Journal of Geophysical Research: Atmospheres

On the influence of North Pacific sea surface temperature on the Arctic winter climate

Authors


Corresponding author: M. M. Hurwitz, NASA Goddard Space Flight Center, Code 614, Greenbelt, MD 20771, USA. (margaret.m.hurwitz@nasa.gov)

Abstract

[1] Differences between two ensembles of Goddard Earth Observing System Chemistry-Climate Model simulations isolate the impact of North Pacific sea surface temperatures (SSTs) on the Arctic winter climate. One ensemble of extended winter season forecasts is forced by unusually high SSTs in the North Pacific, while in the second ensemble SSTs in the North Pacific are unusually low. High – Low differences are consistent with a strengthened Western Pacific atmospheric teleconnection pattern, and in particular, a weakening of the Aleutian low. This relative change in tropospheric circulation inhibits planetary wave propagation into the stratosphere, in turn reducing polar stratospheric temperature in mid- and late winter. The number of winters with sudden stratospheric warmings is approximately tripled in the Low ensemble as compared with the High ensemble. Enhanced North Pacific SSTs, and thus a more stable and persistent Arctic vortex, lead to a relative decrease in lower stratospheric ozone in spring, affecting the April clear-sky UV index at Northern Hemisphere midlatitudes.

1. Introduction

[2] The severity of Arctic ozone depletion is highly dependent on the evolution of polar lower stratospheric temperature in late winter and spring [World Meteorological Organization, 2011]. In 2011, unprecedented Arctic ozone depletion resulted from a sustained period of below-average temperatures and an extremely isolated polar air mass [Manney et al., 2011]. The cold 2010–2011 winter resulted in a large volume of polar stratospheric clouds (PSCs), and thereby, high levels of activated chlorine that catalytically destroyed ozone in the Arctic lower stratosphere [Manney et al., 2011]. It is important to understand the causes of these 2011 conditions, in order to assess how Arctic ozone and ultraviolet (UV) radiation levels will likely evolve as the abundance of ozone-depleting substances declines.

[3] Hurwitz et al. [2011a] considered possible dynamical causes of the 2011 ozone depletion event. March was the focus of the authors' study, because conditions in the Arctic stratosphere were particularly anomalous during that month. Several causes were rejected: Direct radiative cooling by greenhouse gases was dismissed because cooling of the Arctic lower stratosphere observed during the satellite era (0.17 ± 0.14 K year−1, in the MERRA reanalysis) was too weak to explain the roughly 10 K cooling in 2011. Both El Niño/Southern Oscillation (ENSO) and the quasi-biennial oscillation (QBO) modulate the strength of the Arctic vortex but could not explain the anomalous stratospheric cooling in 2011. Relative strengthening of the Arctic vortex during La Niña events, as observed in 2011, weakens and begins to reverse by March. Similarly, the authors found that relative strengthening of the mid-winter vortex during the westerly phase of the QBO, relatively to the easterly phase, does not persist through March. Furthermore, the structure and magnitude of dynamical anomalies in the Arctic stratosphere were similar in March 1997 and March 2011, two episodes of low temperatures and large ozone losses, despite different phases of the QBO.

[4] Hurwitz et al. [2011a]showed that winters with exceptionally high North Pacific SSTs, including 2010–2011, were often characterized by weak planetary wave driving in mid-winter and by cold, persistent Arctic vortices, providing the conditions necessary for severe polar ozone loss. Similarly,Jadin et al. [2010]found positive correlations between mid-winter North Pacific SSTs and the strength of the Arctic vortex. North Pacific SSTs may affect planetary wave driving by modifying the large-scale circulation. The West Pacific (WP) teleconnection pattern is characterized by positive upper tropospheric ridges in the western and central North Pacific [Wallace and Gutzler, 1981]. Strong, positive WP pattern events tend to inhibit planetary wave driving and strengthen the Arctic vortex in winter [Garfinkel and Hartmann, 2008; Orsolini et al., 2009; Woollings et al., 2010; Garfinkel et al., 2010]. North Pacific ridges are an efficient way to reduce planetary wave driving in the Northern Hemisphere because they destructively interfere with the climatological stationary wave pattern [Garfinkel and Hartmann, 2008; Orsolini et al., 2009; Nishii et al., 2010; Garfinkel et al., 2010; Nishii et al., 2011]. Specifically, ridges in the North Pacific (i.e., strengthening of the WP pattern) have been shown to precede prolonged polar stratospheric cooling of up to 6 K [Nishii et al., 2010] and months with very high PSC volumes [Orsolini et al., 2009].

