Abstract:
Objective To explore a high-precision simulation method for ground-level ozone concentration in China based on multiple machine learning models.
Methods Based on multi-source data from 2013 to 2017, a national ground-level ozone concentration simulation model was established using multiple machine learning algorithms.
Results The random forest (RF) model had the best performance with an R2 of 0.752, and RMSE and MAE of 23.264 μg/m3 and 16.094 μg/m3, respectively. The surface downwelling shortwave radiation was the most critical factor for ground-level ozone concentration simulation.
Conclusion The RF model based on multivariate variables such as meteorology, geography, and emission can realize high-precision simulation of ground-level ozone. In the future, the natural source emission data of air pollutants can be further introduced to improve the accuracy of the model.