Snow cover data were obtained from the National Oceanic and Atmospheric Administration (NOAA) Climate Data Record (Robinson et al., 2012), provided by the Rutgers University Global Snow Lab. This original weekly snow dataset, with a horizontal resolution of 25 km, spans from October 1966 to the present. In this study, the snow data were averaged into monthly means and regridded to 2°×2° grids.

Satellite-derived daily ground-level PM_2.5 concentrations over China were obtained from the Long-term Gap-free High-resolution Air Pollutants concentration dataset (LGHAP) version 2 (Bai et al., 2024). The LGHAP PM_2.5 data, covering 2005–2021, have a spatial resolution of 1 km and were developed by integrating satellite retrievals, ground-based observations, and numerical simulations through machine learning methods. Validation against observations from the China National Environmental Monitoring Center (CNEMC) indicates strong reliability, with a high correlation coefficient of 0.95 and a root mean square error (RMSE) of 12.03 µg m⁻³, supporting its suitability for long-term spatiotemporal analysis (Bai et al., 2024).

Meteorological variables including sea surface temperature (SST), geopotential height, zonal and meridional winds at 200 hPa, 500 hPa and the surface, relative humidity, precipitation, planetary boundary layer height, surface sensible heat flux, surface latent heat flux, surface net long-wave radiation flux, surface net short-wave radiation flux, top net long-wave radiation flux, top net short-wave radiation flux, and snow albedo at a spatial resolution of 0.25°×0.25° were obtained from the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 dataset (Zhang et al., 2022; You et al., 2020; Hersbach et al., 2020). The monthly sea ice concentration data, with a horizontal resolution of 1.0°×1.0° was provided by Met Office Hadley Centre. The winter El Niño indices during 2005–2021 were retrieved from NOAA.

To examine the atmospheric response to snow cover anomalies, the total atmospheric column heat source (Q1), was calculated as, following by Zhao and Chen (2001), 1Q1=SH+Rnet+LP where SH denotes surface sensible heat flux, Rnet represents net atmospheric column radiation, and LP is latent heat released through condensation.

Anthropogenic emissions of sulfur dioxide (SO₂) and nitrogen oxides (NO_x) for the period 2005–2021 were obtained from the Multi-resolution Emission Inventory for China (MEIC) (Wu et al., 2024).

The interannual spatiotemporal variability of winter PM_2.5 concentrations over China during 2005–2021 was analyzed using EOF decomposition. The first two EOF modes were emphasized, as they were well separated from higher-order modes based on the North test (North et al., 1982).

To diagnose the propagation characteristics of atmospheric Rossby waves, the horizontal wave activity flux (WAF) was computed using the formulation based on the conservation of wave-activity momentum (Takaya and Nakamura, 2001): 2W=12U‾u‾(Ψx′2-Ψ′Ψxx′)+v‾(Ψx′Ψy′-Ψ′Ψxy′)u‾(Ψx′Ψy′-Ψ′Ψxy′)+v‾(Ψy′2-Ψ′Ψyy′) where ψ denotes the geostrophic stream function, and subscripts represent partial derivatives. U corresponds to the horizontal wind vector, whereas u and v indicate its zonal and meridional components, respectively. W represents the two-dimensional Rossby wave activity flux.

To assess whether a synergistic effect exists between the influences of Niño 1+2 index and Tibetan Plateau snow cover (TPSC) on winter PM_2.5 variability, the statistical diagnostic method proposed by Li et al. (2019) was applied. The Niño 1+2 and TPSC indices were classified into positive (+), neutral (0), and negative (-) phases based on ±0.5 standard deviations. The combinations of Niño 1+2 and TPSC phases are summarized in Table 1.