[5] While observational evidence suggests that North Pacific SSTs can affect polar stratospheric conditions, the Arctic winter climate is highly variable and few anomalous cooling events have occurred during the satellite era. Thus, it is not possible to attribute the dynamical cause(s) of a particular event, such as the unusual meteorology and ozone loss observed in 2011. To pinpoint the role of North Pacific SSTs on the Arctic troposphere and stratosphere in winter, the present study compares two ensembles of chemistry-climate model (CCM) simulations forced by composites of observed SSTs, each providing many samples of the atmospheric response to high or low North Pacific SSTs.Section 2describes the model and experimental set-up.Section 3 will diagnose ensemble mean differences in Arctic geopotential height, eddy heat flux, temperature, ozone and UV index. Also, Section 3 will assess the frequency of strong eddy heat flux events and major sudden stratospheric warmings (SSWs), characterized by a reversal of the climatological westerly zonal winds at 60°N, 10 hPa, in the two ensembles. Section 4 will provide a discussion and summary.

2. Method

[6] The Arctic winter response to changes in North Pacific SSTs is simulated using the Goddard Earth Observing System Chemistry-Climate Model, Version 2 (GEOS V2 CCM). The GEOS V2 CCM couples the GEOS-5 general circulation model (GCM) with a comprehensive stratospheric chemistry module [Bloom et al., 2005; Pawson et al., 2008]. The model has 2° latitude × 2.5° longitude horizontal resolution and 72 vertical layers, with a model top at 0.01 hPa. Predicted distributions of water vapor, ozone, greenhouse gases (CO2, CH4, and N2O) and CFCs (CFC-11 and CFC-12) feed back to the radiative calculations. An earlier version of the GEOS V2 CCM generally performed well in theSPARC CCMVal [2010]detailed evaluation of stratospheric processes, though in the Arctic winter, lower stratospheric temperatures were warm-biased and thus the model tended to underestimate polar ozone loss [seeSPARC CCMVal, 2010, Figures 4.1 and 6.37]. The present formulation of the GEOS V2 CCM is as described by Hurwitz et al. [2011b]. The model formulation includes an updated GCM, with an improved representation of tropical stationary wave patterns, and a new gravity wave drag scheme that allows the model to produce an internally-generated quasi-biennial oscillation (QBO) with realistic periodicity and magnitude.

[7] This study compares two GEOS V2 CCM ensembles, each composed of 40 simulations of an extended Arctic winter season. Each of the 40 pairs of simulations is initialized independently, using a 1st October restart file from a perpetual ENSO neutral (ENSON) simulation [Hurwitz et al., 2011b], and is run through April 30th. The independent set of initial conditions, together with the model's internal QBO, imply that the simulated winter conditions sample the range of possible QBO phases. Both ensembles are forced by greenhouse gas and ozone-depleting substances representative of the 2005 climate (as inHurwitz et al. [2011b]).

[8] The two ensembles differ only by the SST and sea ice boundary conditions imposed north of 20°N. Annually repeating SST and sea ice climatologies are each constructed from the average of two observed winters when North Pacific (40–50°N, 160–200°E) SSTs were either unusually high (1990–1991 and 1996–1997; Figure 1a, blue stars) or unusually low (1986–1987 and 1987–1988; Figure 1a, cyan stars). Note that this experimental design tests the atmospheric sensitivity to forced changes of North Pacific SSTs, rather than the full non-linear interaction of the ocean–atmosphere system. South of 20°N, ENSO neutral SST and sea ice climatologies are prescribed in both ensembles (as in the ENSON simulation, described byHurwitz et al. [2011b]). Prescribing ENSO neutral conditions eliminates the impacts of El Niño and La Niña conditions on the simulated Arctic stratospheric response. SST and sea ice values are obtained from the HadISST1 data set [Rayner et al., 2003].

Figure 1.

(a) Time series of the January/February mean North Pacific SST index. Blue (cyan) stars denote winters used to construct the High (Low) SST and sea ice boundary conditions. The red star denotes the North Pacific SST index in 2011. (b) Time series of North Pacific SST anomalies through the extended winter season, for High – Low (black line) and 2010–2011 anomalies from the 1979–2011 climatology (red line). (c) January/February mean SSTs used as boundary conditions in the High ensemble. (d) January/February mean differences between the High and Low SST fields. White contours indicate zero difference. Black boxes in Figures 1c and 1d indicate the North Pacific region.

