1

Response of the tropical Pacific to abrupt climate change 8,200 1

years ago 2

3

Alyssa R. Atwood1^* David S. Battisti2, C. M. Bitz2, Julian P. Sachs1 4

5 1 University of Washington, School of Oceanography, Seattle, WA 98195, USA 6 2 University of Washington, Dept. of Atmospheric Sciences, Seattle, WA 98195, USA 7

^ Now at: University of California, Berkeley, Dept. of Geography, Berkeley, CA 94720, 8

USA and Georgia Institute of Technology, School of Earth and Atmospheric Sciences, 9

Atlanta, GA 30332, USA 10

11

12

13

14

15

* Corresponding author. Tel: +1 206-694-3143; fax: +1 206-685-3351. 16

E-mail address: aatwood@berkeley.edu (Alyssa Atwood). 17

18

19

2

The relatively stable climate of the Holocene epoch (11,700 yr BP-present) was 20

punctuated by a period of large and abrupt climate change ca. 8,200 yr BP, when an 21

outburst of glacial meltwater into the Labrador Sea drove large and abrupt climate 22

changes across the globe. However, little is known about the response of the tropical 23

Pacific to this event. Here we present the first evidence for large perturbations to 24

the eastern tropical Pacific climate, based on sedimentary biomarker and hydrogen 25

isotopic records from a freshwater lake in the Galápagos Islands. We inform these 26

reconstructions with freshwater forcing simulations performed with the Community 27

Climate System Model version 4 in tandem with an intermediate complexity model 28

of the tropical Pacific. Together, the biomarker records and model simulations 29

provide evidence for a mechanistic link between (1) a southward shift of the 30

Intertropical Convergence Zone in the eastern equatorial Pacific and (2) decreased 31

frequency and/or intensity of Eastern Pacific El Niño events during the 8,200 BP 32

event. These results provide valuable insight into the controls of tropical Pacific 33

climate variability and the mechanisms behind the global response to abrupt climate 34

change. 35

36

The ability of the Earth’s climate system to undergo dramatic and rapid 37

reorganization on decadal timescales is exemplified by abrupt climate transitions, such as 38

those that occurred over the last glacial cycle in association with variations in the strength 39

of the Atlantic Meridional Overturning Circulation (AMOC). These events stretch our 40

understanding of the dynamical principles that govern the climate system, given the lack 41

of these events in the modern record and the inability of climate models to reproduce 42

such variability. The most recent event occurred ca. 8,200 yr BP during the relatively 43

stable climate of the Holocene epoch. Paleoclimate data and modeling studies suggest 44

that this event was caused by the catastrophic drainage of glacial lakes formed by the 45

retreating North American ice sheet into the Labrador Sea, which produced a sufficiently 46

large negative salinity anomaly to weaken the Atlantic Meridional Overturning 47

Circulation (AMOC) and cause dramatic cooling across the North Atlantic1-4. 48

Proxy records indicate that an estimated 163,000 km3 of glacial meltwater was 49

released through the Hudson Bay in one or two pulses during this event2,5-8. Polar ice and 50

3

marine records indicate that annual average surface temperatures dropped by 2-6 °C in 51

central Greenland (Fig. 1B)2,9 and by 1-3 °C in the North Atlantic Ocean and Europe 52

(Fig. 1C)6,7,10,11. The associated climate perturbations are generally thought to have 53

persisted for 100-150 years7,9,12, though some sedimentary records suggest that ocean 54

temperature and circulation anomalies in the North Atlantic may have followed a multi-55

step pattern over several centuries6. 56

The 8200 BP event is marked by reduced concentrations of atmospheric 57

methane13 (Fig. 1D), attributed to a decrease in tropical wetland area1,9 associated with a 58

southward shift of tropical continental rainfall13,14. Consistent with this interpretation, 59

proxy records from continental regions provide evidence for an intensified South 60

American summer monsoon15,16 (Fig. 1K), a weakened Indian (and possibly Asian) 61

summer monsoon15,17-20 (Fig. 1H,I), and arid conditions in the North Africa monsoon 62

region21-23. Taken together, this collection of continental hydroclimate records provide 63

evidence for a large-scale southward shift of tropical continental precipitation during the 64

