Application of Machine-Learning-Based Fusion Model in Visibility Forecast: A Case Study of Shanghai, China

A visibility forecast model called a boosting-based fusion (BFM) was established in this study. The uses machine learning based on multisource data, including air pollutants, meteorological observations, moderate resolution imaging spectroradiometer (MODIS) aerosol optical depth (AOD) and an operational regional atmospheric environmental modeling System for eastern China (RAEMS) outputs. Extreme gradient boosting (XGBoost), light (LightGBM), numerical prediction method, i.e., RAEMS were fused to establish model. Three sets of models, that is, BFM, LightGBM data (LGBM), RAEMS, used conduct tasks. training set from 1 January 2015 31 December 2018 several pre-processing methods, synthetic minority over-sampling technique (SMOTE) resampling, loss function adjustment, 10-fold cross verification. Moreover, apart the basic features (variables), more spatial temporal considered. testing 2019 adopted validate feasibility LGBM, RAEMS. Statistical indicators confirmed methods improved significantly consistently. root mean square error correlation coefficient BFM next 24/48 h 5.01/5.47 km 0.80/0.77, respectively, which much higher than those statistics binary score analysis different areas Shanghai also proved reliability accuracy using particularly low-visibility forecasting. Overall, is suitable tool predicting visibility. It provides accurate 24 48 LGBM results study provide support real-time forecasts.