A new optimized hybrid approach combining machine learning with WRF-CHIMERE model for PM10 concentration prediction

The forecast of particulate matter PM10 concentration is crucial due to its impacts on public health and the environment. Chemical Transport Models (CTM) are used to predict air quality. However, these models are subject to bias because of the precision of inputs. This paper explores a hybrid approach combining CTM (WRF-CHIMERE) predictions with machine learning (ML) to forecast PM10 concentrations. Five ML algorithms were developed: Multiple Linear Regression (MLR), Random Forests (RF), Extreme Gradient Boosting (XGB), Support Vector Regression (SVR), and Artificial Neural Networks (ANN). This hybrid system was trained using hourly data from September to December 2020 on seven Moroccan sites, incorporating nine parameters including meteorological variables, chemical concentrations, and other spatiotemporal variables. The hybrid model was evaluated against PM10 measurements. The results reveal that CHIMERE combined with RF and with XGB presented the best accuracy of predictions of PM10, when compared to the CHIMERE model. These two hybrid models achieved high correlation coefficients of 0.756 and 0.747, and determination coefficients of 57% and 55.7%, respectively. Moreover, they reduced the mean squared error, with CHIMERE-RF decreasing from 42.39 to 21.72 mu g/m3 and CHIMERE-XGB from 42.39 to 22.05 mu g/m3. Additionally, there was improvement in the mean bias, with CHIMERE-RF changing from -24.40 to -1.324 mu g/m3 and CHIMERE-XGB from -24.40 to -0.890 mu g/m3. The significance of these results could be important for air quality monitoring during extreme dust events, as they provide crucial information and simplify the implementation of preventive measures. This would help minimize the health risks associated with PM10.