[9] Figure 1b shows the evolution of North Pacific SST differences throughout the winter. High – Low differences increase from 1 K in October to 2 K in January, and decrease back to 1 K by the end of each simulation. The seasonal evolution of the High – Low SST differences resembles that of the winter 2010–2011 North Pacific SST anomalies, though the magnitude of the imposed High – Low SST differences is approximately twice as large. Figure 1c shows January/February SSTs used as boundary conditions in the High ensemble. In the North Pacific, SSTs range from 5 to 8°C. Figure 1d shows January/February High – Low SST differences. Positive High – Low differences span the Pacific at midlatitudes, with negative SST differences along the west coast of North America and Alaska, and in the subtropical western Pacific. Note that the pattern of High – Low SST differences resembles the pattern of SST anomalies observed during the 2010–2011 winter (not shown).

[10] In order to interpret the High – Low differences, simulated stratospheric ozone and temperature fields are compared with observational data sets. Simulated total ozone is compared with the TOMS/SBUV data set (updated from Stolarski and Frith [2006]). Simulated temperature fields are compared with the Modern Era Retrospective-Analysis for Research and Applications (MERRA) reanalysis. The MERRA reanalysis is based on an extensive set of satellite observations and on the Goddard Earth Observing System Data Analysis System, Version 5 (GEOS-5) [Bosilovich, 2008; Rienecker et al., 2011]. The MERRA reanalysis has vertical coverage up to 0.1 hPa, and for this study, is interpolated to 1.25° × 1.25° horizontal resolution. Both the TOMS/SBUV and MERRA data sets span the 1979–2011 period.

3. Differences Between the High and Low Ensembles

3.1. Troposphere

[11] Figure 2shows High – Low monthly mean geopotential differences at Arctic latitudes in December and March. At 850 hPa, a region of positive geopotential height differences in the North Pacific, co-located with the region of increased SSTs (seeFigure 1d), develops in December (Figure 2a) and persists through the spring (Figure 2d). Positive geopotential height differences are also co-located with the climatological trough in the North Pacific (gray, dashed contours), suggesting that a relative increase in North Pacific SSTs tends to decrease tropospheric wave driving. A region of negative differences develops over the Arctic cap in December (Figure 2a), and persists through the spring (Figure 2d). In January and February, sea level pressure decreases at polar latitudes, and increases in the European and Atlantic sectors (Figure 3). This pattern of sea level pressure differences is consistent with a positive shift in the North Atlantic Oscillation (NAO). Specifically, the sea level pressure difference between 334°E, 38°N (Ponta Delgada, Azores) and 338°E, 64°N (Reykjavik, Iceland) is significantly different in the High and Low simulations at 95% confidence level in January.

Figure 2.

High – Low geopotential height differences (m) at (a and d) 850 hPa, (b and e) 300 hPa and (c and f) 50 hPa, in December (Figures 2a–2c) and March (Figures 2d–2f), poleward of 30°N. Note the different color scale for each pressure level. White contours indicate zero difference. Gray dashed contours show mean geopotential heights in the ENSO neutral simulation. In each panel, the black wedge indicates the North Pacific region. High – Low differences are in general statistically significant at the 95% level, in a two-tailed t-test.

Figure 3.

January/February High – Low sea level pressure differences (hPa), poleward of 30°N. White contours indicate zero difference. Black Xs indicate differences significant at the 95% level, in a two-tailed t-test.

[12] The tropospheric response to the North Pacific SST difference is structurally barotropic. Consistent with Frankignoul and Sennéchael [2007], upper tropospheric (300 hPa; Figures 2b and 2e) geopotential height differences mimic the near-surface (850 hPa;Figures 2a and 2d) geopotential height differences. From January through March, High – Low differences indicate a strengthening of the Arctic Oscillation: negative differences at Arctic latitudes, and positive differences at midlatitudes, particularly in the Pacific and Atlantic sectors.