8200 BP event. 65

However, little is known about the response of rainfall over the oceans, where 66

interactions between the tropical ocean and atmosphere produce low level wind 67

convergence and atmospheric convection that form organized rain bands known as 68

intertropical convergence zones (ITCZ). Robust ocean rainfall reconstructions during this 69

event are hindered by the low temporal resolution of marine sediments and the relatively 70

small changes in ocean chemistry and biology brought about by rainfall variations. In 71

particular, the behavior of tropical Pacific climate variability across the 8200 BP event is 72

poorly constrained. The tropical Pacific has a disproportionately large impact on global 73

climate on interannual timescales resulting from the El Niño/Southern Oscillation 74

(ENSO) and it is believed to play a fundamental role in driving global climate variability 75

on decadal to orbital timescales. Hence, improved constraints are needed on these key 76

features of the climate system to fully understand the nature of the global climate 77

response to this abrupt climate event. 78

Here we present a novel set of hydroclimate proxy records, based on the 79

accumulation rate and hydrogen isotopic composition of sedimentary lipid biomarkers 80

from El Junco Lake in the Galápagos Islands to reconstruct changes in eastern equatorial 81

4

Pacific rainfall over the last 9,100 years. These records provide evidence for large 82

perturbations to tropical Pacific hydroclimate associated with the eastern Pacific ITCZ 83

and El Niño events around the time of the 8200 BP event. 84

To support the interpretation of these rainfall reconstructions, we invoke climate 85

theory and simulations with a state-of-the-art climate system model. While dynamical 86

theory supports the proxy evidence of a climatological southward shift of the ITCZ in 87

response to abrupt North Atlantic cooling, a theoretical basis to support a concomitant 88

change in ENSO variability is less well developed, though a substantial body of literature 89

indicates that ENSO is modulated by changes in the mean state of the tropical 90

Pacific24,e.g. 25,26-28. In order to evaluate the linkage between abrupt North Atlantic 91

cooling, a southward shift of the Pacific ITCZ and changes in ENSO variability (as 92

suggested by the El Junco Lake rainfall reconstructions), we performed simulations with 93

the Community Climate System Model version 4 (CCSM4) in tandem with an 94

intermediate complexity model of the tropical Pacific that explicitly evaluates the 95

influence of tropical Pacific mean state changes on the climate variability. Through 96

pairing such paleoclimate records with dynamical theory and numerical models, we shed 97

new light on the mechanisms behind the global response to abrupt climate change. 98

99

EL JUNCO LAKE BIOMARKER RECORDS 100

El Junco Lake is located on San Cristóbal Island, Galápagos (1°S, 89°W) in the 101

heart of the eastern equatorial Pacific cold tongue (Fig. 2). Rainfall in the modern climate 102

is governed by the annual migration of the ITCZ on seasonal timescales and by increased 103

convection during eastern Pacific El Niño events on interannual timescales29. On long 104

timescales, rainfall at El Junco Lake is thus likely sensitive to long-term changes in both 105

the position of the eastern Pacific ITCZ and the intensity and frequency of eastern Pacific 106

El Niño events. 107

Based on the findings presented in Atwood and Sachs (2014) and updated in Atwood 108 30, a novel set of proxies based on the hydrogen isotopic composition 109

(δ2H=((2H/1H)sample/(2H/1H)std-1)*1000) and accumulation rate of phytoplankton lipids in 110

sediments are used to infer changes in (1) climatological total mean annual rainfall at El 111

5

Junco Lake ∆ (Fig. 3A), and (2) climatological rainfall at El Junco Lake associated 112

with El Niño events∆ (Fig. 3B; i.e. the total rain that fell during El Niño events 113

divided by the total amount of time in the sedimentary interval). Changes in 114

climatological rainfall that are not associated with changes in El Niño rainfall (which we 115

term residual rainfall, ∆ ; Fig. 3C) can be inferred for periods in which the inferred 116

change in total rainfall opposes the change in rainfall associated with El Niño events, 117

where we limit our interpretation of ∆ to its sign during such periods (see 118

Supplementary Material). ∆ is based on δ2H values of dinosterol, a biomarker 119

produced by a dinoflagellate of the genus Peridinium in El Junco Lake 31, while ∆ is 120

based on the accumulation rate of C34 botryococcene ( ), a biomarker produced by 121

a single race of the green algae Botryococcus braunii that appears to bloom in response to 122

decreased nutrient concentrations associated with overflowing conditions in El Junco 123

