Modelling of greenhouse gas emissions in wastewater treatment plants: a data-driven approach

Abstract

With the impacts of global warming on our planet, strategies are developed to increase sustainability by reducing greenhouse gas (GHG) emissions. While these gases are essential to keep the Earth’s warmth and support life, too much of them are emitted due to human activity, causing disruptive changes to the planet’s climate. Although having a smaller contribution compared to other sectors such as energy, transportation or agriculture, sewage treatment has a notable impact on GHG emissions, generating a range of gases including methane (CH4 ), carbon dioxide (CO2 ), and nitrous oxide (N2 O). Data-driven approaches are employed to understand more of the GHG emissions’ behaviour and to develop machine learning models that can predict these emissions and perform control actions in the plant to reduce them. This work explores data-driven techniques to analyse and model GHG emissions in a Wastewater Treatment Plant (WWTP) located in Helsinki, Finland. The research follows three key steps: data analysis, pre-processing, and modelling. First, a detailed analysis of the dataset enhances process understanding and identifies how variations in predictor variables influence GHG emissions. Next, pre-processing techniques are applied to improve the modelling process. Finally, machine learning models - both linear and non-linear - are developed to predict nitrous oxide (N2 O) and carbon dioxide (CO2 ) emissions. Results demonstrate that employing rolling window approaches, even with simple models, generally improve model accuracy, offering promising perspectives for emission modelling in WWTPs.