A machine learning-based method for identifying the meteorological field potentially inducing ozone pollution

meteorological conditions are critical for ozone pollution. When numerical forecasts are provided for future meteorological fields, it is possible to determine whether they will cause ozone pollution, which is important for local administration to take precautions to reduce pollution. A new machine-learning-based identification method is proposed in this study to quickly identify meteorological fields associated with ozone pollution. This algorithm consists of five steps: (1) Using the Local Linear Embedding (LLE) method to reduce the dimension of the ERA5 historical data from 2015 to 2021 to obtain a historical sample library containing the characteristics of the pressure field (P), humidity field (R), u-component of wind and v-component of wind (U, V), temperature field (T), at the surface as well as 850 hPa; (2) after the new model-output data is generated, its features will be extracted by dimensionality reduction to form the sample to be identified; (3) by measuring the Euclidean distance between the sample to be identified and each sample in the historical sample library, several historical samples most similar to the sample to be identified can be determined; and (4) by checking the ozone concentration on the actual date corresponding to the most similar historical samples, we can determine whether the sample to be identified belongs to the meteorological field that is prone to cause ozone pollution; (5) calibrating the identification results by using a local meteorological index, which describe whether local surface meteorological conditions are prone to ozone pollution. The effectiveness of this method is tested using the ERA5 model data of 2022 with Guangzhou as the research object, and the identification accuracy of this method is 80.27%, and the missed alarm rate of 3.17%, which has certain application prospects.