Machine Learning Algorithms Show Promise in Estimating Nitrous Oxide Emissions in Agricultural Landscapes

Strategic Insights -

Research conducted at the University of Guelph in Ontario, Canada, has demonstrated the potential of machine learning algorithms in estimating nitrous oxide (N2O) emissions in agricultural landscapes. A team of researchers, led by Uttam Ghimire, employed multiple algorithms, including random forest regression (RFR), support vector regression (SVR), and artificial neural network (ANN), to test their performance in estimating N2O emissions in data-sparse agricultural landscapes.

Key Takeaways:

  • The study found that the artificial neural network (ANN) algorithm was the most accurate in estimating N2O emissions, capturing 94% of the variability in emissions within the training dataset.
  • The machine learning algorithms were able to capture 92% of high emissions (10 gm/ha/day) within their predictive intervals of 95% confidence.
  • The study concluded that Random Forest Regression (RFR) and Artificial Neural Network (ANN) are recommended for estimating N2O emissions in similar agricultural landscapes in future studies due to their computational cost, ease of fine-tuning, and stable model performance.
  • The study utilized two scenarios, High Input (HI) and Low Input (LI), to evaluate the performance of the machine learning algorithms, with HI using discrete measurements of N2O, rainfall, temperature, fertilizer application dates, soil nitrate, ammonium content, and pH values.
  • The researchers found that the High Input (HI) scenario was better at capturing high emissions compared to the Low Input (LI) scenario.

Statistics:

  • 64% (66%) of the variability of emissions was captured by RFR in the High Input (HI) scenario within the training (testing) datasets.
  • 59% (63%) of the variability of emissions was captured by SVR in the High Input (HI) scenario within the training (testing) datasets.
  • 94% (43%) of the variability of emissions was captured by ANN in the High Input (HI) scenario within the training (testing) datasets.
  • 92%, 29%, and 75% of high emissions were captured by RFR, SVR, and ANN in the High Input (HI) scenario within their predictive intervals of 95% confidence.

Sources:

  • Application of Machine Learning Algorithms in Nitrous Oxide (N2O) Emission Estimation in Data-Sparse Agricultural Landscapes. Atmosphere, 2025,16(6):703. (Atmosphere - http://www.mdpi.com/journal/atmosphere/)
  • University of Guelph, School of Engineering, Guelph, ON N1G2W1, Canada (Contact: Uttam Ghimire)
  • NewsRx. Findings from University of Guelph Broaden Understanding of Machine Learning [Application of Machine Learning Algorithms in Nitrous Oxide (N2O) Emission Estimation in Data-Sparse Agricultural Landscapes]. Agriculture Week. July 10, 2025; p 1250.