Biochar application to soil is a potential climate change mitigation strategy. In addition to long-term sequestration of the carbon content of the biochar itself, its application may reduce the emissions of other greenhouse gases (GHGs) from the soil. However, the reported effects of biochar application on soil GHG fluxes exhibit inconsistent results. Prediction of such effects is an important gap that needs to be addressed in biochar research. In this study, rule-based machine learning models were developed based on rough-set theory. Data from the literature were used to generate the rules for predicting the effects of biochar application on soil GHG (CO2, N2O, and CH4) fluxes. Four rule-based models for CO2 fluxes, two rule-based models for N2O fluxes, and three rule-based models for CH4 fluxes were developed. The validity of these models was assessed based on both statistical measures and mechanistic plausibility. The final rule-based models can guide the prediction of changes in soil GHG fluxes due to biochar application, and thus serve as a decision-support tool to maximize the benefits of biochar application as a negative emission technology (NET). In particular, mechanistically plausible rules were identified that predict triggers for GHG fluxes that can offset carbon sequestration gains. This knowledge allows biochar application to be calibrated to local conditions for maximum efficacy.Graphical
Read full abstract