[13] Eddy heat flux at 40–80°N, 100 hPa is a measure of the planetary wave activity entering the Arctic stratosphere. Time-integrated eddy heat flux in this region is highly correlated with polar lower stratospheric temperatures, with a 1–2 month lag [Newman et al., 2001]. Figure 4 shows the mean eddy heat flux in the preceding 45 days, in the High (blue (Figure 4, top)) and Low (cyan) ensembles, as well as the mean differences between the two ensembles (dotted black line (Figure 4, bottom)). High – Low differences are negative from early January through mid-February, suggesting a relative High – Low decrease in eddy heat flux in December and January; inFigure 4, statistical significance of mean differences is based on a two-tailed t-test. By early February, eddy heat flux in the Low ensemble is both larger and more variable than in the High ensemble; the significance of the difference in variability is based on an F-test. The sign of High – Low eddy heat flux differences flips in mid-April (Figure 4, bottom).

Figure 4.

(top) Eddy heat flux (K m s−1) at 40–80°N, 100 hPa in the preceding 45 days (i.e., cumulative eddy heat flux). Solid blue and cyan lines indicate the High and Low ensemble mean values. Dashed blue and cyan lines indicate the 10th- and 90th-percentile values for each ensemble. Black Xs below the plot indicate days when the High – Low eddy heat flux difference is significant at the 95% level, in a two-tailed t-test. Black Xs above the plot indicate days when the difference in variability between the High and Low ensembles is significant at the 95% level, in an F-test. (middle) Like Figure 4 (top) but for polar cap temperature (K) at 50 hPa. (bottom) High – Low ensemble mean differences in eddy heat flux (dotted black line) and temperature (solid black line).

[14] Not only do the 45-day mean eddy heat fluxes differ between the two ensembles, so does the frequency of strong wave events.Table 1 lists the total of days where the eddy heat flux at 40–80°N, 100 hPa is greater or equal to 25 K m s−1. Note that days when eddy heat flux is greater or equal to 25 K m s−1 represent approximately 5% of the simulated period. Strong eddy heat flux “events” are defined as the persistence of 40–80°N, 100 hPa eddy heat flux, greater or equal to 25 K m s−1, for two or more consecutive days. The total number of strong eddy heat flux events is larger in the Low ensemble than the High ensemble, as is the number of strong eddy heat flux days. In December and February, note that the number of strong heat flux days is roughly doubled in the Low ensemble as compared with the High ensemble.

Table 1. Number of Days When Eddy Heat Flux at 40–80°N, 100 hPa is Equal to or Greater Than 25 K m s−1, in December, January and February, in the High and Low Ensemblesa
 Number of Events TotalNumber of Days
TotalDecJanFeb
  • a

    Events are defined as the persistence of 40–80°N, 100 hPa eddy heat flux, greater or equal to 25 K m s−1, for two or more consecutive days. Totals represent the November–March extended winter season.

High143414898248
Low15349713891103

3.2. Polar Stratosphere

[15] Polar geopotential height at 50 hPa is a measure both of the strength of the Arctic vortex, and of the mean temperature below 50 hPa. Negative High – Low geopotential height differences over the polar cap (Figure 2f) persist from January through March. Negative geopotential height differences shift toward the Siberian sector in April. Polar cap temperature at 50 hPa is used to examine the seasonal evolution of the High – Low changes. The polar cap cools in the mid- to late winter, in the High ensemble relative to the Low ensemble (Figure 4, bottom). 50 hPa polar cap temperature differences are statistically significant in January and February, consistent with the negative eddy heat flux differences, and in late March (black and gray Xs below Figure 4(center)). Note that approximately 2/3 of the 40 pairs of simulated winters show a strong High – Low cooling of the polar stratosphere in March. Polar cap temperature differences reverse sign in mid-April (similar to the seasonal evolution of total ozone in 2010 and 2011 [Balis et al., 2011]). Early winter temperature differences are negligible.

[16] North Pacific SSTs modulate not only the mean state of the polar stratosphere, but also the distribution of possible winter conditions. In Figure 4(middle), blue (cyan) dashed lines indicate the 90th-percentile (fourth-highest) and 10th-percentile (fourth-lowest) temperature values of the High (Low) ensemble. The 10th-percentile values are similar in the two ensembles. However, the 90th-percentile values are significantly warmer (denoted by the Xs above the plot) in the Low ensemble than in the High ensemble in January and early February. This suggests that mid-winter polar stratospheric variability is enhanced when North Pacific SSTs are relatively lower.Section 3.3 will relate this result to the relative frequency of sudden stratospheric warmings (SSWs) in the two ensembles.

[17] The mean and distribution of seasonal mean January–February–March (JFM) polar cap temperature at 50 hPa differs robustly between the two ensembles. Figure 5a shows histograms of the JFM mean polar cap temperature; the mean High – Low difference is 1.91 K. Note that temperatures in the Low ensemble (cyan bars) have a tail toward high values.

