Greenhouse gas rebound effects on disruptive events via SHAP-assisted neural networks

Greenhouse gas (GHG) rebound and its lagging effects, resulting from disruptive events such as the COVID-19 pandemic, are often underestimated and sensitive to socioeconomic fluctuations. Artificial intelligence models such as artificial neural networks are widely adopted in research studies for GHG forecasting. In this study, a Bayesian-optimized artificial neural network model, complemented with Shapley Additive exPlanations (SHAP), is applied to forecast sectoral GHG emissions and rebound effects under disruptive events based on highly correlated socioeconomic indicators, marking a first-of-its-kind approach. Malaysia is chosen as the case study. Nine sectoral emissions are forecasted under pandemic and non-pandemic circumstances. The forecasted overall GHG emissions indicate that a GHG rebound in Malaysia is apparent by 2022, with 9.56% higher GHG emissions in the pandemic scenario compared to the non-pandemic scenario. The “Industrial processes and product use” sector exhibits a high rebound, surpassing pre-pandemic levels, while the “Transportation” sector experiences a moderate rebound, both contributing to an increase in overall GHG emissions until 2030. SHAP analysis reveals that socioeconomic indicators have varying influences over sectoral emissions, based on ranking and SHAP value magnitude differences. The findings underscore the need for policymakers to reassess their climate goals, considering the repercussions of disruptive events like COVID-19, while also narrowing the emission gap between climate goals and post-pandemic emissions. Sectoral policies are tailored based on the obtained results from the socioeconomic-explanatory ANN model and are ranked following the GHG emissions proportion contributing to the rebound and its lagging effect.