On the Observed Relationships between Variability in Gulf Stream Sea Surface1

Temperatures and the Atmospheric Circulation in the North Atlantic2

Samantha M. Wills⇤ and David W. J. Thompson3

Department of Atmospheric Science, Colorado State University, Fort Collins, CO 805214

Laura M. Ciasto5

Geophysical Institute, University of Bergen and Bjerknes Centre for Climate Research,6

Bergen, Norway7

⇤Corresponding author address: Samantha M. Wills, Department of Atmospheric Science, Colorado State8

University, Fort Collins, CO 805219

E-mail: smwills@atmos.colostate.edu10

1

ABSTRACT11

The advent of increasingly high-resolution satellite observations and numerical models12

has led to a series of advances in our understanding of the role of midlatitude sea surface13

temperature (SST) in climate variability, especially near western boundary currents (WBC).14

Observational analyses suggest that ocean dynamics play a central role in driving interannual15

SST variability over the Kuroshio-Oyashio and Gulf Stream extensions. Numerical experi-16

ments suggest that variations in the SST field within these WBC regions may have a much17

more pronounced influence on the atmospheric circulation than previously thought.18

In this study, the authors examine the observational support for (or against) a ro-19

bust atmospheric response to midlatitude SST variability in the Gulf Stream extension.20

To do so, they apply lead/lag analysis based on daily-mean data to assess the evidence21

for two-way coupling between SST anomalies and the atmospheric circulation on transient22

timescales, building o↵ of previous studies that have utilized weekly data. A novel decom-23

position approach is employed to demonstrate that atmospheric circulation anomalies over24

the Gulf Stream extension can be separated into two distinct patterns of midlatitude atmo-25

sphere/ocean interaction: 1) a pattern that peaks 2-3 weeks before the largest SST anomalies26

in the Gulf Stream extension, which can be viewed as the “atmospheric forcing” and 2) a27

pattern that peaks several weeks after the largest SST anomalies, which the authors argue28

can be viewed as the “atmospheric response”. The latter pattern is linearly independent of29

the former, and is interpreted as the potential response of the atmospheric circulation to30

SST variability in the Gulf Stream extension.31

32

KEYWORDS: midlatitude ocean-atmosphere interaction, Gulf Stream, atmospheric vari-33

ability34

2

1. Introduction35

The ocean is an integral part of the climate system, but its impact on the atmosphere36

varies greatly from one region of the globe to another. In the tropics, variations in sea surface37

temperatures (SST) are largely balanced by vertical motion [e.g., Hoskins and Karoly 1981].38

Hence, the linear atmospheric response to tropical SST anomalies can readily extend into39

the free tropospheric circulation and have a notable impact on global climate [e.g., Horel and40

Wallace 1981]. In contrast, variations in midlatitude SST anomalies are readily balanced by41

small changes in the horizontal wind field [e.g., Hoskins and Karoly 1981], and thus the linear42

atmospheric response to midlatitude SST anomalies may be relatively shallow and weak.43

Not surprisingly, the e↵ects of midlatitude SST anomalies on the large-scale atmospheric44

circulation have proven di�cult to isolate and quantify in both numerical experiments and45

observations, as summarized in the review by Kushnir et al. [2002].46

Nevertheless, over the past decade, analyses of increasingly high-resolution satellite47

observations and numerical models have revealed a potentially more important role of the48

midlatitude ocean in extratropical climate than previously thought. The most robust e↵ects49

of midlatitude SSTs on the large-scale atmospheric circulation have been found in the context50

of the climatological-mean circulation. Analyses of high resolution SST and surface wind51

stress observations reveal that the climatological-mean near-surface wind field is strongly52

influenced by large horizontal gradients in the SST field, such as those associated with53

the major western boundary currents [e.g., O’Neill et al. 2003; Nonaka and Xie 2003;54

Chelton et al. 2004; Chelton and Xie 2010]. The associated patterns of convergence in the55

atmospheric boundary layer seemingly extend to vertical motion in the free troposphere and56

thus precipitation [e.g., Minobe et al. 2008, 2010]. Results from numerical experiments57

run with and without sharp gradients in the SST field suggest that the climatological-mean58

ocean fronts play a key role in determining the location and amplitude of the extratropical59

storm tracks [e.g., Nakamura et al. 2008; Sampe et al. 2010].60

To what extent variability in midlatitude SSTs influences the atmospheric circulation61

3

is less clear, but evidence is building that the influence may not be trivial. Observational62

analyses suggest that variations in SSTs in the vicinity of the Northern Hemisphere western63

boundary currents are linked to significant changes in the large-scale atmospheric circulation64

[e.g., Czaja and Frankignoul 2002; Ciasto and Thompson 2004; Frankignoul et al. 2011;65

Kwon and Joyce 2013]. Numerical simulations imply that variations in midlatitude SST66

gradients are linked to changes in the amplitudes of the storm tracks [e.g., Brayshaw et al.67

2008; Nakamura et al. 2009; O’Reilly and Czaja 2015]. Importantly, a very recent numerical68

experiment suggests that the atmospheric response to midlatitude SST anomalies may vary69

dramatically depending on the spatial resolution of the atmospheric model [e.g., Smirnov70

et al. 2014b]: in a low resolution version of the CAM5 atmospheric general circulation71

model, the atmospheric response to midlatitude SST anomalies is dominated by horizontal72

temperature advection in the lowermost troposphere, but in a high-resolution version, it73

includes substantial changes in vertical motion and thus potentially the hemispheric-scale74

circulation. The link between variations in midlatitude SSTs and vertical motion is also75

found in numerical experiments of the extratropical storm response to SST anomalies [e.g.,76

