Prediction of particulate matter pollution using a long short-term memory model in Zhejiang Province, China

Air pollution, one of the most serious environmental issues that people face, affects the standard of living in urban areas. Strategies for assessing and alerting the public to anticipated hazardous levels of air pollution can be developed using particulate matter (PM) forecasting models. Accurate assessments of pollutant concentrations and forecasts are essential components of air quality evaluations and serve as the foundation for making informed strategic decisions. In the current study, the Long Short-Term Memory (LSTM) model, a deep learning approach, was employed to forecast PM pollution along with meteorological variables in Zhejiang Province, China. The model’s performance was assessed using the cross-validation (CV), mean absolute error (MAE), root mean squared error (RMSE), and the coefficient of determination (R²). According to our findings, the model performed well in predicting PM₁₀ (R² = 0.76, RMSE = 11.51 µg/m³, and MAE = 8.74 µg/m³) and PM_2.5 (R² = 0.74, RMSE = 7.06 µg/m³, and MAE = 5.41 µg/m³) concentrations. Moreover, there was a downward trend in PM concentrations from 2019 to 2022, but Zhejiang Province experienced an increase in PM levels in 2023. These results are reliable and underscore the need for increased efforts to reduce air pollution in the future.