Decadal Breakdown of Northeast Pacific SST–Arctic Stratospheric Ozone Coupling

Abstract

Using multiple reanalysis datasets, this study investigates the decadal variability in the relationship between Northeast Pacific Sea surface temperature (SST) and Arctic stratospheric ozone (ASO), with a focus on the role of atmospheric dynamics in mediating this connection. A significant decadal shift is identified around the year 2000, characterized by a weakening of the previously strong negative correlation between January–February SST anomalies and February–March ASO. Prior to 2000 (1980–2000), warm SST in the northeastern Pacific suppressed upward planetary wave propagation, resulting in decreased stratospheric wave activity and a weakened Brewer–Dobson circulation. The weakened BD circulation reduced poleward transport of tropical ozone and heat, yielding a colder, ozone-poor polar vortex. The strong relationship enabled skillful seasonal predictability of ASO using SST precursors in a linear regression model. However, post-2000 (2001–2022), the weakened planetary wave response to SST anomalies resulted in a breakdown of this relationship, yielding non-significant predictive skill. The findings highlight the non-stationary nature of ocean-stratosphere coupling and underscore the importance of accounting for such decadal shifts in climate models to improve projections of Arctic ozone recovery and its surface climate impacts.

1. Introduction

Stratospheric ozone plays a dual critical role in the earth system, serving both as a protective shield against harmful ultraviolet radiation for life [1,2] and as a key modulator of global climate through radiative-dynamic coupling [3]. This coupling mechanism enables stratospheric ozone variability to exert downward influences on surface weather extremes [4,5,6,7,8,9,10,11,12,13,14,15]. For instance, the recent record extreme hot Australian summer of 2012/13 was linked to anomalously high ozone levels in the preceding November [8], while ozone depletion has been associated with the intensification of austral summer precipitation extremes [7]. Given the significant impact of stratospheric ozone on Earth’s climate, numerous studies have investigated its variability and explored the factors and processes that influence ozone concentrations [16,17,18,19].

Extensive research has established that stratospheric ozone variability is governed by both anthropogenic forcing and natural climate variability. Key anthropogenic factors include ozone-depleting substances (ODS) [20,21,22] and greenhouse gases (GHGs) [21,23], while natural modes of variability encompass the El Niño-Southern Oscillation (ENSO) [17,24,25], Quasi-Biennial Oscillation (QBO) [26,27], Indian Ocean Dipole (IOD) [28,29], and Pacific Decadal Oscillation (PDO) [30]. The pronounced global ozone decline observed prior to 1998 was primarily attributable to anthropogenic emissions. Specifically, ODS undergo photo-dissociation upon reaching the stratosphere under intense UV radiation, releasing chlorine and bromine radicals that catalytically destroy ozone molecules through chain reactions [31,32]. The implementation of the Montreal Protocol successfully averted uncontrolled global ozone depletion and consequent dramatic increases in surface UV-B radiation [33,34]. In recent years, ozone has begun to recover, but the trends in ozone concentration vary by altitude. While the upper stratosphere has demonstrated robust recovery trends [35,36], the lower stratosphere (60°S–60°N) has exhibited persistent ozone decreases since 1998 [37,38]. These complex spatial patterns likely reflect the dominant influence of internal climate variability. For instance, ENSO modulates tropical lower stratospheric ozone concentrations, particularly during boreal summer [39]. During western QBO phases, double peaks of positive ozone anomalies are observed in the tropics, accompanied by negative anomalies in the extra-tropics [27].

Although Arctic stratospheric ozone depletion is not as severe as in Antarctica, the variability in ozone loss has always been much greater in the Arctic due to variability in wave forcing [40]. It is well known that Arctic ozone has been depleting, but since the signing of the Montreal Protocol, it began to slowly recover after the 1990s. It is projected that Arctic ozone will return to 1980 levels by 2040 [41]. In addition to trend changes, inter-annual variability has been a major focus of current research and discussions, as recent studies have shown that inter-annual variations in Arctic ozone significantly influence surface weather and climate in the Northern Hemisphere [9,10,11,12,13,42,43,44,45,46]. Understanding the inter-annual variability of Arctic ozone and its potential influencing factors is of great practical importance. The inter-annual variations in polar ozone are intrinsically linked to stratospheric polar vortex dynamics. Previous studies have extensively examined the relationship between the Arctic polar vortex and Pacific sea surface temperature (SST) [47,48,49,50,51]. For instance, weak stratospheric polar vortex events occur more frequently during positive PDO phases compared to its negative phases [50,51], with approximately 25% of the polar vortex strength attributable to central North Pacific SST [48]. However, research on ozone-SST relationships remains relatively scarce, with limited studies suggesting that North Pacific warming contributes to Arctic lower stratospheric ozone depletion [52,53]. While some researchers pointed out that negative SST anomalies in the central North Pacific could enhance the upward propagation of planetary wavenumbers 1 and 2, weakening the Brewer–Dobson circulation and subsequently reducing ozone concentrations in the Arctic lower stratosphere [54].