Czaja and Blunt 2011; Sheldon and Czaja 2013].77

The goal of this contribution is to re-examine the observational evidence for midlati-78

tude ocean-atmosphere interaction, with a focus on variations in SSTs over the Gulf Stream79

extension. We exploit daily-mean data to examine the lead/lag relationships between vari-80

ability in the atmospheric circulation and SST variability in the Gulf Stream extension on81

subseasonal timescales. The key novel result is that SST anomalies in the Gulf Stream ex-82

tension are associated with two distinct and independent patterns of atmospheric variability:83

1) a pattern that leads the SST field and is interpreted as the atmospheric forcing of the84

SST anomalies and 2) a pattern that lags the SST field and is interpreted as the atmospheric85

response. The former pattern is expected and is consistent with previous results. As far as86

we know, the latter pattern has not been documented in association with atmosphere/ocean87

interaction over the North Atlantic. Section 2 describes the data. Section 3 explores the88

4

patterns of atmospheric variability associated with variations in SSTs over the Gulf Stream89

extension. Section 4 provides a physical interpretation of the results. Conclusions are pro-90

vided in Section 5.91

92

2. Data93

All results are based on daily-mean output from the ECMWF Interim Reanalysis94

(ERA-Interim). The analyses were performed on a 1.5� ⇥ 1.5� mesh and over the 35 year95

period 1979 - 2013. Anomalies of SST, potential temperature (✓), wind (~v), and geopoten-96

tial height (Z) were formed by removing the long-term mean seasonal cycle from the data.97

Throughout the study, SLP is expressed as geopotential height at 1000hPa (Z1000), as de-98

picted in the figures. Note that sea surface temperature is a prescribed boundary condition99

in ERA-Interim and is a collection of several di↵erent observational data products [e.g., Dee100

et al. 2001, c.f., Table 1]. As noted in the text, key results are reproduced using SST data101

from the NOAA Optimum Interpolated dataset [e.g., Reynolds et al. 2002]. Apart from the102

removal of the seasonal cycle, the data are not filtered in any way in the analyses.103

104

3. Observed lead/lag relationships between the atmospheric circulation and105

SSTs in the Gulf Stream extension106

Figure 1 reviews key aspects of the climatological-mean circulation and SST field over107

the North Atlantic during the NH winter months of December-February (DJF). Panel (a)108

shows the DJF-mean SST and 850hPa wind fields; panel (b) shows the standard deviation109

of daily-mean SST anomalies during DJF. As noted extensively in previous studies [e.g.,110

Nakamura et al. 1997; Nonaka and Xie 2003; Ciasto and Thompson 2004; Deser et al. 2010;111

Smirnov et al. 2014a, etc.], the standard deviation of midlatitude SSTs peaks in the region112

of largest horizontal temperature gradients. Variations in SSTs in the Gulf Stream region113

and its extension can arise from forcing by the atmospheric flow, particularly in association114

with temperature advection from the cold continental regions to the west [e.g., Frankignoul115

5

1985; Haney 1985; Kushnir et al. 2002]. They can also arise from forcing by the ocean116

circulation itself, especially near western boundary currents [e.g., Frankignoul and Reynolds117

1983; Smirnov et al. 2014a]. The focus of this paper is on the two-way interactions between118

the large-scale atmospheric circulation and SST anomalies over the region of large SST119

variance in the Gulf Stream extension (Fig. 1b).120

To investigate the linkages between the atmospheric circulation and SSTs in the Gulf121

Stream extension, we first generate a time series of daily-mean SST anomalies averaged over122

the region 37.5�N - 45�N, 72�W - 42�W (indicated by the box in Fig. 1b). The index123

(hereafter GSST ) is standardized so that it has a mean of zero and standard deviation of one.124

By construction, positive values of the index correspond to warmer than normal SSTs in the125

Gulf Stream extension, and vice versa. We then compute lag relationships between various126

fields and the GSST index using daily-mean data. Note that the results are not sensitive to127

the specific domain used to define the Gulf Stream extension.128

The left column in Fig. 2 shows daily-mean SST (shading) and SLP (contours) anoma-129

lies regressed onto the GSST index time series as a function of lag. Negative lags indicate130

results where the SLP and SST fields precede peak amplitude in the GSST index, and vice131

versa for positive lags. The GSST index is always centered on the 90-day period from Dec.132

1 - Feb. 28, whereas the fields being regressed shift from Nov. 11 - Feb 8. (for the lag -20133

regressions) to Dec. 21 - Mar. 20 (for the lag +20 regressions). The statistical significance134

of the SLP anomalies at positive lag is indicated in Fig. 3.135

The SST regression coe�cients shown in the left column indicate the evolution of the136

SST field. By construction, the SST anomalies peak at lag 0 and in the vicinity of the Gulf137

Stream extension. The amplitudes of the SST anomalies are comparable to the standard138

deviations in Fig. 1b (also by construction since the GSST index is standardized). They139

decay only slightly with increasing lag, consistent with the relatively large thermal inertia140

of the ocean mixed layer.141

The SLP regression coe�cients indicate the attendant evolution of the atmospheric142

6

circulation. The most pronounced circulation anomalies are found at negative lags (i.e.,143

the atmosphere leading variations in GSST ), when positive SLP anomalies span much of the144

North Atlantic basin. As indicated in previous analyses based on pentad and weekly-mean145

SST data [e.g., Deser and Timlin 1997; Ciasto and Thompson 2004], the SLP anomalies at146

negative lags are consistent with forcing of the SST field by anomalies in the atmospheric147

circulation. In regions of large SST gradients (Fig. 1a), periods of anomalously warm SSTs148

over the Gulf Stream extension are consistent with anomalously southerly flow (Fig. 2, top149

panels in left column), and vice versa.150

While the large SLP anomalies that lead variations in GSST are entirely expected, the151