Lake during eastern Pacific El Niño events32 (see the Supplementary Material for more 124

information). 125

126

RECONSTRUCTED TROPICAL PACIFIC CLIMATE CHANGES DURING THE 8200 127

BP EVENT 128

The El Junco Lake biomarker records indicate that total mean annual rainfall at El 129

Junco Lake (∆ ) varied substantially over timescales of centuries to millennia during the 130

past 9,100 years (Fig. 3B). Of the 9100-year record, the highest inferred values of 131 ∆ occur from ca. 8500-8200 yr BP and are preceded and followed by lower rainfall ca. 132

8700 yr BP and 8100-7100 yr BP, respectively. Concomitant with this inferred increase 133

in total climatological rainfall, the sedimentary flux of botryococcenes decreased to some 134

of the lowest values of the 9100-yr record ca. 8500-8000 yr BP (Fig. 3B). We interpret 135

this low botryococcene flux as indicative of reduced frequency and/or intensity of El 136

Niño-driven B. braunii blooms during this time. From these El Junco Lake biomarker 137

records, we thus infer increased total rainfall but reduced rainfall associated with El Niño 138

events around the time of the 8200 BP event. 139

6

Wet conditions in the Galápagos Islands around the time of the 8200 BP event 140

coincide within age uncertainty to reduced climatological rainfall in Costa Rica33 (Fig. 141

1J), and increased climatological wind-driven upwelling in the Cariaco Basin34 (Fig. 1L), 142

which was interpreted by the authors as strengthened trade winds off the coast of 143

Venezuela. Collectively, these paleoclimate records provide strong evidence for a 144

climatological southward shift of the ITCZ in the eastern Pacific and western Atlantic 145

Oceans during the 8200 BP event (Fig. 1; Fig. 2). Importantly, the El Junco Lake 146

biomarker records provide the first evidence of a southward shifted ITCZ from a 147

maritime region that is free of monsoonal influences. 148

Concomitant with the inferred increase in climatological rainfall at El Junco Lake 149

(∆ ) ca. 8500-8000 yr BP (Fig. 1M), the low sedimentary flux of botryococcenes 150

provides evidence for reduced frequency and/or intensity of eastern Pacific El Niño 151

rainfall during this time (Fig. 1E). Further evidence of reduced El Niño rainfall comes 152

from grainsize records in El Junco Lake sediment35 (Fig. 1G). Akin to the botryococcene 153

flux variations, sedimentary sand fraction (% by mass) was interpreted by the authors as a 154

proxy for changes in the frequency and/or intensity of heavy rainfall that occurs at El 155

Junco Lake during strong El Niño events. The covariance between sand content and 156

botryococcene flux supports the existence of such a shared driver of variations in these 157

two proxy records through the Holocene (Fig. 1E,G). Low sand content coincident with 158

reduced botryococcene flux thus provides an independent estimate of reduced El Niño 159

rainfall around the time of the 8200 BP event. 160

A final line of evidence of reduced ENSO variability during the 8200 BP event 161

comes from a high-resolution speleothem record in Malaysian Borneo36 (Fig. 1F and Fig. 162

S1). Located in the western equatorial Pacific (4°N, 114°E), interannual rainfall 163

variability at the Bukit Assam cave site is primarily driven by ENSO, with rainfall 164

deficits and positive rainfall δ18O anomalies occurring during El Niño years36,37. Variance 165

in the 2-7-year bandpass filtered δ18Ocalcite (i.e. ENSO band) is shown in Fig. 1F 166

(comprised of six sub-annually resolved records of length 60-210 years spanning 8320-167

8140, 6740-6580, 5710-5630, 5230-5160, 3380-3230, 2600-2390 yr BP). These three 168

highly disparate proxy records (botryococcene concentration and sand content from a 169

Galápagos lake and speleothem δ18O from Borneo) show good overall correspondence 170

