Ground-level ozone pollution is a significant environmental and health concern in China, causing premature deaths and harming ecosystems. Predicting surface ozone concentrations is crucial for effective mitigation strategies, especially long-term predictions that enable proactive planning. While precursor emissions are a key factor, large-scale ocean-atmosphere circulations also have substantial impacts. Previous research has shown the potential of using large-scale climatic patterns for seasonal ozone prediction in other regions, linking them to long-term memory in ocean-atmospheric systems. Meteorological conditions like drought, high temperatures, and intense sunlight also influence ozone formation, and their relationship with broader climatic patterns suggests potential for seasonal prediction. Studies have explored connections between springtime warming in various ocean regions and ozone pollution in China, along with the indirect influence of Arctic sea ice and the role of the East Asian summer monsoon. However, existing literature mainly focuses on specific regions or long-term variability, without directly addressing long-term prediction across all of China. This study aims to identify crucial fluctuation patterns in surface ozone across China and their associations with global SSTA, leveraging the long-lasting memory of SSTA for seasonal prediction. Eigenanalysis, including PCA and EOF, is employed to decompose multiple fluctuation patterns into independent principal modes, focusing on predicting the time series of the dominant pattern to address the complexity of ozone pollution across various regions.

Numerous studies have linked ground-level ozone pollution to significant health impacts worldwide, with hundreds of thousands of premature deaths annually attributed to it. Surface ozone, a secondary pollutant, is formed through photochemical reactions involving nitrogen oxides and hydrocarbons. In recent years, ozone has become a major pollutant in China's urban areas, with some regions experiencing substantial increases. However, in other regions, the trends are less clear, highlighting the role of natural and climate variability. The influence of large-scale ocean-atmosphere circulations on ozone levels is well-established, making these circulations critical for long-term forecasting. Studies have demonstrated successful seasonal prediction of surface ozone concentrations in the United States, leveraging the connection between large-scale patterns and inherent long-term memory. Similar relationships have been observed in Europe, where the North Atlantic Oscillation influences ozone pollution by modulating photochemical reactions. Meteorological conditions are also conducive to ozone formation, making their link to climate patterns relevant for prediction. Research has connected springtime ocean warming to shifts in the frequency of heat waves and ozone pollution in China, identified Arctic sea ice as an indirect influence, and highlighted the role of Rossby waves and the East Asian summer monsoon. Existing studies often focus on specific regions or aspects of ozone variability, leaving a gap in the understanding of long-term prediction across China. The authors highlight that predicting ozone-related meteorological conditions is not equivalent to predicting ozone pollution itself and that patterns of ozone fluctuation result from correlations between different regions.

The study uses the maximum daily 8-hour average (MDA8) surface ozone concentration dataset (version 2) over China from June 2013 to February 2023, sourced from the Tracking Air Pollution in China (TAP) project. This dataset integrates ground measurements, satellite retrievals, and model outputs. Global daily average Sea Surface Temperature (SST) data from the ERA5 Reanalysis are also utilized. Data preprocessing involves detrending: removing the mean seasonal cycle (calculated from the training period, June 2013 to February 2018) from the ozone data for both training and testing periods (June 2018 to February 2023). For SST, data from 1991 to 2023 are divided into three parts to obtain SSTA (Sea Surface Temperature Anomalies). Eigenanalysis is performed on the detrended ozone field for summer and winter to identify principal modes of ozone fluctuation. The first principal component (PC1) represents the dominant pattern. The cross-correlation function is calculated between PC1 and global SSTA time series to quantify their relationships, considering time delays. Significance tests, using the White Noise Null Model (WNNM) and Monthly Autocorrelation-preserving Null Model (MANM), are employed to distinguish true correlations from spurious ones. Significant correlations are used to identify critical SSTA regions. Four clusters are formed from spatially contiguous critical SSTA nodes, indicating teleconnections between SSTA and ozone levels. The corresponding atmospheric circulation indices linked to these clusters are identified and analyzed to explore the physical mechanisms. A multivariate linear regression model is used to predict the time series of PC1, using the identified SSTA clusters as predictors. The model is trained on data from June 2013 to February 2018 and tested on data from June 2018 to February 2023. The significance of the predictive performance is evaluated using the MANM test.

Eigenanalysis revealed that the first principal component (PC1) captures the dominant pattern of ozone fluctuation in China during both summer and winter. The spatial distribution of PC1 shows that Northern China experiences higher ozone concentrations in summer, while Southern China dominates in winter. Cross-correlation analysis identified four significant SSTA clusters influencing PC1. In summer, these clusters are linked to the Walker circulation, North Pacific High, West Pacific Subtropical High, and Pacific-North American teleconnection pattern. In winter, they are associated with the Southern Oscillation, North Atlantic Oscillation, Amundsen Sea Low, and Madden-Julian Oscillation. The multivariate regression model, utilizing these SSTA clusters as predictors, successfully predicts the time series of PC1 with a lead time of at least 3 months. The model achieved high prediction accuracy (R-values of 0.89 and 0.81 for summer and winter respectively) in the training dataset and moderate accuracy (R-values around 0.5 for both seasons) in the testing dataset. The MANM test confirmed the statistical significance of the model's predictive performance. The study also shows that some SSTA regions exhibit long-term memory behavior, with delay times exceeding 90 days, suggesting the potential for even longer lead-time predictions.

The findings demonstrate a strong link between large-scale climate patterns (represented by SSTA clusters) and the dominant pattern of surface ozone pollution in China. The successful prediction of ozone patterns using the multivariate regression model, with a lead time of at least three months, highlights the potential for using SSTA as a predictor for seasonal ozone forecasting. The identification of specific atmospheric circulations associated with each SSTA cluster provides valuable insight into the underlying physical mechanisms driving ozone pollution variability. The differences in spatial patterns and associated atmospheric circulations between summer and winter emphasize the seasonality of ozone pollution and the importance of considering seasonal factors in mitigation strategies. While the model's predictive power is significant, further research is needed to improve accuracy and potentially extend lead times. This could involve incorporating additional climate variables or refining the model's structure.

This study successfully established a link between global SSTA and the seasonal predictability of surface ozone pollution in China. The developed multivariate regression model demonstrates the potential for forecasting dominant ozone patterns with at least a three-month lead time, offering valuable insights for effective mitigation strategies. Future research could focus on improving model accuracy by including additional factors and exploring the potential for even longer-lead-time predictions. The findings highlight the importance of considering large-scale climate patterns in air quality management.

The study's predictive model focuses on the dominant ozone pattern (PC1), potentially overlooking regional variations not captured by this primary mode. The model's accuracy might vary depending on the year and specific SSTA cluster, as shown by the interannual variability in the testing period. The study primarily relies on statistical relationships, and further research is needed to fully elucidate the underlying physical mechanisms. Improved accuracy might be achieved by integrating more detailed information on precursor emissions and high-resolution meteorological data.