Application of artificial neural network forecasting ambient sulfur dioxide concentration: a case of Mae Moh District, Lampang Province

This study investigates the application of a Long Short-Term Memory (LSTM) network to predict 24-hour ahead ambient sulfur dioxide (SO2) concentrations in the vicinity of a coal-fired power plant. Recognizing SO2 as a significant pollutant associated with coal-based electricity generation and its potential impacts on the surrounding environment and communities, the research developed and evaluated an LSTM model using historical meteorological and air quality time-series data from monitoring stations around the Mae Moh power plant. The model’s predictive performance assessed using Coefficient of Determination (R2), Mean Absolute Error (MAE), and Mean Squared Error (MSE). SHapley Additive exPlanations (SHAP) analysis revealed a spatially heterogeneous influence of input variables, with direct emission-related pollutants being more influential near the plant, while meteorological factors played a greater role at more remote stations. These findings underscore the importance of considering localized environmental factors and potentially implementing station-specific model adjustments to enhance the accuracy of SO2 forecasting for effective environmental monitoring and management in the region.