Using machine learning with case studies to identify practices that reduce greenhouse gas emissions across Australian grain production regions

Elizabeth Meier, Peter Thorburn, Jody Biggs, Jeda Palmer, Nikki Dumbrell, Marit Kragt

Research output: Contribution to journalArticlepeer-review


It is difficult to identify farm management practices that consistently provide greenhouse gas (GHG) abatement at different locations because effectiveness of practices is greatly influenced by climates and soils. We address this knowledge gap by identifying practices that provide abatement in eight case studies located across diverse conditions in Australian’s grain-producing areas. The case studies focus on soil-based emissions of nitrous oxide (N2O) and changes in soil organic carbon (SOC), simulated over 100 years for 15 cropping management scenarios. Average changes in the balance of GHG from both N2O emissions and SOC sequestration (∆GHG balance) and gross margins compared to a high emissions baseline were determined over 25 and 100 simulated years. Because scenarios providing the greatest abatement varied across individual case studies, we aggregated the data over all case studies and analysed them with a random forest data mining approach to build models for predicting ∆GHG balance. Increased cropping intensity, achieved by including cover crops, additional grains crops, or crops with larger biomass in the rotation, was the leading predictor of ∆GHG balance across the scenarios and sites. Abatement from increased cropping intensity averaged 774 CO2-e ha−1 year−1 (25 years) and 444 kg CO2-e ha−1 year−1 (100 years) compared to the baseline, with reduced emissions from SOC sequestration offsetting increased N2O emissions for both time frames. Increased cropping intensity decreased average gross margins, indicating that a carbon price would likely be needed to maximise GHG abatement from this management. To our knowledge, this is the first time that the random forest approach has been applied to assess management practice effectiveness for achieving GHG abatement over diverse environments. Doing so provided us with more general information about practices that provide GHG abatement than would have come from qualitative comparison of the variable results from the case studies.

Original languageEnglish
Article number29
JournalAgronomy for Sustainable Development
Issue number2
Publication statusPublished - Apr 2023


Dive into the research topics of 'Using machine learning with case studies to identify practices that reduce greenhouse gas emissions across Australian grain production regions'. Together they form a unique fingerprint.

Cite this