Despite previous studies having explored Arctic stratospheric ozone variability and its potential drivers, few have discussed the decadal relationship between Arctic ozone concentration and SST in the Pacific. Recently, Wang et al. (2023) examined decadal changes in the connection between Arctic stratospheric ozone and SST responses [55]. It is worthwhile to explore whether there is a decadal relationship between Arctic stratospheric ozone and preceding SST. This study aims to characterize the decadal relationship between Arctic stratospheric ozone and Pacific SST variability, to identify key connection regions, and to elucidate the underlying physical mechanisms governing these relationships.

2. Data and Methods

This study employs three ozone datasets to ensure robust analysis: the monthly Stratospheric Water and Ozone Satellite Homogenized (SWOOSH) dataset at 2.5° longitude resolution, which provides a merged record of stratospheric ozone and water vapor measurements from limb sounding and solar occultation satellites and comprises data from the SAGE-II/III, UARS HALOE, UARS MLS, and Aura MLS instruments spanning from 1984 to present [56]; the monthly Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2) ozone dataset with a uniform 2.5° latitude × 2.5° longitude spatial grid, produced by NASA’s Global Modeling and Assimilation Office (GMAO), which assimilates Solar Backscatter Ultraviolet Radiometer (SBUV) partial column ozone retrievals (1980–2004) and later incorporates Aura Microwave Limb Sounder (MLS) ozone profiles and Ozone Monitoring Instrument total column ozone (2004–present) [57]; and the monthly fifth-generation European Centre for Medium-Range Weather Forecasts (ERA5) reanalysis provided at 2° latitude × 2° longitude resolution, which uses Level 2 ozone from multiple instruments [58]. Additionally, atmospheric variables—including geopotential height, wind, and temperature—are also derived from ERA5 with a uniform 1° latitude × 1° longitude spatial grid, while monthly SST data are obtained from the Met Office Hadley Centre’s HadISST dataset on a 1° latitude × 1° longitude global grid [59]. In this study, we mainly focus on the period of 1980–2022.

The propagation of wave activity uses the following formulas according to Andrews et al. (1987) [60]. The meridional (F_y) and vertical (F_z) components of the Eliassen–Palm (EP) flux and the EP flux divergence (D_F) are expressed as follows:

$$F_y={\rho }_{0}a\cos \phi ({\overline{u}}_z\overline{v^{\prime }{\theta }^{\prime }}/{\overline{\theta }}_z-\overline{u^{\prime }{v}^{\prime }})$$

(1)

$$F_z={\rho }_{0}a\cos \phi \{[f-{\left(a\cos \phi \right)}^{-1}{\left(\overline{u}\cos \phi \right)}_{\phi }]\overline{v^{\prime }{\theta }^{\prime }}/{\overline{\theta }}_z\}$$

(2)

$$D_F={\displaystyle \frac{\nabla ·F}{{\rho }_{0}a\cos \phi }}={\displaystyle \frac{\partial \left(F_y\cos \phi \right)/a\cos \phi \partial \phi +\partial F_z/\partial z}{{\rho }_{0}a\cos \phi }}$$

(3)

where ρ₀ is the density of air; a is the radius of Earth; φ is the latitude; f is the Coriolis parameter; u and υ are the zonal and meridional components of the wind, respectively; and $\theta$ is the potential temperature. The overline denotes the zonal mean. Superscript symbols indicate deviations from the zonal mean.

The EP flux indicates the direction of wave propagation, quantifying planetary wave activity in the stratosphere. Increased wave propagation from the troposphere causes stronger EP flux convergence in the stratosphere. The EP flux convergence drives a stronger Brewer–Dobson circulation via downward control arguments.

