Prediction and evaluation of spatial distributions of ozone and urban heat island using a machine learning modified land use regression method

In summer, Ozone (O₃) pollution and urban heat island (UHI) pose serious health risks to humans. To obtain the spatial distributions of ozone and urban heat island in Xi'an in summer and develop a simultaneous control strategy of ozone and urban heat island, the land use regression model is modified and improved using the machine learning random forest algorithm. The LUR-Kriging-RF integrated prediction model is then established. The land use regression and kriging are used to extract the feature variables, while random forest is used to establish a regression model. The spatial distribution maps of ozone and urban heat island in Xi'an are obtained by regression mapping of the prediction model, and the spatial relationships between them are analyzed. The SHapley Additive explanation (SHAP) and partial dependence plot (PDP) are adopted to explain the way feature variables act on ozone and urban heat island. Based on the spatial distribution and interaction mode, a simultaneous control strategy of ozone and urban heat island in Xi'an is put forward. For ozone, the R² of the integrated prediction model (0.65) is higher than that of land use regression (0.4), while the RMSE (28.18) of the integrated model is lower than that of land use regression (35.66). For temperature, the R² of the integrated model (0.93) is higher than that of land use regression (0.8), while its RMSE (0.92) is lower than that of land use regression (1.52). The performance of the LUR-Kriging-RF integrated prediction model is better than that of land use regression. This study reveals the spatial interactions between ozone and land use regression in the central urban areas. The suitable strategies for mapping ozone pollution and urban heat island control include reducing VOCs emissions from industrial sources and agricultural sources, increasing plants with low VOCs emissions, and spray humidification. This study can be used to evaluate ozone exposure and thermal exposure, provide scientific support for environmental protection and urban heat island control policies, contribute to reducing public health threats, promote the sustainability of urban environments, and promote the practical application of machine learning in this field.