Figure 5.

Histograms of the January-February-March (JFM) seasonal mean polar cap temperature (K) at 50 hPa: (a) High and Low ensembles, and (b) QBO-easterly years and QBO-westerly years. The black-tipped bars indicate the location of the ensemble mean values; the High and Low ensemble mean difference is significant at the 95% level, in a two-tailed t-test.

[18] In contrast, the JFM mean polar cap 50 hPa temperature is independent of the modeled QBO phase. Figure 5b shows histograms of the JFM polar cap temperature in both the High and Low ensembles, partitioned by QBO phase (i.e., the magnitude of JFM zonal winds at 30 hPa, between 10°S and 10°N, following Hurwitz et al. [2011b]). QBO-westerly years are characterized by JFM zonal winds greater than 2 m s−1, while QBO-easterly years are characterized by zonal winds less than −2 m s−1. The means of the QBO-W and QBO-E composites are statistically indistinguishable, in a two-tailed t-test. In their observational analysis of polar stratospheric conditions in March,Hurwitz et al. [2011a]found a similar lack of dependence on the QBO phase. Note that the QBO does modulate the strength of the Arctic stratosphere in early winter (i.e., the Holton-Tan relation [Holton and Tan, 1980]), in this model version. Also note the relative abundance of simulated QBO-E years, as compared with QBO-W.

[19] Polar stratospheric temperature anomalies descend from the upper stratosphere in mid-winter to the lower stratosphere in spring. The evolution of High – Low temperatures at 80°N through the extended winter season is summarized byFigure 6. Temperature differences are negligible in November through mid-December. By late December, relative cooling develops at and above 10 hPa. In January, relative cooling of the middle stratosphere exceeds 8 K. The anomalous cooling slowly descends, with weaker cooling extending to the 400-hPa level in February. Lower stratospheric cooling is strongest in March and early April. In March, relative cooling of 1–2 K extends to the surface layer, suggestive of downward stratosphere-troposphere coupling. Stratospheric cooling is followed by relative warming of 2–4 K, with an approximately two-month lag. The Arctic stratospheric temperature response to ENSO events has a similar structure [Manzini et al., 2006].

Figure 6.

High – Low temperature differences (K) at 80°N, as a function of date and altitude. White contours indicate zero difference. Black Xs indicate differences significant at the 95% confidence level, in a two-tailed t-test.

[20] The breakup of the Arctic vortex, as defined by the transition from mean westerlies to easterlies at 60°N, occurs later in the High ensemble than in the Low ensemble. At 10 hPa, the mean difference in the breakup date at 10 hPa is approximately 4 days (not shown). The relatively later breakup in the High ensemble is consistent with the relatively colder polar lower stratosphere in late winter (see Figures 4 through 6).

[21] Interannual variability of mid-winter North Pacific SSTs explains some of the observed interannual variability in polar lower stratospheric conditions in late winter.Figure 7 shows the observed February/March volume of polar stratospheric clouds (VPSC) time series for the 380–550 K potential temperature layer, updated from Rex et al. [2004, 2006], versus the January/February North Pacific SST index (as in Hurwitz et al. [2011a]) in the same winter. VPSC represents the volume of polar air that is below the formation threshold for polar stratospheric clouds (approximately 195 K at 50 hPa). Thus, February/March VPSC is a metric of the potential for late winter ozone depletion. The VPSCcorrelation with the January/February North Pacific SST index is greater for February/March (r = 0.38) than for the December–March average (r = 0.33; not shown). Furthermore, the SST-VPSC correlations increase when only the largest VPSC values are considered: r = 0.76 for February/March VPSC ≥ 15 × 106 km3; this result is consistent with the finding that blocking in the North Pacific tends to precede very high monthly mean VPSC values [Orsolini et al., 2009]. However, while VPSCand North Pacific SSTs were both unusually high in 2011, elevated North Pacific SSTs do not explain the near-zero VPSC in e.g., 1991.

Figure 7.

January/February North Pacific SST anomaly (K) versus February/March VPSC in the 380–550 K layer (106 km3), for each winter in the 1980–2011 period. SST anomaly and VPSC values are denoted by year number (e.g., ‘11’ denotes 2011).

