Prediction of aerosol scattering and absorption coefficients based on machine learning

Abstract

Aerosol scattering and absorption coefficients are important parameters that characterize the optical properties of aerosols, which have significant impacts on the radiation balance, air quality, and climate change of the Earth. In order to further improve the understanding of the relationship between aerosol optical properties and meteorological parameters in the offshore areas of Guangdong Maoming, the scattering and absorption coefficients of aerosols as well as meteorological parameters such as temperature, humidity, pressure, wind speed, wind direction, and visibility were measured. In this study, a prediction model of aerosol scattering and absorption coefficients based on the CatBoost algorithm was proposed using the measured data. Firstly, the measured data was preprocessed, and then a CatBoost algorithm model based on ensemble learning was constructed and trained. The Optuna framework was used to optimize the hyperparameters of the model to obtain the final aerosol scattering and absorption coefficient prediction model. Finally, the machine learning model was used to predict the scattering and absorption coefficients of aerosols in the offshore areas of Maoming. The model was compared with XGBoost and LightGBM algorithm models, and the mean squared error (MSE) and mean absolute error (MAE) were used as evaluation metrics to assess the accuracy of the model predictions. Based on the evaluation metrics, the CatBoost algorithm model based on Optuna automatic hyperparameter optimization performed the best among several models. The experimental results showed that when the training and testing data came from the same region, the MAE of the CatBoost algorithm model based on Optuna hyperparameter optimization was about 5.33, and the MSE was about 48.764, achieving a prediction accuracy of 90.88% for aerosol scattering and absorption coefficients.