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

Greenhouse gases (GHGs) emission from composting was a major concern, however it is still unpredictable due to the complex characteristics of organic waste and differentiated composting conditions. In order to raise the predictability of GHGs emissions from composting, 501 filed-monitoring based datasets of methane and nitrous oxide effluxes and five influential factors were collected from seven decentralized composting sites in China, and then explainable machine learning method was applied for predicting their GHGs emissions. It was found that methane and nitrous oxide effluxes varied between 2.57 × 10^-5∼32.41 mg (CH₄)·m^-2·min^-1 and 1.98 × 10^-4∼2.27 mg(N₂O)·m^-2·min^-1, while Adapt Boosting and Gradient Boosting model can achieve the highest predicting accuracy. Pile temperature and C/N ratio were the key drivers for methane and nitrous oxide emissions. The lifecycle GHGs emission factors suggested herein were 4.5 %∼15.0 % of those in IPCC average defaults. This work provides an innovative approach to understand and control GHGs emission from composting.