3.3. Sudden Stratospheric Warmings

[22] Major sudden stratospheric warmings (SSWs) are characterized by reversals in the zonal wind direction at 60°N, 10 hPa. The observed occurrence of SSW events, during years when January/February North Pacific SSTs were unusually high or low, is compiled from a list of historical SSWs [Butler and Polvani, 2011]. Distinct SSWs, and thus distinct zonal wind reversals, require at least a 20-day separation, and must be distinct from the final warming (i.e., breakup of the Arctic vortex) [Charlton and Polvani, 2007]. SSWs occurred during all of the four winters with the lowest North Pacific SSTs between 1979 and 2011 (see Figure 1a), while no major SSWs were observed during the four winters with the highest North Pacific SSTs. The GEOS V2 CCM ensembles test whether or not this strong observed sensitivity of the SSW frequency to North Pacific SSTs is robust to a larger sample size.

[23] Table 2 shows the number and frequency of SSWs in the two GEOS V2 CCM ensembles. Simulated SSW identification follows Charlton and Polvani [2007] and Butler and Polvani [2011]. The relative frequency of simulated SSWs is consistent with the mean strength of the Arctic vortex: the frequency of winters with at least one SSW (i.e., active winters) is approximately tripled in the Low ensemble (20 active winters) as compared with the High ensemble (7 active winters). There are two winters each with two SSWs in both ensembles. The increased SSW frequency in the Low ensemble is consistent with the overall increase in variability in January (see Figure 4). The frequency of active winters during perpetual ENSO neutral conditions (10 active winters in a 40-year sample) lies between that for the North Pacific SST extremes.

Table 2. Simulated Number and Frequency of Winters With at Least One SSW, and Number and Frequency of Winters With Two SSWsa
 GEOS V2 CCM Simulations1979–2011 Frequency of Active Winters
Number of Active WintersFrequency of Active WintersNumber of Winters with Two SSWsFrequency of Two-SSW Winters
  • a

    Simulated number and frequency of winters with at least one SSW: 2nd and 3rd columns; number and frequency of winters with two SSWs: 4th and 5th columns. Forty winters for each of the High, Low and ENSO neutral simulations are considered. The 6th column shows the frequency of active winters during the 1979–2011 period, based on a list of historical SSWs [Butler and Polvani, 2011]. For the historical SSWs, the four years with the highest (lowest) January/February North Pacific SSTs represent the High (Low) case.

High70.1820.050
Low200.5020.051
ENSO Neutral100.25000.42

[24] The dynamical situation preceding SSWs in the High ensemble is distinct from that preceding SSWs in the Low ensemble. Composites of the zonally asymmetric component of the 300-hPa geopotential height anomalies in the 5–20 days preceding SSWs, at Arctic latitudes, highlight differences between the ensembles: The High ensemble has a wave number–2-like structure (Figure 8a) while the Low ensemble has a wave number–1-like structure (Figure 8b). These structures are, respectively, consistent with the tropospheric precursors preceding split and displacement type SSWs [Charlton and Polvani, 2007; Cohen and Jones, 2012].

Figure 8.

Zonally asymmetric component of the 300-hPa geopotential height anomalies in the 5–20 days preceding SSWs in the (a) High and (b) Low ensembles, poleward of 30°N. White contours indicate zero difference.

3.4. High-Latitude Stratospheric Ozone and UV

[25] High – Low polar ozone differences are shown in Figure 9. In January through March, ozone differences in the middle and lower stratosphere are not statistically significant, despite strong polar cooling (Figure 6) and weakened polar downwelling (positive w* differences; see colored contours, Figure 9, and Balis et al. [2011]). The reversal in the sign of High – Low 50 hPa temperature differences in mid-April (seeFigures 4 and 6) does not occur in the ozone difference field. Weak, negative ozone differences of 0.1–0.2 ppmv, centered at 50 hPa, occur in late March and April, consistent with the relative cooling at 80°N (Figure 6).

Figure 9.

High – Low differences in ozone mixing ratio [ppmv] and in the residual vertical velocity (w*) (mm s−1) in the 70–90°N region, as a function of date and altitude. Filled, colored contours show w* differences; white contours indicate zero w* difference. Black contours indicate ozone differences; thick, black contours indicate zero ozone difference.