We employed standard linear regression models to examine the relationships between preceding SST and ASO. To determine the robustness of our linear model in predicting ASO and sidestep the issue of data overfitting. A leave-one-out cross-validation method is employed [9,61]. Specifically, a time point is selected from a time series of length N as a hindcasting target point, while the remaining (N–1) points are treated as training points to develop the linear model. Repeat the above procedure until each point in the time series has served as the hindcasting target point. Consequently, a set of N predictive models will be constructed, culminating in an ensemble hindcast.

Throughout our study, we exclusively employed Pearson’s correlation coefficients to assess linear relationships between variables. Given the relatively small sample size in our analysis, we employed a nonparametric bootstrap approach to assess the statistical significance of Pearson’s correlation coefficients [62]. We performed 1000 bootstrap resamples to generate robust confidence intervals while maintaining computational efficiency. The strength of correlation was determined by examining the 95% confidence interval derived from these resampled coefficients: if the interval excluded zero and the original correlation coefficient fell within this range, the correlation was considered statistically significant (p < 0.05, strong correlation coefficients). Conversely, if the 95% confidence interval included zero, the correlation was deemed weak or moderate.

3. The Weakening Relationship Between the Arctic Ozone and Northeast Pacific SST Since the 2000s

To investigate the ozone variability, ozone concentrations in ppmv and Dobson Units (DU) derived from the SWOOSH dataset are analyzed. Figure 1 presents the climatology of zonal-mean ozone mixing ratios (ppmv) and their corresponding DU values (Figure 1a,b). The results for the ozone mixing ratio (ppmv) reveal a pronounced maximum centered in the tropical stratosphere near 10 hPa, with values decreasing progressively toward higher latitudes and altitudes (Figure 1a). Notably, the standard deviation is higher in the lower tropical stratosphere and the lower stratosphere in the polar regions (Figure 1c). Conversion to DU yields distinct spatial patterns compared to mixing ratios. The DU distribution exhibits an arch-like structure, with peak ozone occurring in the Arctic lower stratosphere (Figure 1b). Maximum variability is observed between 30 and 150 hPa over both polar regions (Figure 1d). These findings demonstrate that the Arctic lower stratosphere contains not only the highest DU-based ozone but also exhibits substantial variability. This large Arctic ozone variability has important implications, as growing evidence suggests that Arctic stratospheric ozone anomalies can significantly influence surface weather and climate across northern mid-to-high latitudes [9,10,11,12,13,42,43,44,45,46].

To further characterize the evolution of Arctic stratospheric ozone, we define an Arctic Stratospheric Ozone (ASO) index. This index is computed as the area-weighted average over the polar cap region (60–90°N) and incorporates mass-weighting through vertical integration across the 1–150 hPa pressure. Figure 2a presents the time series of ASO anomalies derived from three independent datasets (SWOOSH, MERRA2, and ERA5), which shows remarkable consistency. The records reveal a pronounced ozone depletion trend prior to 1997, consistent with increasing atmospheric concentrations of ODS during this period, followed by a stabilization after the implementation of the Montreal Protocol. To exclude the influences from the long-term ODS-driven trend, Figure 2b displays the detrended ASO anomalies, which exhibit substantial seasonal variability. To further investigate these seasonal differences, Figure 2c presents the standard deviation of the detrended ASO anomalies from January to December. The analysis reveals maximum variability during winter months (December–March), with peak values occurring in February and March. This seasonal variability likely reflects enhanced dynamical variability associated with a stronger polar vortex during winter in early spring [63]. The pronounced winter/spring variability carries particular significance, as numerous studies have demonstrated that late winter/early spring Arctic ozone anomalies can influence tropospheric temperature and East Asian precipitation variability [12,13].

Figure 3 presents the February–March ASO anomaly time series. Similarly to the annual dataset, a significant ozone depletion trend is evident before 1997, while the post-1997 period shows continued but weaker depletion. The detrended anomalies in Figure 3b exhibit pronounced inter-annual variability, with amplitude exceeding 20 DU during extreme years. Previous studies have established decadal linkage between ozone variability and SSTs [55]. They have documented a weakened connection between March Arctic ozone and April North Pacific SST since 2000. This raises the question of whether the influence of preceding SST on Arctic ozone also exhibits inter-decadal variations. To address this issue, and considering that the impact of SST on the polar circulation may have a lead of about one month, Figure 3c,d displays the correlation between the detrended February–March ASO anomalies and January–February SST anomalies, with the pre-2000 period (1980–2000) showing a significant negative correlation over the Northeast Pacific, while no significant positive correlation is observed post-2000. We further examine the relationship through 21-year running correlation between January–February SST and February–March ASO (Figure 3e), which reveals a significant shift in their correlation structure around the year 2000. This suggests that the influence of SST on late winter or early spring ozone also exhibits decadal changes. To further analyze this relationship, the time series of detrended February–March ASO anomalies and January–February SST anomalies, averaged over 20–40°N and 105–127°W (SST__NEP), are shown in Figure 4a,b. It is clear that they are significantly negatively correlated before 2000, with a correlation coefficient of −0.57, while after 2000, no significant correlation is observed. Figure 4c,d present scatter plots of standardized detrended ASO versus SST__NEP anomalies. Similarly to Figure 4a,b, SST__NEP and ASO show a significant negative correlation before 2000, but no significant linear correlation is found post-2000.