SLP anomalies that lag variations in GSST are more intriguing. The regressions suggest that152

the several week period following anomalously warm conditions in the Gulf Stream extension153

is marked by anomalously low SLP over the Gulf Stream region and anomalously high SLP154

centered to the south of Iceland. The pattern of SLP anomalies at positive lags is statistically155

significant at the 95% confidence level (Fig. 3) and reproducible in other datasets (i.e., the156

NOAA Optimum Interpolated dataset, not shown). Importantly, the anomalous circulations157

leading and lagging variations in GSST have distinct spatial structures, as quantified below.158

The di↵erences between the patterns of SLP anomalies that precede and follow vari-159

ations in GSST are quantified using a linear decomposition method as follows. First, we160

define the pattern of SLP anomalies at lag -20 as the “atmospheric forcing” pattern (here-161

after denoted SLP-20). We then decompose the SLP regression maps at all lags into two162

components: 1) a pattern that is linearly congruent with the “atmospheric forcing” pattern163

(i.e., the “fit” to SLP-20) and 2) a “residual” pattern that is linearly independent of the164

SLP-20 regression map. That is, the SLP regression map at lag t is given as:165

SLPt = ↵t · SLP-20 + SLP⇤t (1)

where SLPt is the total SLP regression map at lag t, ↵t is the spatial regression coe�cient166

7

found by projecting SLPt onto SLP-20 (↵ varies as a function of lag), and SLP⇤t is the residual167

SLP pattern at lag t.168

The middle column of Fig. 2 shows the components of the SLP regressions that are169

linearly congruent with (or “fitted” to) SLP-20, i.e., the middle column shows the evolution of170

the “atmospheric forcing” pattern as given by ↵t·SLP-20. The SST anomalies are reproduced171

from the left column. By construction, at all lags the patterns in the middle column are172

identical to the total regression at lag -20 and vary only in amplitude. The amplitude of173

the “atmospheric forcing” pattern decays on a timescale of several weeks, and has only weak174

amplitude at positive lags.175

The right column in Fig. 2 shows the components of the SLP regressions that are176

linearly independent of SLP-20, i.e., the column shows the “residual” SLP patterns given177

by SLP⇤t . Note that the spatial correlation coe�cient between SLP⇤

t and SLP-20 is zero at178

all lags, hence the SLP⇤t maps reflect the components of the regressions that are linearly179

independent of the “atmospheric forcing” pattern.180

The residual SLP patterns in the right column are not constrained to have similar spa-181

tial structure at all lags, but they do. This is important since it suggests that the space/time182

evolution of SLP anomalies associated with variations in SSTs in the Gulf Stream extension183

region can be viewed as the superposition of two structures: 1) a pattern that peaks in am-184

plitude several weeks before peak amplitude in the GSST index and is consistent with forcing185

of the SST field by the atmospheric circulation (middle column of Fig. 2) and 2) a linearly186

independent pattern of SLP variability that grows in amplitude from ⇠lag 0 through the187

several week period after peak amplitude in the GSST index. As noted above and explored188

in Section 4, the former, “atmospheric forcing” pattern is consistent with anomalous tem-189

perature advection by the anomalous horizontal flow. In the next section, we will argue that190

the “residual” pattern may be interpreted as the “atmospheric response” to the underlying191

SST anomalies.192

Figure 4 explores the associated time varying structures in the 500hPa geopotential193

8

height (Z500) and 850hPa potential temperature (✓850) fields. The figure is constructed in194

an identical manner to Fig. 2, but in this case the analyses and decomposition procedure195

are based on lag regressions between Z500 anomalies and the GSST index (contours) and be-196

tween ✓850 anomalies and the GSST index (shading). The results in Fig. 4 indicate that the197

patterns of atmospheric variability that lead and lag GSST both include a barotropic compo-198

nent. They also indicate that the temperature changes in the lower troposphere lie directly199

over the SST anomalies when the atmosphere leads GSST , but are shifted to the northeast of200

the SST anomalies when the atmosphere lags GSST . As noted in the following section, the201

structure of the lower tropospheric temperature anomalies may provide an important clue202

regarding the development of the circulation anomalies that follow variations in GSST .203

204

4. Discussion205

The results in Figures 2-4 suggest that wintertime SST variability in the Gulf Stream206

extension is associated with two, linearly independent patterns of atmospheric variability: 1)207

a pattern of circulation anomalies that peaks prior to largest amplitude in the SST field and208

2) a very di↵erent pattern of circulation anomalies that peaks after largest amplitude in the209

SST field. The linear independence of the two patterns is highlighted by the decomposition210

applied in the middle and right columns of Figs. 2 and 4. But in practice, the linear decom-211

position is not required to identify the pattern of circulation anomalies that lags GSST (i.e.,212

the lower left and lower right panels in Figs. 2 and 4 are nearly identical). In this section,213

we explore key aspects of the pattern of circulation anomalies that forms during the several214

week period after largest amplitude in the GSST time series.215

216

a. Comparison with results based on an atmospheric index217

There are at least two possible physical explanations for the pattern of atmospheric218

circulation anomalies that lags GSST : 1) the pattern may simply reflect the evolution of219

the atmospheric circulation that would occur even in the absence of the underlying SST220

9

anomalies; or 2) the pattern may reflect the response of the atmospheric circulation to221

SST anomalies in the Gulf Stream extension. To test against the former possibility, we222

repeated the analyses in Fig. 2, but for regressions based on the expansion coe�cient time223

series of the SLP-20 regression map. The expansion coe�cient time series was formed by224

projecting daily-mean SLP anomalies onto the SLP-20 map on all days of the analysis and225

then standardizing the resulting index. By construction, the time series (hereafter Gatmos)226

indicates the temporal evolution of the SLP pattern shown in the top left panel of Fig 2.227