[26] Weak High – Low ozone differences result from a polar lower stratospheric warm bias in the GEOS V2 CCM. Figure 10 shows that the High and Low ensemble mean temperatures at 80°N, 50 hPa remain above the threshold for PSC formation (approximated by the black dotted line), and thus for chemical ozone depletion, throughout the winter. In contrast, observed temperatures during the e.g., 2010–2011 winter (solid black line) dropped below the PSC formation threshold for sustained periods in February and March. Though the simulated area where polar temperatures are below 195 K at 50 hPa is never large enough to match the ozone loss observed in e.g., 2011 (see Figure 9 and Manney et al. [2011]), the 50-hPa PSC area in the High and Low ensembles are significantly different in January, February and March (seeTable 3).

Figure 10.

Time series of zonal mean temperature at 80°N, 50 hPa during the 2010–2011 winter (black), as compared with the High (blue) and Low (cyan) ensemble means. The approximate threshold for PSC formation (195 K) is denoted by the dotted black line.

Table 3. Ensemble and Monthly Mean Area Where 50-hPa Polar Temperature is Less Than 195 K (106 m2)a
 DecJanFebMar
  • a

    Differences between the High and Low areas are significantly significant at the 90% level, in a two-tailed t-test, in January, February and March.

High2.6 ± 1.219.2 ± 2.627.3 ± 2.512.1 ± 2.2
Low3.7 ± 1.211.4 ± 2.516.4 ± 2.86.9 ± 1.7

[27] In April, High – Low ensemble mean total ozone differences are consistent with the observed total ozone anomalies in 2011. In the GEOS CCM simulations (Figure 11a), positive differences of 5–10 DU are seen in the Canadian sector, while negative differences of up to 37 DU are seen in the European and Siberian sectors. Weaker, negative total ozone differences span the midlatitude Pacific and the eastern USA. The observed total ozone anomalies in April 2011 (Figure 11b) match the pattern of the simulated differences, but with two main differences: First, the observed anomalies are approximately three times larger. Second, the observed response lacks the simulated negative total ozone response in the midlatitude East Asia and Pacific sector.

Figure 11.

April total ozone differences (DU) (a) for the High – Low ensemble means and (b) for 2011 – climatology from the TOMS/SBUV data set, poleward of 30°N. In Figure 11a, black Xs indicate differences significant at the 95% level in a two-tailed t-test. White letters denote the approximate location of the four cities: London (L), Moscow (M), Calgary (C) and Washington (W).

[28] Decreased total ozone in spring enhances the flux of clear-sky UV radiation to the earth's surface, while increased total ozone diminishes the UV flux. Simulated April UV differences are calculated for four cities: Washington, Calgary, London and Moscow. These cities span the latitudinal circle at northern midlatitudes.Figure 12 shows the April UV index in the four cities, in two years when North Pacific SSTs were unusually low, three years when North Pacific SSTs were unusually high, as well as the High and Low ensemble mean values. In Washington, London and Moscow, North Pacific SSTs and the April UV index are positively correlated: UV index is highest when North Pacific SSTs are elevated and Arctic ozone depletion is relatively more severe. The statistical significance of High – Low UV differences is highest for Washington. In Calgary, the UV index exhibits the opposite sensitivity to North Pacific SSTs, consistent with the total ozone response in April.

Figure 12.

April clear-sky UV index in four cities: Washington (approximately 283°E, 38°N), Calgary (246°E, 51°N), Moscow (37°E, 55°N) and London (0°E, 51°N). For each city, the first three vertical bars represent the UV index calculated based on April total ozone from the TOMS/SBUV data set: the mean of 1987 and 1988 (unusually low North Pacific SSTs), the mean of 1991 and 1997 (high), and 2011 (high). The last two vertical bars represent the mean UV index calculated based on April total ozone for the High and Low simulations. Differences in the April UV index between the High and Low ensembles are significant at the 95% confidence level in Washington and at the 90% confidence level in London, in two-tailed t-tests.

4. Conclusions

[29] GEOS V2 CCM simulations show that SSTs in the North Pacific region significantly affect the Arctic troposphere and stratosphere in winter. Two ensembles, each of 40 extended winter season simulations, differ only by the SST boundary conditions imposed north of 20°N. The largest, positive SST differences are in the North Pacific region. By imposing no tropical SST differences, the impact of ENSO on Arctic winter variability is removed. While the phase of the QBO varies between each of the ensemble members, differences in the simulated Arctic late winter response during QBO-easterly versus QBO-westerly conditions are negligible.

