Seasonal predictability of the dominant surface ozone pattern over China linked to sea surface temperature

Surface ozone (O3) in the lower troposphere is a harmful pollutant linked to substantial health and ecological impacts in China. O3 is a secondary pollutant formed photochemically from nitrogen oxides and hydrocarbons. Recent years have seen O3 emerge as a major pollutant in Chinese urban areas, with notable increases in northern China; yet regional trends also reflect significant natural and climate variability. Long-lead predictions are valuable for planning emission controls months in advance. Beyond emissions, large-scale ocean–atmosphere circulations substantially influence O3 levels, offering potential predictability due to their inherent memory. Prior studies have leveraged large-scale climate patterns to predict seasonal O3 in the United States and shown the role of teleconnections such as the North Atlantic Oscillation in Europe. In China, meteorological conditions conducive to O3 (heat, drought, strong insolation) are linked to broader climate variability, including Western Pacific and Indian Ocean warming, Arctic sea ice changes, Rossby wave activity, and monsoonal influences; ENSO also modulates ozone through dynamical pathways. However, most prior work focuses on specific regions or on O3-related meteorology rather than directly predicting nationwide O3 pollution itself. Furthermore, O3 concentrations exhibit coherent fluctuation patterns across regions. Therefore, this study aims to: (1) identify dominant fluctuation patterns of surface O3 across China using eigenanalysis; (2) quantify their associations with global sea surface temperature anomalies (SSTA); and (3) develop and evaluate a seasonal prediction model for China’s dominant O3 pattern based on SSTA memory.

The paper reviews evidence that large-scale climate variability influences regional air quality and O3. Studies in the U.S. showed seasonal predictability of summertime O3 using large-scale climate patterns. In Europe, NAO-driven circulation anomalies modulate O3 via photochemistry. In China, meteorological drivers (drought, heat, insolation) correlate with broader climate modes; springtime warming in the Western Pacific/Indian Ocean/Ross Sea relates to summer co-occurrence of heat waves and O3 pollution. Arctic sea ice variability can affect northern China O3 via Eurasian circulation changes, and Rossby waves promote anticyclones and stagnation conducive to O3 build-up. The East Asian summer monsoon influences interannual O3 variability. ENSO affects total ozone column via tropopause variations. Recent work identified dominant O3 patterns in eastern China and links to the West Pacific. Collectively, these studies motivate leveraging SSTA’s long-term memory for seasonal O3 prediction, but prior efforts largely target regions or O3-related meteorology rather than directly predicting national-scale O3 fluctuation modes.

Data: The study uses China-wide MDA8 surface O3 concentration dataset (TAP, version 2) at 0.1°×0.1° resolution from June 2013–February 2023 for summer (JJA) and winter (DJF). Global daily SST (ERA5, 0.25°×0.25°) from 1991–2023 provides SSTA. Data detrending: O3 is split into training (Jun 2013–Feb 2018) and testing (Jun 2018–Feb 2023). For both periods, the mean seasonal cycle (computed from the training period) is removed for each calendar day. SST is partitioned into 1991–2000, 2001–2010, and 2011–2023; SSTA are computed by removing seasonal cycles from each decade, with the last segment using the first 7 years (excluding the testing year) to avoid data leakage. Principal component analysis: The detrended O3 field Y (N grids × M times) is used to form covariance C=YᵀY/M. Eigenvalues λ and eigenvectors u are obtained from Cu=λu; principal components Vn=Yu define temporal modes. The first principal component (PC1) captures the dominant fluctuation pattern. Network construction and correlation analysis: For each SSTA grid i with time series s_i(t), cross-correlation with O3 PC1 v1(t) is computed over lags 0–365 days. The maximum absolute correlation |CC_{v1,si}| and corresponding lag τ* are recorded. Statistical significance is assessed using two null models: White Noise Null Model (WNNM; complete shuffling) and Monthly Autocorrelation-preserving Null Model (MANM; shuffling month-long blocks to preserve intra-month autocorrelation). A significance threshold λ is set at the 97.5th percentile of MANM correlations. Network links are defined where |CC|>λ and τ*>90 days, indicating potential predictability beyond one season. Spatially contiguous significant SSTA nodes with long lags are grouped into clusters; the four largest clusters are retained for each season. Regression model: For cluster C_j, a weighted SSTA index WS_j(t)=Σ_{i∈Cj} s_i(t)|CC_{v1,si}| / Σ_{i∈Cj}|CC_{v1,si}| is formed. A multiple linear regression predicts PC1 with 15-day moving average: V1(t)=Σ_{j=1}^{Nc} a_j·WS_j(t−τ_j)+b, where τ_j (>90 days) are the identified lags. Coefficients a_j and intercept b are fit using a rolling five-year window from the training period and then fixed to predict the following year; parameters are updated year by year. Model performance is evaluated on training (2013–2018) and testing (2018–2023) periods, and significance of test correlations is assessed by MANM via shuffling within clusters.

