML models offer a valuable tool for assessing the efficacy of various slurry management practices in mitigating ammonia release.
- The choice of ML algorithm is critical and impacts model sensitivity and predictive accuracy for ammonia emission assessment.