The Community Earth System Model version 2.1.3 (CESM v2.1.3) was employed to investigate how circulation patterns and PM_2.5 concentrations respond to SST and snow cover anomalies. The model was configured with a horizontal resolution of 0.94°×1.25° and 70 vertical layers (Gent et al., 2011), using the FWHIST component. The atmospheric processes were simulated using the Community Atmosphere Model version 6, while chemical and land processes were represented by the Whole Atmosphere Community Climate Model version 6 (Gettelman et al., 2019). The aerosol microphysical and chemical processes in CESM have been extensively developed and validated in previous studies (Gettelman et al., 2019; Danabasoglu et al., 2020; Liu et al., 2021). We assessed CESM's ability to reproduce observed surface air temperature, relative humidity, and PM_2.5 concentrations in the control simulation (CESM_ctrl), as shown in Fig. S1 in the Supplement. The results demonstrate that CESM_ctrl generally captures the spatial patterns and interannual variability of these variables. Quantitatively, the mean fractional biases (MFBs) and mean fractional errors (MFEs) for PM_2.5 fall within the widely adopted model performance criteria of ±60% for MFB and below +75% for MFE (Boylan and Russell, 2006), indicating acceptable simulation skill. CESM reproduces meteorological variables better than aerosol concentrations, which are subject to larger uncertainties arising from emission inputs, chemical processing, and aerosol-cloud interactions (Liu et al., 2021; Danabasoglu et al., 2020).

SST was prescribed to isolate the effects of SST forcing. CESM_ctrl was forced with monthly varying climatological SSTs. Two additional sensitivity experiments were conducted: (1) CESM_ElNiño, in which composite winter SST anomalies associated with Niño 1+2 were imposed; and (2) CESM_TPSC, in which surface albedo over the northern Tibetan Plateau (86–94° E, 35–40° N) was set to a minimum of 0.8 to represent enhanced snow cover conditions (Cohen and Rind, 1991), following Wang et al. (2021). Specifically, when the model-simulated albedo falls below 0.8, it is reset to 0.8 to represent increased snow cover over the northern Tibetan Plateau. Grid points where albedo already exceeds this threshold retain their original values. All simulations covered the period from November 2010 to February 2011, during which both Niño 1+2 and TPSC were in neutral phases (Table 1). Surface PM_2.5 concentrations were derived from the model output by extracting the lowest vertical level of the simulated three-dimensional PM_2.5 field. Given the uncertainties in CESM-simulated PM_2.5 concentrations, model outputs are interpreted in terms of the direction and spatial pattern of PM_2.5 changes rather than their absolute magnitudes. Statistical significance was evaluated using daily model output during the winter simulation period.

Figure 1 presents the leading two EOF modes of winter PM_2.5 concentrations over China. Both modes are distinctly separated according to the North test, indicating their statistical robustness. The dominant mode (EOF1), accounting for 56.2 % of the total variance, exhibits a spatially coherent pattern characterized by uniformly elevated PM_2.5 concentrations across eastern China. The corresponding normalized principal component (PC1) shows a pronounced increase from 2005 to 2013, followed by a marked decline after 2014. This temporal evolution closely mirrors trends in anthropogenic emissions, with rapid growth in energy consumption and emissions during 2005–2013 while the introduction of stringent clean air policies since 2013. Quantitatively, PC1 is almost perfectly correlated with regional mean PM_2.5 concentrations (R=0.99) and also shows strong correlations with anthropogenic NO_x and SO₂ emissions (R=0.70 and 0.71, respectively). These relationships indicate that EOF1 primarily represents the influence of anthropogenic emission variability on winter PM_2.5 over eastern China.