4. Diagnosing the Possible Mechanisms Behind the Decadal Changes

To further investigate the physical processes through which SST influences ASO variability, Figure 5 presents the correlation analyses between the SST__NEP index and zonal-mean zonal winds at 500 hPa and 200 hPa for both pre- and post-2000 periods. During the pre-2000 era, significant negative correlations prevailed across mid-latitudes at both pressure levels, indicating a pronounced weakening of westerly winds that would have suppressed vertical propagation of planetary waves into the stratosphere (Figure 5a,c). This large-scale circulation response would favor a stronger, more persistent polar vortex, creating conditions that inhibit the meridional transport of ozone from tropical to polar regions, consistent with the strong negative ASO-SST relationships observed during this era. In contrast, the post-2000 correlations exhibit a markedly different pattern, with negative correlations confined primarily to the North Pacific sector while positive correlations emerge across other mid-latitude regions. Such changes in the circulation field likely have a weaker impact on wave propagation, corresponding with the documented weakening of ASO-SST relationships (Figure 4b). These results demonstrate that, while pre-2000 SST variability exerted a strong hemispheric-scale influence on circulation patterns, significantly affecting Arctic ozone, the post-2000 regime has shifted to a state where SST forcing produces more limited and localized dynamical responses, with a reduced potential to influence stratospheric ozone variability.

Furthermore, to better understand the process through which planetary waves forced by tropical SST anomalies modulate polar ozone, we employed EP flux to examine stratospheric wave activity propagation. The stratospheric polar vortex variability is intrinsically linked to vertically propagating planetary waves originating from tropospheric sources. Figure 6 presents daily computed planetary wave responses between positive and negative SST__NEP events, revealing distinct decadal changes around 2000. During the pre-2000 period (Figure 6a), we observed reduced upward wave propagation at 60°N accompanied by EP flux divergence in the high-latitude stratosphere (10–30 hPa), a configuration that would cool and strengthen the northern polar vortex by reducing Brewer–Dobson circulation-induced adiabatic heating. In contrast, the post-2000 period shows weaker EP flux convergence in the polar stratosphere, with enhanced equatorward EP flux transport from high latitudes but no significant changes in vertical wave (Figure 6b). Further decomposition into stationary and transient wave components (Figure 7) reveals more complex dynamics. The pre-2000 period demonstrates fewer planetary waves transported upward into the stratosphere in both stationary and transient wave at mid-latitudes, though with opposing signs in EP flux divergence (Figure 7a,c), indicating that the stratospheric EP flux propagation and divergence were jointly influenced by both wave types. After 2000, similar to the overall changes in EP flux (Figure 6b), the EP flux did not exhibit significant vertical changes, while the divergence (Figure 7b) and convergence (Figure 7d) counteracted each other, leading to a weaker response in the total EP flux divergence (Figure 6b). To quantitatively assess the relative contributions of stationary waves and transient waves, we calculated the vertical flux averaged over the 45–75°N region at 50 hPa. The results show that during 1980–2000, stationary waves accounted for 68% of the total wave activity, while transient waves contributed 32%. In contrast, during 2001–2022, stationary waves only accounted for 38%, with transient waves dominating at 62%. Notably, the amplitude of vertical wave activity post-2000 was significantly weaker, measuring only 13% of the pre-2000 level. These analyses demonstrate that January–February Northeast Pacific SST anomalies can significantly modulate stratosphere-entering planetary waves through combined stationary and transient wave effects prior to 2000, while the post-2000 period exhibits markedly weaker planetary wave responses to SST variability.