[30] In the troposphere, enhanced North Pacific SSTs tend to weaken the Aleutian low (consistent with Latif and Barnett [1996], Liu et al. [2006], and Frankignoul and Sennéchael [2007]), inhibiting planetary wave propagation into the stratosphere. Arctic geopotential heights are relatively lower, from December through March. Furthermore, enhanced North Pacific SSTs can affect winter weather in the Northern Hemisphere: relatively lower surface pressure at polar latitudes coupled with relatively higher pressure at midlatitudes imply a strengthening of the North Atlantic Oscillation and Arctic Oscillation (consistent with correlation analysis by Kim and Ahn [2012], for the autumn season). However, SST differences between the High and Low ensembles are non-zero in the North Atlantic, and thus, imposed Atlantic SST differences may have contributed to the simulated tropospheric response.

[31] Reduced tropospheric planetary wave driving, in winters with relatively warmer North Pacific SSTs, leads to a less disturbed Arctic vortex. Lower temperatures in the Arctic stratosphere begin in the upper stratosphere in January, and slowly descend to the lower stratosphere by March and early April. The JFM seasonal mean difference in polar cap temperature at 50 hPa is 1.91 K. The breakup of the Arctic vortex is delayed by approximately 4 days at 10 hPa. When North Pacific SSTs are relatively low, not only is the seasonal mean Arctic vortex weakened, the frequency of simulated SSWs is considerably larger (consistent with Jadin et al. [2010]). Furthermore, the tropospheric dynamical situation preceding SSWs that occur when North Pacific SSTs are anomalously cool (warm) tend to have a wave number–1 (wave number–2) structure. The reduction in SSW frequency by warmer North Pacific SSTs is consistent with the reduction in planetary wave driving.

[32] The Arctic ozone response to enhanced North Pacific SSTs is consistent with the dynamical response. Relative cooling of the Arctic stratosphere corresponds with a relative decrease in polar ozone. In the lower and middle stratosphere, negative ozone differences are centered over the polar cap from January through March. Though the pattern of simulated ozone differences is consistent with the dynamical response, ozone differences are not statistically significant, likely because of the GEOS V2 CCM's mid-winter warm bias.

[33] In April, negative High – Low ozone differences shift away from the pole and toward the Siberian sector. The pattern of simulated total ozone differences is similar to the pattern of ozone anomalies observed in April 2011 [see also Balis et al., 2011]; however, the magnitude of the simulated differences is approximately a third of that observed. In April, negative total ozone differences exceeding 30 DU are simulated in Europe and the Siberian sector (leading to small increases in clear-sky UV index in e.g., London and Moscow), while total ozone increases exceed 20 DU over northern Canada (slightly decreasing the UV index in e.g., Calgary). Weaker negative differences are seen at northern midlatitudes (increasing the April UV index in e.g., Washington).

[34] The results of the present study suggest that one of the keys to predicting the behavior of the Arctic stratosphere in a future climate is to understand variability and trends in the North Pacific Ocean. Interannual variability in North Pacific SSTs is primarily driven by a “subarctic mode” of decadal variability that is independent of ENSO and the Pacific Decadal Oscillation (PDO) [Nakamura et al., 1997; Frankignoul and Sennéchael, 2007; Orsolini et al., 2009; Nishii et al., 2010]. The Coupled Model Intercomparison Project Phase 5 (CMIP5; http://cmip-pcmdi.llnl.gov/cmip5/) simulations of likely 21st century climate scenarios will provide updated forecasts of future variability and trends in North Pacific SSTs.

[35] The above model results support the hypothesis that anomalously warm North Pacific SSTs contributed to the strong and prolonged cooling of the polar lower stratosphere, and severe ozone depletion, that were observed during the 1996–1997 and 2010–2011 winters [Pawson and Naujokat, 1999; Manney et al., 2011; Hurwitz et al., 2011a]. January/February North Pacific SSTs are well correlated with late winter VPSC, explaining some of the “coldest winters” as identified by Rex et al. [2004, 2006] and Manney et al. [2011]. However, elevated North Pacific SSTs alone do not predict polar stratospheric cooling in late winter: The warmest North Pacific SSTs of the satellite era were observed during the 1990–1991 winter, while VPSCwas near-zero. Similarly, in the GEOS V2 CCM simulations, not all winters forced by high North Pacific SSTs lead to late winter stratospheric cooling, because stochastic atmospheric variability also plays a role in modulating the Arctic winter climate.

Acknowledgments

[36] The authors thank S. Frith for providing the updated TOMS/SBUV data and processing the model output, M. Rex and P. von der Gathen for supplying VPSC time series, A. Karpechko and three anonymous reviewers for helpful comments, and NASA's ACMAP program for funding.