Lag regressions of SST and SLP onto the Gatmos time series were then calculated for lags228

spanning 0 to +40 days. The SLP regressions hence indicate the time evolution of the SLP-20229

pattern with no (direct) information from the SST field. The lags range from 0 to 40 days230

so that the results can be compared directly with those based on the GSST time series (i.e.,231

lag 0 in the Gatmos time series corresponds to lag-20 in the GSST time series).232

Figure 5 shows the results of the regression analysis. The SLP anomalies associated233

with Gatmos decay on a time scale of about 10-20 days, consistent with the time scale of234

the SLP anomalies in the middle column of Fig. 2. However, results based on the Gatmos235

time series di↵er from those based on the GSST time series in two important ways: 1) the236

SST anomalies derived from regressions onto Gatmos are relatively weak, which suggests a237

substantial fraction of the variance in the SST field is unrelated to this particular pattern of238

SLP forcing; and 2) the SLP anomalies associated with the Gatmos time series do not evolve239

into a pattern of circulation anomalies reminiscent of the structure that lags peak amplitude240

in GSST . Hence the pattern of residual SLP anomalies observed at positive lags in Fig. 2241

does not appear to simply reflect the evolution of the atmospheric circulation. Rather, it is242

seemingly dependent on the inclusion of information from the SST field itself.243

244

b. Signature in temperature advection245

What physical processes might give rise to the pattern of SLP anomalies at positive246

lag? Here we argue that they may reflect the circulation response to the poleward and247

10

upward advection of anomalously warm air from the Gulf Stream extension.248

As shown in Fig. 4, the circulation anomalies associated with GSST are accompanied by249

two distinct patterns of lower tropospheric temperature anomalies: 1) positive temperature250

anomalies that overlie the Gulf Stream extension during the weeks preceding variations in251

GSST ; and 2) positive temperature anomalies to the northeast of the Gulf Stream extension252

during the weeks following variations in GSST . As shown below, the former temperature253

anomalies are consistent with temperature advection by the anomalous circulation across254

the climatological-mean gradients in lower tropospheric temperature, whereas the latter are255

in part dependent on temperature advection by the climatological-mean circulation across256

the anomalous gradients in temperature. As such, the former are not dependent on the257

existence of anomalies in the SST field, but the latter are.258

The relative roles of temperature advection by the anomalous and climatological-mean259

atmospheric circulations can be quantified by decomposing the total anomalous horizontal260

temperature advection at 850hPa as follows:261

(I) (II) (III)

(~u · ~5✓)TOT = ~uA~5✓C + ~uC

~5✓A + ~uA~5✓A,

(2)

where ~u and ✓ represent the horizontal components of the wind and potential temperature262

at 850hPa, respectively, C denotes the climatological mean, and A denotes the anomalies.263

The total anomalous temperature advection is given by the left hand side (LHS), and the264

terms on the right hand side (RHS) denote I) advection by the anomalous flow across the265

climatological-mean temperature gradients; II) advection by the climatological-mean flow266

across the anomalous temperature gradients; and III) advection by the anomalous flow across267

the anomalous temperature gradients.268

Figure 6 shows the patterns of temperature advection associated with all three terms269

at lag 0. The top panel shows results for the first term on the RHS of Eq. 2. Contours270

indicate the climatological-mean isotherms at 850hPa; vectors indicate the anomalous flow at271

11

850hPa; and shading indicates the associated anomalous temperature advection. As inferred272

in Section 3, advection by the anomalous flow across the climatological-mean temperature273

gradients gives rise to a pattern of temperature tendencies at 850hPa that peaks over the274

region of largest SST anomalies, consistent with forcing of the SST field by the anomalous275

atmospheric circulation.276

The middle panel of Fig. 6 shows results for the second term on the RHS of Eq. 2. Here,277

contours indicate the 850hPa temperature anomalies; vectors indicate the climatological-278

mean flow at 850hPa; and shading indicates the associated temperature advection. Advection279

by the climatological-mean flow across the anomalous temperature gradients gives rise to a280

very di↵erent pattern of temperature tendencies than that shown in the upper panel. The281

anomalous temperature advection in the middle panel has comparable amplitude to that in282

the top panel, but projects onto the atmospheric temperature and circulation anomalies over283

the central North Atlantic rather than the Gulf Stream extension. Since it is dependent on284

the anomalies in lower tropospheric temperature, the pattern of temperature tendencies in285

the middle panel of Fig. 6 (and thus the lower tropospheric temperature anomalies over the286

North Atlantic following GSST ) derive from the warming of the lower troposphere over the287

Gulf Stream region.288

The bottom panel of Fig. 6 shows the corresponding results for the third term on the289

RHS of Eq. 2. Advection by the anomalous flow across the anomalous temperature gradients290

has a relatively small contribution to temperature advection in the lower troposphere.291

292

c. Signature in vertical motion and the hemispheric scale circulation293

To the extent that the underlying SST field influences variations in lower tropospheric294

temperatures over the Gulf Stream extension, it follows that the pattern of temperature ad-295

vection in the middle panel of Fig. 6 may be at least partially attributed to the underlying296

temperature anomalies in the SST field. Figure 7 reveals that the resulting positive temper-297

ature anomalies over the central North Atlantic during the period following peak amplitude298

12

in GSST are also associated with anomalous rising motion.299

Figure 7 shows meridional and vertical circulation anomalies regressed on GSST at300

lag+20 (i.e. when the atmosphere lags the ocean) over the boxed region indicated in the301

second panel of Fig. 6. The warming to the northeast of the Gulf Stream extension is marked302

by positive temperature anomalies that extend throughout the atmospheric column at posi-303

tive lag. Notably, the positive temperature anomalies also coincide with anomalous upward304

motion between ⇠45 and 60 degrees latitude. The anomalous rising motion is consistent305

with results by Smirnov et al. [2014b], who noted that SST anomalies over the Kuroshio-306