EOF2, accounting for 13.2 % of the total variance, reveals a pronounced north-south dipole pattern in winter PM_2.5 concentrations over eastern China, characterized by positive anomalies in northern China and negative anomalies in southern China. PC2 displays strong interannual variability, with predominantly negative values before 2012 and mainly positive values during 2013–2020. We examined the correlations between PC2 and anthropogenic SO₂ and NO_x emissions across eastern, northern, and southern China. The correlation coefficients between PC2 and SO₂ (NOx) emissions are -0.31 (-0.01), -0.37 (0.07), and -0.33 (-0.11) in eastern, northern, and southern China, respectively. None of these correlations are statistically significant, and critically, they exhibit no north-south dipole structure that could mirror the EOF2 pattern. This confirms that the spatially heterogeneous emission controls do not drive the EOF2 dipole structure. To elucidate the meteorological conditions underlying the EOF2 dipole structure, regression maps of atmospheric circulation and meteorological fields onto the normalized PC2 are shown in Fig. 2. At the 500 hPa level, a pronounced anticyclonic circulation anomaly is observed over northeastern China and Japan, accompanied by a weaker cyclonic anomaly over the southern Tibetan Plateau. This configuration induces anomalous southerly flow across eastern China. In southern China, these southerly winds transport humid air from the South China Sea, enhancing precipitation by approximately 0.2 mm per month and promoting wet removal of PM_2.5. In contrast, over northern China, the southerly anomalies weaken atmospheric ventilation, as reflected by reduced surface wind speeds (-0.05 m s⁻¹) and a suppressed planetary boundary layer height (-30 m), favoring pollutant accumulation and increasing PM_2.5 levels. As a result, anomalous southerly flow exerts opposite influences on winter PM_2.5 concentrations in northern and southern China, which is consistent with previous findings (Zhang et al., 2022; An et al., 2022). These results confirm that EOF2 dipole pattern primarily reflects large-scale meteorological forcing rather than anthropogenic emission variability.

To further elucidate the large-scale climate drivers underlying the EOF2 mode, correlation analyses were performed between PC2 and winter SST anomalies (Fig. 3). Strong correlations are found over the central and eastern equatorial Pacific, indicating a close linkage with ENSO, consistent with its well-documented influence on East Asian climate (Zhao et al., 2022; Xie et al., 2022). The correlation coefficients between PC2 and the Niño 1+2, Niño 3, and Niño 3.4 indices are 0.53, 0.51, and 0.51, respectively (all significant at the 95 % confidence level). Significant correlations are also identified between PC2 and SST anomalies in the Southern Hemisphere, resembling the pattern associated with the Antarctic Oscillation (AAO). The correlation coefficient between PC2 and the AAO index reaches -0.48 (P<0.05). To isolate the ENSO contribution, partial correlation analysis was performed by removing the ENSO signal with Niño 1+2 index, which exhibits the strongest correlation with PC2. After excluding the ENSO signal, the SST-PC2 relationship over the Northern Hemisphere becomes statistically insignificant (Fig. 3c). Consistently, the partial correlation between PC2 and the AAO index decreases to -0.22 and is no longer significant (P>0.1), attributed to the strong coupling between ENSO and the AAO (Han et al., 2017). Although Arctic sea ice has been reported to influence PM_2.5 pollution in China (An et al., 2023; Yin et al., 2021; Zhang et al., 2025), no significant correlation is detected between Arctic sea ice and PC2 (Fig. S2). In contrast, snow cover over the northern Tibetan Plateau becomes significantly correlated with PC2 after the ENSO signal is removed (Fig. 3d), suggesting that Tibetan Plateau snow anomalies exert an additional and independent modulation of the north-south dipole pattern of winter PM_2.5 over eastern China.

As noted above, PC2 exhibits the strongest correlation with the Niño 1+2 index and also shows statistically significant correlations with other Niño indices. El Niño-related warming of the tropical Pacific can trigger a Rossby wave propagating into East Asia. This induces a strong negative and weak positive geopotential anomaly over South and North of China, respectively, which closely resembles the wave train regressed by PC2 (Fig. S3). Figure 4 illustrates the regression patterns of meteorological conditions onto the normalized Niño 1+2 index. During strong El Niño winters, positive geopotential height anomalies emerge over the western Pacific (Fig. 4a), favoring southwesterly flow over southern China and southeasterly flow over northern China. As a result, moisture transport from adjacent oceans into southern China is enhanced, leading to a significant increase in relative humidity exceeding 2 % over the PRD (Fig. 4b) and increased precipitation of approximately 0.4 mm per month (Fig. 4c). The more humid conditions promote wet scavenging of aerosols, thereby reducing PM_2.5 concentrations in southern China. In contrast, over northern China, the anomalous southeasterly winds, which are opposite to the prevailing northwesterly flow, partly weaken near-surface wind speeds (Fig. 4d). Similar meteorological conditions are also found for Niño 3.4 index (Fig. S4). This meteorological contrast induced by El Niño supports the observed north-south dipole pattern of winter PM_2.5 over eastern China identified in EOF2.