These results demonstrate distinct decadal shifts in the Northeast Pacific SST’s influence on stratospheric dynamics. During the pre-2000 period, positive SST anomalies in the northwestern Pacific effectively suppressed upward planetary wave propagation (Figure 6a), leading to a cooler and stronger polar vortex (Figure 8a,c). This dynamical response established atmospheric conditions that effectively suppressed the poleward transport of both ozone and heat from tropical regions. The chain of physical processes—from suppressed EP flux (Figure 6a) to vortex strengthening (Figure 8c) and ultimately to ozone loss—explains the robust negative ASO-SST relationships observed during this era. In contrast, the post-2000 period shows markedly different behavior. The weakened planetary wave response to SST variations (Figure 6b) corresponds with non-significant changes in stratospheric temperatures and circulation patterns (Figure 8b,d,f). This decoupling suggests fundamental changes in the ocean-atmosphere-stratosphere interaction pathways, where Northeast Pacific SST anomalies no longer effectively modulate the stratospheric polar vortex through wave-mean flow interactions. The absence of coherent dynamical responses helps explain the breakdown of ASO-SST relationships in recent decades. These findings highlight the importance of considering non-stationarity in stratosphere–troposphere coupling when assessing long-term ozone variability and its climate connections.

To quantitatively assess the predictive potential of northwestern Pacific SST anomalies for Arctic ozone variability, we developed a linear regression model with January–February SST__NEP as the predictor for February–March polar ozone intensity. The pre-2000 period demonstrates robust predictive skill, with the full-period (1980–2000) regression model yielding a significant correlation coefficient of 0.57 (p < 0.05) between fitted and observed ASO indices (Figure 9a). The model’s stability was further verified through a split-sample approach: when trained on 1980–1994 data, the model maintained strong skill with a correlation of 0.56 (p < 0.05) for the training period and even higher skill (r = 0.62, not significant due to small sample) during the independent test period (1995–2000, Figure 9c), demonstrating the temporal robustness of this relationship. In stark contrast, the post-2000 period shows markedly reduced predictive capability. The full-period (2001–2022) regression model achieves only marginal skill (r = 0.24, Figure 9b), while the split-period analysis reveals similarly limited performance (training period 2001–2016: r = 0.24; prediction period 2017–2022: r = 0.33). As shown in Figure 9e,f, the observed ASO index (blue line, MERRA-2) and predicted values based on the leave-one-out cross-validation method (green line) also demonstrate strong agreement during 1980–2000, with a statistically significant correlation coefficient of 0.47. However, this relationship becomes insignificant during 2001–2022. These findings are fully consistent with the results shown in Figure 9a–d. This substantial decline in predictive skill corresponds with the documented weakening of dynamical connections between Pacific SST and Arctic stratospheric variability, further confirming the decadal shift in stratosphere–troposphere coupling around 2000.

5. Conclusions and Discussion

This study analyzed the influence of Northeast Pacific SST anomalies from late winter (January–February) on the inter-annual variations in Arctic stratospheric ozone in late winter and early spring (February–March) using observations. It was found that there is a weakening in the connection between Northeast Pacific SST and ASO, marked by a distinct shift around 2000. Prior to 2000, the January–February SST are strong negative correlated with February–March ASO anomalies (r = −0.57), while this relationship became statistically insignificant in the subsequent two decades. The weakening of this linkage likely stems from significant alterations in atmospheric wave propagation, which modulate stratospheric ozone concentrations.

The underlying physical mechanism involves SST-induced modulation of stratospheric wave driving. In the pre-2000 period, warm SST anomalies in the northwestern Pacific effectively suppressed upward planetary wave propagation, weakening Brewer–Dobson circulation and reducing poleward transport of tropical ozone and heat. These dynamical processes resulted in robust negative correlations between January–February SSTs and subsequent February–March ASO anomalies, enabling skillful seasonal predictability in regression models. However, post-2000, this teleconnection weakened significantly due to altered atmospheric wave responses, with SST anomalies producing only localized circulation changes that failed to coherently modulate stratospheric dynamics. Consequently, the predictive power of SST precursors diminished, highlighting the non-stationary nature of ocean-stratosphere coupling. The observed shift in SST-ASO relationships, potentially linked to PDO or anthropogenic climate change, carries important implications for understanding Arctic ozone changes and improving seasonal forecasting frameworks in a changing climate. The breakdown of the previously robust relationship suggests that climate models capable of capturing such non-stationary behavior may achieve more accurate projections of future stratospheric ozone trends and their impacts on surface climate.