- PCA identified dominant O3 fluctuation modes: the first eigenvalue explains 22.8% (summer) and 36.8% (winter) of variance. The first spatial mode is positive over most of China, emphasizing high O3 in northern China (Hebei, Shanxi) in summer and in southern China in winter; the second mode (~11.3% summer; ~11.7% winter) delineates a north–south contrast. - Cross-correlation networks between PC1 and global SSTA revealed significant teleconnections with lags >90 days, enabling potential lead-times beyond one season. Four major SSTA clusters per season were identified. - Summer clusters correspond to atmospheric circulation indices: Niño 1+2 (Walker circulation; CS1), North Pacific High (CS2), West Pacific Subtropical High (CS3), and Pacific–North American pattern (CS4). Correlations (lagged) between indices and PC1 are strong: 0.74±0.18 (Niño1+2–PC1), −0.65±0.21 (NPH–PC1), 0.79±0.16 (WPSH–PC1), 0.71±0.19 (PNA–PC1). Corresponding correlations with SSTA clusters are similarly high (e.g., 0.72±0.18 for Niño1+2–CS1; 0.79±0.16 for WPSH–CS3). - Winter clusters align with SOI (CW1), NAO (CW2), Amundsen Sea Low (CW3), and Madden–Julian Oscillation (CW4). Lagged correlations with PC1: −0.63±0.21 (SOI), −0.42±0.25 (NAO), −0.53±0.23 (ASL), −0.58±0.22 (MJO). Correlations with SSTA clusters are of comparable magnitude (e.g., −0.66±0.20 for SOI–CW1). - Physical mechanisms: Identified circulation anomalies produce high-pressure, warm, dry, low-cloud conditions over China (especially north China in summer) or strengthen north-to-south transport in winter, fostering photochemical O3 formation and pollutant transport. - Predictive model performance: Multiple linear regression using delayed SSTA cluster indices achieves high training correlations with observed PC1 (R0=0.89 summer; 0.81 winter). On the independent testing period, correlations remain around 0.5 (R1≈0.54 summer; 0.46 winter), exceeding the 97.5th percentile of MANM-based null distributions, indicating statistical significance. Predictive skill varies across years and clusters.

The study addresses whether global SSTA, via their long memory and teleconnections, can provide seasonal predictability of China’s dominant surface O3 pattern. By extracting PC1 of O3 fluctuations, the authors isolate coherent large-scale behavior distinct between seasons. Significant, time-delayed correlations between PC1 and specific SSTA clusters demonstrate robust links to key circulation systems (Walker circulation, NPH, WPSH, PNA in summer; SOI, NAO, ASL, MJO in winter). These circulations shape meteorological environments (pressure, humidity, temperature, radiation, cloud cover, and winds) favorable for O3 formation or transport. Incorporating these SSTA clusters into a lagged multivariate regression yields skillful forecasts of the dominant O3 mode at lead times >3 months. The model captures seasonal-to-interannual variability and broad trends in PC1, though it is less effective for short-term fluctuations. The significance tests against autocorrelation-preserving null models strengthen confidence that the predictability arises from physical teleconnections rather than spurious correlations. The results support using ocean memory to inform preseasonal air quality planning in China.

This work identifies and predicts China’s dominant surface O3 fluctuation patterns using eigenanalysis and SSTA-based teleconnections. The first O3 mode explains substantial variance and exhibits distinct seasonal spatial structures. Four key SSTA clusters per season, linked to well-known atmospheric circulation indices, show significant, delayed correlations with the O3 mode, enabling predictive lead times beyond one season. A simple multivariate regression driven by weighted, lagged SSTA cluster indices achieves high training skill and statistically significant testing performance (R≈0.5), indicating practical value for preseasonal O3 outlooks and mitigation planning. Future research could: (1) extend predictors to additional ocean–atmosphere indices and land-surface variables; (2) integrate chemistry–transport model outputs with statistical frameworks; (3) explore nonlinear or machine learning models and multi-mode predictions beyond PC1; (4) assess robustness under climate change and across more seasons/regions; and (5) investigate regime-dependent predictability and year-to-year variability in cluster influence.

- The predictive framework targets only the first principal component (PC1), not full spatial O3 fields or higher-order modes, potentially omitting region-specific dynamics. - Skill primarily reflects broad trends; the authors note limited capability in capturing short-term fluctuations. - Predictive power varies across years and among SSTA clusters, indicating nonstationarity and potential regime dependency. - Dependence on detrending choices and moving averages (15-day) may smooth extremes. - Interdependence among atmospheric circulation indices (e.g., ASL influence on MJO and SOI) may complicate attribution. - Training data span is relatively short (about 5 years for fitting windows), which can constrain model generalizability. - Affiliations and data sources suggest robust datasets, but prediction relies on SSTA memory; unforeseen shifts in ocean dynamics under climate change may affect stability of teleconnections.