PM2.5 Forecast System by Using Machine Learning and WRF Model, A Case Study: Ho Chi Minh City, Vietnam
Vo Thi Tam Minh, Trần Trung Tín, Tô Thị Hiền
Abstract
Predicting has necessary implications as part of air pollution alerts and the air quality management system. In recent years, air quality studies and observations in Vietnam have shown that pollution is increasing, especially the concentration of PM2.5. There are warnings about excessively high concentrations of PM2.5 in the two major cities of Vietnam as Ho Chi Minh City and Hanoi. Projections for PM2.5 concentrations in these cities will provide short-term predictive data on air quality. Using the WRF model to forecast PM2.5 in Ho Chi Minh City is new research for providing forecast information on air pollution. Experiments with six machine learning algorithms show that the Extra Trees Regression model gives the best forecast with statistical evaluation indicators including RMSE = 7.68 µg m-3, MAE = 5.38 µg m-3, R-squared = 0.68, and the confusion matrix accuracy of 74%. The experimental setting of the Extra Trees Regression algorithm to predict PM2.5 for the next two days with WRF's simulated meteorological data compared with the forecast with observed data showing high accuracy of over 80%. The results show that machine learning with the WRF model can predict PM2.5 concentration, suitable for early warning of pollution and information provision for air quality management system in large cities as Ho Chi Minh City.