Oyashio extension are associated with warm, rising air when the atmosphere lags the SST307

field by several weeks. Similar results were also noted by Czaja and Blunt [2011] and Sheldon308

and Czaja [2013], who argued that SST-induced heating in the Gulf Stream region can be309

advected upward and poleward in the warm sector of extratropical storms.310

The anomalous rising motion indicated in Fig. 7 is important for three reasons. One,311

the coexistence of heating and rising motion indicates that anomalous heating may be viewed312

as forcing, rather than responding, to the changes in vertical motion (if the anomalous313

motion was downwards, then the positive temperature anomalies in the free troposphere314

would be consistent with adiabatic compression). Two, it suggests that the heating due315

to extratropical SST anomalies is being balanced, at least in part, by anomalous vertical316

motion. A similar conclusion was reached by Smirnov et al. [2014b] in their simulations of317

the atmospheric response to Kuroshio-Oyashio extension SST anomalies. Third, the changes318

in vertical motion suggest that the anomalous heating of the lower troposphere in regions to319

the northeast of the Gulf Stream extension extends to the upper tropospheric circulation.320

If the heating of the lower troposphere by the SST field is balanced by vertical motion,321

it follows that it will lead to the generation of circulation anomalies at upper tropospheric322

levels. As demonstrated in Fig. 4, the lower tropospheric heating anomalies to the northeast323

of the Gulf Stream extension are, in fact, associated with higher than normal geopotential324

heights at 500hPa, consistent with hydrostatic balance of the column of air. As shown in Fig.325

13

8, the free tropospheric geopotential height anomalies associated with GSST also appear to326

extend downstream beyond the Gulf Stream extension. Figure 8 shows the regression of the327

SLP and SST fields onto the GSST index at lag+20 for the entire hemisphere. The results328

are identical to those shown in the left column of Fig. 2, but include regression coe�cients329

beyond the North Atlantic sector. The pattern of circulation anomalies associated with GSST330

is consistent with a wavetrain extending across much of Europe and western Russia. A very331

similar pattern of geopotential height anomalies was recently found in Ciu et al. [2015] in332

their analysis of covariability between climate variability over central Asia and the North333

Atlantic sector.334

335

5. Conclusions336

The results in this study suggest that SST variability in the Gulf Stream extension is337

associated with two distinct patterns of tropospheric circulation anomalies: 1) a pattern that338

peaks in amplitude several weeks before the largest anomalies in Gulf Stream extension SSTs,339

and is consistent with forcing of the SST field by the anomalous atmospheric circulation; and340

2) a very di↵erent pattern that peaks in amplitude several weeks after the largest anomalies341

in Gulf Stream extension SSTs. As far as we know, the latter pattern has not been identified342

in previous observational analyses of atmosphere/ocean interaction of the North Atlantic343

sector. Lead/lag regressions do not prove causality, but several observations suggest that344

the pattern of circulation anomalies that lag the SST field may reflect the atmospheric345

response to SST anomalies in the Gulf Stream region: the pattern of circulation anomalies346

at positive lags has a very di↵erent spatial structure than the pattern at negative lags (Fig.347

2), it is highly statistically significant (Fig. 3), and it does not emerge from analyses that348

do not include (direct) information from the SST field (Fig. 5).349

We have argued that the pattern of circulation anomalies that follows variations in350

Gulf Stream extension SSTs is driven by anomalous vertical motion in the region to the351

northeast of the Gulf Stream extension (Fig. 6): e.g., positive lower tropospheric temperature352

14

anomalies over the Gulf Stream region are advected northeastward by the climatological flow353

(Fig. 6), where they are at least partially balanced by anomalous rising motion (Fig. 7).354

The anomalous rising motion is important, since it suggests heating over the Gulf Stream355

extension is capable of perturbing the free tropospheric circulation. It is also robust: a356

similar pattern of vertical motion anomalies emerges in analyses of the atmospheric response357

to SST anomalies over the Kuroshio-Oyashio extension [e.g., Smirnov et al. 2014b], and in358

analyses of the extratropical storm response to SST anomalies in the Gulf Stream region359

[e.g., Czaja and Blunt 2011; Sheldon and Czaja 2013].360

The results shown here are derived from lead/lag analysis of daily-mean data. We361

believe that the use of lag regressions based on daily-mean data may provide insight into362

the nature of extratropical atmosphere/ocean coupling in the same way that it has led363

to new insights into the nature of stratosphere/troposphere coupling [e.g., Baldwin and364

Dunkerton 2001]. The distinction between the patterns of circulation anomalies at negative365

and positive lags identified here would be much more di�cult to extract from regressions366

based on monthly or seasonal-mean data. However, it is interesting to emphasize that both367

patterns are embedded in such regressions.368

For example, the left panel in Fig. 9 shows results derived by regressing winter sea-369

son (NDJFM) monthly-mean Z500 and ✓850 onto standardized, monthly-mean values of the370

GSST index. The middle panel shows the component of the regression map that is linearly371

congruent with the “atmospheric forcing” pattern, as defined from the lag -20 day regression372

maps shown in Fig. 4. The right panel (the residual of the regression) shows the di↵erences373

between the left and middle panels. The residual pattern in the right panel bears strong374

resemblance to the pattern of circulation anomalies that lags variations in GSST on daily375

timescales. Hence the winter season monthly-mean regression map in the left panel reflects376

the juxtaposition of two distinct patterns of circulation anomalies: the pattern that leads377

variations in GSST by several weeks, and the pattern that lags variations in GSST by several378

weeks. We believe the use of daily-mean data is key for future analyses of extratropical379