To further verify the role of El Niño-like SST anomalies, sensitivity simulations were conducted by imposing warm SST anomalies associated with the Niño 1+2 index (Fig. S5). Figure 5a shows the differences between the CESM_ElNiño and CESM_ctrl experiments, which isolate the atmospheric and PM_2.5 response to El Niño forcing. Consistent with the statistical analysis, the imposed El Niño-like SST pattern strengthens the western Pacific subtropical high, induces anomalous southerly winds over eastern China, and enhances moisture transport into southern China (Fig. S6a, b). Consequently, surface PM_2.5 concentrations decrease markedly in southern China, with reductions of approximately 12 µg m⁻³ over the Sichuan Basin and 6 µg m⁻³ over the Yangtze River Delta (Fig. 5a). In contrast, PM_2.5 concentrations increase slightly over northeastern China by about 3 µg m⁻³, further confirming the El Niño-driven north-south dipole structure captured by EOF2.

After excluding the ENSO signal, PC2 remains significantly correlated with snow cover over the northern TP (Fig. 3d), indicating that interannual variability in winter PM_2.5 over China is also modulated by TP snow anomalies. To quantify this relationship, we define a Tibetan Plateau Snow Cover (TPSC) index as the wintertime area-averaged snow cover over the northern TP (86–94° E, 35–40° N), described by the blue box in Fig. 3d. The TPSC index associated snow albedo exhibits pronounced interannual variability throughout 1979–2021, with notably elevated values in recent years (Fig. S7). The correlation coefficient between the Niño 1+2 index and the TPSC index is -0.28 (p>0.24), indicating that the linear dependence between the two predictors is weak and statistically insignificant. To formally evaluate the degree of multicollinearity among PC2, Niño 1+2, and TPSC, we computed the Variance Inflation Factor (VIF) for each variable. The resulting VIF values are 1.87, 1.90, and 1.47, respectively, all well below the commonly adopted threshold of 5 (O'Brien, 2007), confirming that multicollinearity does not pose a meaningful threat to the robustness of the partial correlation results. Partial correlation analysis reveals a significant positive correlation between the TPSC index and PC2 (r=0.49, p<0.05). In contrast, no statistically significant relationship emerges when a domain-wide TP snow cover index is employed, suggesting that the dynamically relevant snow signal is regionally confined to the northern TP rather than reflecting a plateau-wide forcing.

Figure 6 illustrates anomalies in surface heat fluxes and the total atmospheric column heat source regressed onto the TPSC index. Enhanced snow cover over the northern TP increases snow albedo by approximately 0.02, leading to an increase in upward shortwave radiation exceeding 4 W m⁻² through the snow-albedo feedback. Meanwhile, a reduction of about 3 W m⁻² in upward longwave radiation is observed over the northern TP, driven by lower surface temperatures associated with reduced short-wave radiation absorption by the surface. In contrast, anomalies in sensible and latent heat fluxes remain weak (<1 W m⁻²). As a consequence, a pronounced cooling effect on the atmospheric column is found over the northern TP, exceeding 5 W m⁻².

The cooling effect over TP modulates the meridional temperature gradient and weakens the subtropical westerly jet north of the Plateau (Fig. 7a). Associated with enhanced snow cover, 200 hPa zonal wind speeds are reduced by up to 15 m s⁻¹, inducing an anomalous anticyclonic circulation over northern China (Fig. 7b, c). Under the influence of positive geopotential height anomalies, near-surface wind speeds decrease by more than 0.1 m s⁻¹, and the planetary boundary layer height is reduced by approximately 15 m. These stagnant conditions favor pollutant accumulation, while concurrently elevated surface relative humidity enhances aerosol hygroscopic growth (Fig. 7d), together contributing to increased PM_2.5 concentrations over northern China.

