Municipal solid waste (MSW) landfills significantly contribute to global methane gas production, underscoring the critical need for accurate emission gas estimation within an effective gas management strategy. While first-order models such as LandGEM are essential for estimating gas emissions, their lack of accuracy has spurred numerous studies to enhance core parameters, specifically methane generation rate constant (k) and potential methane generation capacity (L0). In this study, various machine learning models were used to generate modified LandGEM model parameters to reduce the error of methane gas estimations by the model. Using inverse modeling, we calculated k-inverse values based on methane generation data from gas collection systems and their efficiencies. A dataset was then created, incorporating average annual precipitation as an independent variable and k-inverse as a dependent variable. After training various machine learning models, including the k-nearest neighbors (KNN) algorithm, we achieved the best correlation coefficient (R2) of 0.62 between k-inverse (derived through inverse modeling) and k-predicted (from the machine learning model) values. Despite the low R2, the KNN model's methane generation predictions were significantly more accurate than the Inventory and CAA defaults of the LandGEM model. The 54% and 84% error reductions for Inventory and CAA default parameter values of the LandGEM model were achieved, respectively. This study highlights the potential of machine learning models to predict methane generation in landfills more accurately than LandGEM software, thereby enhancing the effectiveness of policies and strategies to manage landfill emissions.
Read full abstract