15

atmosphere/ocean interaction on subseasonal timescales, and that the results have poten-380

tial implications for understanding the response of the atmosphere to variations in SSTs on381

longer timescales.382

383

Acknowledgements384

S.M.W. is funded by the NASA Physical Oceanography program under Grant NNX13AQ04G.385

D.W.J.T. is funded by the NASA Physical Oceanography program and NSF Climate Dy-386

namics program. L.M.C. is supported by the Centre for Climate Dynamics at the Bjerknes387

Centre through a grant to the WaCyEx project.388

16

REFERENCES389

Baldwin, M.P., and T.J. Dunkerton, 2001: Stratospheric harbingers of anomalous weather390

regimes. Science, 294, 581 - 584.391

Barsugli, J.J., and D.S. Battisti, 1998: The Basic E↵ects of Atmosphere-Ocean Thermal392

Coupling on Midlatitude Variability. J. Atmos. Sci., 55, 477 - 493.393

Brayshaw, D.J., B. Hoskins, and M. Blackburn, 2008: The Storm-Track Response to Ideal-394

ized SST Perturbations in an Aquaplanet GCM. J. Atmos. Sci., 65, 2842 - 2860.395

Chelton, D.B., M.G. Schlax, M.H. Freilich, and R.F. Milli↵, 2004: Satellite Measurements396

Reveal Persistent Small-Scale Features in Ocean Winds. Science, 303, 978 - 983.397

Chelton, D.B., and S.-P. Xie, 2010: Coupled ocean-atmosphere interaction at oceanic398

mesoscales. Oceanography, 23, 52 - 69.399

Ciasto, L.M., and D.W.J. Thompson, 2004: North Atlantic Atmosphere-Ocean Interaction400

on Intraseasonal Time Scales. J. Climate, 17, 1617 - 1621.401

Czaja, A., and C. Frankignoul, 2002: Observed Impact of Atlantic SST Anomalies on the402

North Atlantic Oscillation. J. Climate, 15, 606 - 623.403

Czaja, A., and N. Blunt, 2011: A new mechanism for ocean-atmosphere coupling in mid-404

latitudes. Q. J. R. Meteorol. Soc., 137, 1095 - 1101.405

Dee, D.P., S.M. Uppala, A.J. Simmons, P. Berrisford, P. Poli, S. Kobayashi, U. Andrae,406

M.A. Balmaseda, G. Balsamo, P. Bauer, P. Bechtold, A.C.M. Beljaars, L. van de407

Berg, J. Bidlot, N. Bormann, C. Delsol, R. Dragani, M. Fuentes, A.J. Geer, L.408

Haimberger, S.B. Healy, H. Hersbach, E.V. Holm, L. Isaksen, P. Kallberg, M. Kohler,409

M. Matricardi, A.P. McNally, B.M. Monge-Sanz, J.-J. Morcrette, B.-K. Park, C.410

Peubey, P. de Rosnay, C. Tavolato, J.-N. Thepaut, and F. Vitart, 2011: The ERA-411

17

Interim reanalysis: configuration and performance of the data assimilation system.412

Q. J. R. Meteorol. Soc., 137, 553 - 597.413

Deser, C., and M.S. Timlin, 1997: Atmosphere-Ocean Interaction on Weekly Timescales in414

the North Atlantic and Pacific. J. Climate, 10, 393 - 408.415

Deser, C., M.A. Alexander, S.-P. Xie, and A.S. Phillips, 2010: Sea Surface Temperature416

Variability: Patterns and Mechanisms. Annu. Rev. Mar. Sci., 2, 115 - 143.417

Frankignoul, C., and R.W. Reynolds, 1983: Testing a Dynamical Model for Mid-Latitude418

Sea Surface Temperature Anomalies. J. Phys. Oceanogr., 13, 1131 - 1145.419

Frankignoul, C., 1985: Sea Surface Temperature Anomalies, Planetary Waves, and Air-Sea420

Feedback in the Middle Latitudes. Rev. Geophys., 23, 357 - 390.421

Frankignoul, C., N. Sennechael, Y-O. Kwon, and M.A. Alexander, 2011: Influence of the422

Meridional Shifts of the Kuroshio and the Oyashio Extensions on the Atmospheric423

Circulation. J. Climate, 24, 762 - 777.424

Haney, R.L., 1985: Midlatitude Sea Surface Temperature Anomalies: A Numerical Hind-425

cast. J. Phys. Oceanogr., 15, 787 - 799.426

Horel, J.D., and J.M. Wallace, 1981: Planetary-Scale Atmospheric Phenomena Associated427

with the Southern Oscillation. Mon. Wea. Rev., 109, 813 - 829.428

Hoskins, B.J., and D.J. Karoly, 1981: The Steady Linear Response of a Spherical Atmo-429

sphere to Thermal and Orographic Forcing. J. Atmos. Sci., 38, 1179 - 1196.430

Kushnir, Y., W.A. Robinson, I. Blade, N.M.J. Hall, S. Peng, and R. Sutton, 2002: Atmo-431

spheric GCM Response to Extratropical SST anomalies: Synthesis and Evaluation.432

J. Climate, 15, 2233 - 2256.433

18

Kwon, Y.-O., and T. M. Joyce, 2013: Northern Hemsiphere Winter Atmospheric Transient434

Eddy Heat Fluxes and the Gulf Stream and Kuroshio-Oyashio Extension Variability.435

J. Climate, 26, 9839 - 9859.436

Minobe, S., A. Kuwano-Yoshida, N. Komori, S-P. Xie, and R.J. Small, 2008: Influence of437

the Gulf Stream on the troposphere. Nature, 452, 206 - 210.438

Minobe, S., M. Miyashita, A. Kuwano-Yoshida, H. Tokinaga, and S-P. Xie, 2010: Atmo-439

spheric Response to the Gulf Stream: Seasonal Variations. J. Climate, 23, 3699 -440

