Predicting greenhouse gases emissions from decentralized composting by applying explainable machine learning method

Area of Science:

• Environmental Science
• Waste Management
• Climate Change Research

Background:

• Greenhouse gas (GHG) emissions from composting are a significant environmental concern.
• Predicting these emissions is difficult due to the complex nature of organic waste and varying composting conditions.

Purpose of the Study:

• To enhance the predictability of methane (CH4) and nitrous oxide (N2O) emissions from composting.
• To identify key factors influencing GHG emissions during the composting process.

Main Methods:

• Collected 501 filed-monitoring datasets on methane and nitrous oxide effluxes from seven decentralized composting sites in China.
• Applied explainable machine learning methods, including Adapt Boosting and Gradient Boosting, to predict GHG emissions.
• Analyzed the influence of factors such as pile temperature and C/N ratio on emissions.

Main Results:

• Methane effluxes ranged from 2.57 × 10⁻⁵ to 32.41 mg·m⁻²·min⁻¹, and nitrous oxide effluxes ranged from 1.98 × 10⁻⁴ to 2.27 mg·m⁻²·min⁻¹.
• Adapt Boosting and Gradient Boosting models achieved the highest prediction accuracy.
• Pile temperature and C/N ratio were identified as the primary drivers for methane and nitrous oxide emissions.
• Lifecycle GHG emission factors were found to be 4.5%–15.0% of IPCC average defaults.

Conclusions:

• Machine learning offers a powerful approach for predicting and understanding GHG emissions from composting.
• Controlling pile temperature and C/N ratio can help mitigate GHG emissions.
• The findings provide lower, site-specific emission factors compared to IPCC defaults, aiding in more accurate environmental assessments.