Because TP snow cover is itself influenced by local temperature, precipitation, and large-scale circulation variability, the statistical relationship may overestimate its impact on meteorological conditions and PM_2.5. To assess the causal impact of TP snow cover, we performed a targeted CESM experiment (CESM_TPSC) in which surface albedo over the northern TP (86–94° E, 35–40° N) was prescribed with a minimum value of 0.8. Surface albedo represents the reflectivity of the land surface, integrating contributions from all surface components including snow, soil, and vegetation. A minimum value of 0.8 was imposed to simulate persistent high snow cover conditions, consistent with the characteristically high reflectivity of snow-dominated surfaces (Cohen and Rind, 1991). The CESM results reproduce a coherent cooling response over the northern TP and a corresponding weakening of the subtropical westerly jet, which leads to increases in surface relative humidity and reductions in both planetary boundary layer height and near-surface wind speed over northern China (Fig. S6), supporting positive PM_2.5 anomalies over the BTH (Fig. 5b). These modeled responses are consistent with the observational regression analysis, reinforcing the physical credibility of the identified snow-forcing mechanism.

The individual impacts of El Niño and the TP snow cover on winter PM_2.5 in China have been confirmed, however, whether these two factors exert synergistic effects remains unclear. Following Li et al. (2019), we disentangle their respective and combined influences on circulation patterns associated with PM_2.5 pollution. We focus on atmospheric circulation rather than PM_2.5 concentrations directly, as meteorological variables are available over a longer period (1979–2021), whereas PM_2.5 observations are limited to 2005–2021.

Composite geopotential height anomalies for different combinations of Niño 1+2 and TPSC phases are shown in Fig. 8. Following Li et al. (2019), the symbol “&” denotes years when two indices occur simultaneously in the specified phase (e.g., “Niño 1+2⁺&TPSC⁺” indicates years when both Niño 1+2 and TPSC are positive), whereas the symbol “\” denotes years when only the first index is in the specified phase while the second index is not (e.g., “Niño 1+2+\TPSC⁺” indicates years when Niño 1+2 is positive but TPSC is neutral or negative). During Niño 1+2⁺&TPSC⁺ (Niño 1+2⁻&TPSC-) years, strong positive (negative) geopotential height anomalies develop over northern China and the western Pacific, accompanied by negative (positive) anomalies over the TP. This circulation configuration strongly favors the north-south dipole pattern of winter PM_2.5 over eastern China (Fig. 8a, b).

By contrast, during Niño 1+2+\TPSC⁺ or TPSC+\Niño 1+2⁺ years, similar circulation structures emerge but with substantially weaker anomaly amplitudes (Fig. 8c, e). Notably, during Niño 1+2+\TPSC⁺ years, El Niño also induces a negative geopotential height anomaly located over TP (Fig. 8c), similar to the impacts of TP snow (Fig. 7a). Previous studies confirmed that winter El Niño has positive impacts on TP snow by increasing storm activity and resultant snowfall (Shaman and Tziperman, 2005), which reinforce snow-albedo feedbacks and strengthen the associated circulation response. This process amplifies both the spatial extent and magnitude of circulation anomalies in Niño 1+2⁺&TPSC⁺, indicating significant synergistic effects of El Niño and TP snow cover during their positive phases.

During Niño 1+2-\TPSC⁻ years, positive geopotential height anomalies are observed over Northeast Asia (Fig. 8d), which appear inconsistent with the regression results (Fig. 4a). This discrepancy arises because the Niño 1+2-\TPSC⁻ composite includes three years with neutral TPSC conditions and four years with positive TPSC anomalies (Table 1). As a result, the circulation pattern is partly influenced by positive TPSC signals and exhibits geopotential height features resembling those associated with TPSC+ years (Fig. 8e). This inconsistency reflects a methodological limitation related to the small sample size inherent in the phase-classification approach (Li et al., 2019). Although geopotential height anomalies during Niño 1+2⁻&TPSC⁻ years are stronger than those in &TPSC-\Niño 1+2⁻ years, suggesting that simultaneous negative phases of Niño 1+2 and TPSC may also interact synergistically to enhance the circulation response, the synergistic effect is more robust and dynamically coherent during their positive phases. Therefore, we primarily conclude that Niño 1+2 and TP snow cover interact synergistically when both are in their positive phases.