3719.441

Nakamura, H., G. Lin, and T. Yamagata, 1997: Decadal Climate Variability in the North442

Pacific during the Recent Decades. Bull. Amer. Meteor. Soc., 78, 2215 - 2225.443

Nakamura, H., T. Sampe, A. Goto, W. Ohfuchi, and S-P. Xie, 2008: On the im-444

portance of midlatitude oceanic frontal zones for the mean state and domi-445

nant variability in the tropospheric circulation. Geophys. Res. Lett., 35, L15709,446

doi:10.1029/2008GL034010.447

Nakamura, M., and S. Yamane, 2009: Dominant Anomaly Patterns in the Near-Surface448

Baroclinicity and Accompanying Anomalies in the Atmosphere and Oceans. Part I:449

North Atlantic Basin. J. Climate, 22, 880 - 904.450

Nonaka, M., and S-P. Xie, 2003: Covariations of Sea Surface Temperature and Wind over451

the Kuroshio and Its Extension: Evidence for Ocean-to-Atmosphere Feedback, J.452

Climate, 16, 1404 - 1413.453

O’Neill, L.W., D.B. Chelton, and S.K. Esbensen, 2003: Observations of SST-Induced Per-454

turbations of the Wind Stress Field over the Southern Ocean on Seasonal Timescales.455

J. Climate, 16, 2340 - 2354.456

19

O’Reilly, C.H., and A. Czaja, 2015: The response of the Pacific storm track and atmospheric457

circulation to Kuroshio Extension variability. Q. J. R. Meteorol. Soc., 141, 52 - 66.458

Palmer, T.N., and Z. Sun, 1985: A modeling and observational study of the relationship459

between sea surface temperature in the north-west Atlantic and the atmospheric460

general circulation. Q.J.R. Met. Soc., 111, 947 - 975.461

Reynolds, R.W., N.A. Rayner, T.M. Smith, D.C. Stokes, and W. Wang, 2002: An improved462

in situ and satellite SST analysis for climate. J. Climate, 15, 1609 - 1625.463

Sampe, T., H. Nakamura, A. Goto, and W. Ohfuchi, 2010: Significance of a Midlatitude464

SST Frontal Zone in the Formation of a Storm Track and an Eddy-Driven Westerly465

Jet. J. Climate, 23, 1793 - 1814.466

Sheldon, L., and A. Czaja, 2013: Seasonal and interannual variability of an index of deep467

atmospheric convection over western boundary currents. Q. J. R. Meteorol. Soc.468

DOI:10.1002/qj.2103.469

Smirnov, D., M. Newman, and M.A. Alexander, 2014a: Investigating the Role of Ocean-470

Atmosphere Coupling in the North Pacific Ocean. J. Climate, 27, 592 - 606.471

Smirnov, D., M. Newman, M.A. Alexander, Y-O. Kwon, and C. Frankignoul, 2014b: In-472

vestigating the Local Atmospheric Response to a Realistic Shift in the Oyashio Sea473

Surface Temperature Front. J. Climate, 28, 1126 - 1147.474

20

Figure Captions475

Figure 1: North Atlantic wintertime (DJF) a) climatological-mean SST (contours) and ~u850476

(vectors) and b) standard deviation of SST (�SST ). The boxed region spans 37.5�N -477

45�N, 72�W - 42�W and indicates the region used to calculate the GSST index. Units for478

SST and �SST are in Kelvin. Units for ~u are in m/s.479

480

Figure 2: (Left column) wintertime lag regressions of Z1000 (contours) and SST (shading)481

onto the standardized GSST index, with negative (positive) lags denoting Z1000/SST anoma-482

lies leading (lagging) GSST . (Middle, right columns) linear decomposition of Z1000 where483

the anomalous field is decomposed into two parts: (middle column) the linear fit of Z1000 to484

the total -20-day lag regression map and (right column) the residual Z1000. The total SST485

regression at each lag is shown in all three columns. Z1000 contours are spaced at 4 meters486

(-6, -2, 2, 6...m), where solid (dashed) lines indicate positive (negative) anomalies. Note that487

at each lag, mapleft = mapmiddle +mapright.488

489

Figure 3: Regressions (contours) and correlations (shading) of Z1000 against the standardized490

GSST index at a lag of +20 days. Stippling indicates significance at the 95% confidence level491

using the 2-tailed Student’s t-test. Degrees of freedom are calculated using the method in492

Santer et al. 2000.493

494

Figure 4: Same as in Fig. 2, except for the linear decomposition of both ✓850 (shading)495

and Z500 (contours). Z500 contours are spaced at 6 meters (-9, -3, 3, 9...m).496

497

Figure 5: Daily lag regressions of SST (shading) and Z1000 (contours) onto the standard-498

ized Gatmos index. Z1000 contours with solid (dashed) lines represent positive (negative)499

values at an interval spacing of 10 meters (-15, -5, 5, 15...m). Positive lags indicate the500

Z1000/SST lagging the Gatmos index.501

21

502

Figure 6: 850hPa wintertime patterns of anomalous horizontal temperature advection asso-503

ciated with all three terms on the RHS of Eq. 2 at lag 0: (I) advection of the climatological-504

mean temperature gradient by the anomalous flow, (II) advection of the anomalous tem-505

perature gradient by the climatological-mean flow, and (III) advection of the anomalous506

temperature gradient by the anomalous flow. Contours represent the spatial temperature dis-507

tribution, vectors represent the wind, and shading represents temperature advection. Units508

for SST, ~u, and temperature advection are in Kelvin, m/s, and K/day, respectively. The509

purple box in panel II indicates the region averaged for the cross-section in Figure 7.510

511

Figure 7: 60�W - 30�W averaged vertical cross-section of ✓ (shading) and v, w (vectors)512

regressed onto the standardized GSST index at a lag of +20 days. v,w vectors are in units513

of ms, with w scaled by a factor of 2⇥ 103 for qualitative comparison.514

515

Figure 8: Same as in Fig. 2 (bottom left), except for SST and Z1000 regressed onto the516

standardized GSST index over the Northern Hemisphere at a lag of +20-days.517

518

Figure 9: Similar to Fig. 4, expect for the winter season (NDJFM) monthly-mean a) total519

regression of ✓850 (shading) and Z500 (contours) onto the standardized monthly-mean GSST520

index. Panels b, c show the linear decomposition of the seasonal ✓850 and Z500 anomalies521

into b) a linear fit to the -20-day lag pattern (from the daily decomposition in Fig. 4, top522

left), and c) a residual. Z500 contours are spaced at 6 meters (-9, -3, 3, 9...m) where solid523

(dashed) lines represents positive (negative) values.524

22

Figures525

(a) SST and ~u850

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

K

270 275 280 285 290 295

(b) �SST

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

K

0 0.5 1 1.5

Figure 1: North Atlantic wintertime (DJF) a) climatological-mean SST (contours) and ~u850 (vectors) andb) standard deviation of SST (�SST ). The boxed region spans 37.5�N - 45�N, 72�W - 42�W and indicatesthe region used to calculate the GSST index. Units for SST and �SST are in Kelvin. Units for ~u are in m/s.

23

Lag-20

Z1000 and SST

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

Z1000 fit and SST

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

Z1000 residual and SST

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

-10

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

0

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

+10

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

+20

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

K

σ

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1

Figure 2: (Left column) wintertime lag regressions of Z1000 (contours) and SST (shading) onto the stan-dardized GSST index, with negative (positive) lags denoting Z1000/SST anomalies leading (lagging) GSST .(Middle, right columns) linear decomposition of Z1000 where the anomalous field is decomposed into twoparts: (middle column) the linear fit of Z1000 to the total -20-day lag regression map and (right column)the residual Z1000. The total SST regression at each lag is shown in all three columns. Z1000 contours arespaced at 4 meters (-6, -2, 2, 6...m), where solid (dashed) lines indicate positive (negative) anomalies. Notethat at each lag, mapleft = mapmiddle +mapright.

24

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

r

−0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15 0.2

Figure 3: Regressions (contours) and correlations (shading) of Z1000 against the standardized GSST index ata lag of +20 days. Stippling indicates significance at the 95% confidence level using the 2-tailed Student’st-test. Degrees of freedom are calculated using the method in Santer et al. 2000.

25

Lag-20

✓850 and Z500

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

✓850 and Z500 fit

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

✓850 and Z500 residual

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

-10

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

0

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

+10

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

+20

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

K

σ

−1 −0.5 0 0.5 1

Figure 4: Same as in Fig. 2, except for the linear decomposition of both ✓850 (shading) and Z500 (contours).Z500 contours are spaced at 6 meters (-9, -3, 3, 9...m).

26

Lag0

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

+10

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

+20

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

+30

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

+40

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

!

K

σ

" #$% # #$% "

Figure 5: Daily lag regressions of SST (shading) and Z1000 (contours) onto the standardized Gatmos index.Z1000 contours with solid (dashed) lines represent positive (negative) values at an interval spacing of 10meters (-15, -5, 5, 15...m). Positive lags indicate the Z1000/SST lagging the Gatmos index.

27

(I) ~uA~5TC

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

(II) ~uC~5TA

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

(III) ~uA~5TA

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

K

day

−0.5 −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5

Figure 6: 850hPa wintertime patterns of anomalous horizontal temperature advection associated with allthree terms on the RHS of Eq. 2 at lag 0: (I) advection of the climatological-mean temperature gradient bythe anomalous flow, (II) advection of the anomalous temperature gradient by the climatological-mean flow,and (III) advection of the anomalous temperature gradient by the anomalous flow. Contours represent thespatial temperature distribution, vectors represent the wind, and shading represents temperature advection.Units for SST, ~u, and temperature advection are in Kelvin, m/s, and K/day, respectively. The purple boxin panel II indicates the region averaged for the cross-section in Figure 7.

28

30 40 50 60 70

200

400

600

800

1000

Latitude

Pre

ssu

re (

hP

a)

!

K

σ

" #$% # #$% "

Figure 7: 60�W - 30�W averaged vertical cross-section of ✓ (shading) and v, w (vectors) regressed onto thestandardized GSST index at a lag of +20 days. v,w vectors are in units of m

s, with w scaled by a factor of

2⇥ 103 for qualitative comparison.

29

120

o W

60 o

W

0o

60

o E

120 o

E

180oW

!

K

σ

" #$% # #$% "

Figure 8: Same as in Fig. 2 (bottom left), except for SST and Z1000 regressed onto the standardized GSST

index over the Northern Hemisphere at a lag of +20-days.

30

(a) Seasonal ✓850 and Z500

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

(b) Fit to map-20

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

(c) Residual ✓850 and Z500

90 oW

60oW 30

oW

0o

30 oN

45 oN

60 oN

K

σ

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1

Figure 9: Similar to Fig. 4, expect for the winter season (NDJFM) monthly-mean a) total regression of ✓850(shading) and Z500 (contours) onto the standardized monthly-mean GSST index. Panels b, c show the lineardecomposition of the seasonal ✓850 and Z500 anomalies into b) a linear fit to the -20-day lag pattern (fromthe daily decomposition in Fig. 4, top left), and c) a residual. Z500 contours are spaced at 6 meters (-9, -3,3, 9...m) where solid (dashed) lines represents positive (negative) values.

31