7

through the Holocene, with variable activity during the early Holocene (6-9 kyr BP), low 171

variability during the mid-Holocene (4-6 kyr BP) and increasing variability during the 172

late Holocene (2-4 kyr BP). The correspondence between these three proxy records 173

bolsters our interpretation of a common ENSO signal in these records. In addition, the 174

variance in Borneo speleothem δ18O reaches a minimum from 8220-8170 yr BP, relative 175

to all other 50-year intervals in the Holocene record, which falls within the (broader) 176

local minima in the El Junco Lake botryococcene concentration and sand records (Fig. 177

1E-G; Fig. S1). Taken together, these three records provide strong evidence of weak 178

ENSO variability across the tropical Pacific during the 8200 BP event. 179

TROPICAL PACIFIC RESPONSE TO ABRUPT NORTH ATLANTIC COOLING FROM 180

THEORY AND MODELS 181

It has been well established from theory and idealized climate model simulations 182

that the atmosphere responds to extratropical heating and cooling through meridional 183

displacements of the zonal mean ITCZ towards the heated hemisphere, in order to 184

balance the imposed hemispherically-asymmetric energy input38,39. A similar response is 185

seen in GCMs under a prescribed freshwater input to the North Atlantic, where a 186

southward shift of the zonal mean climatological position of the ITCZ follows a 187

weakening of the AMOC and expansion of Arctic sea ice40-43. However, while the zonal 188

mean ITCZ response is robust across models of varying complexity, the spatial 189

heterogeneity of the ITCZ response can be large. In particular, while the southward shift 190

of the Atlantic ITCZ is large and robust, the Pacific ITCZ response is typically weaker 191

and less robust across models. Importantly, the position of the Pacific ITCZ represents a 192

key feature of the background state of the tropical Pacific that modulates the 193

characteristics of ENSO. 194

While theory and modeling studies support the proxy evidence for a southward 195

shifted ITCZ during the 8200 BP event, a theoretical basis for the concomitant reduction 196

in El Niño rainfall variability is less clear. In order to evaluate this aspect of the tropical 197

Pacific response to a freshwater perturbation in the North Atlantic and thereby provide a 198

more robust interpretation of the El Junco Lake hydroclimate reconstructions, we 199

performed simulations with CCSM4 in tandem with the Linearized Ocean Atmosphere 200

8

Model44-46, an intermediate complexity model of the tropical Pacific that explicitly 201

evaluates the influence of tropical Pacific mean state changes on the coupled system. 202

In response to the large (1 Sv) prescribed freshwater perturbation in the North 203

Atlantic, the AMOC weakens in CCSM4 by 85% within 25 years after the hosing is 204

initiated and the Southern Ocean overturning cell extends into the North Atlantic (Fig. 205

S2). The AMOC weakening produces a large reduction in northward ocean heat transport 206

that causes the areal extent of sea ice to expand throughout the Arctic and sea surface 207

temperatures to decrease across the NH (and increase in the SH; Fig. S3A-F). The 208

resultant hemispherically asymmetric changes in surface energy fluxes that are initiated 209

by the weakening of the AMOC and intensified by Arctic sea ice growth produce a 210

strengthening (weakening) of the NH (SH) Hadley cell and a southward shift of the mean 211

annual position of the ITCZ. In association with these changes to the Hadley circulation, 212

an anomalous surface pressure gradient develops between the North and South Atlantic 213

that drives northerly wind anomalies in the tropical Atlantic (Fig. S3G-I). The evolution 214

of anomalous SSTs, surface pressure and winds is consistent with the mechanism of 215

equatorward cooling propagation proposed by Chiang and Bitz 47 and Cheng, et al. 41 216

whereby cooling is propagated from the North Atlantic high latitudes to the subtropics 217

through the ocean gyre circulation and further propagated to the tropical Atlantic through 218

the wind-evaporation-SST (WES) feedback.. 219

Large changes in the background state of the tropical Pacific are simulated in 220

CCSM4 in response to the freshwater perturbation. The WES feedback appears to 221

propagate the cooling into the tropical Pacific, whereby the increased surface pressure 222

gradient between the cooled tropical Atlantic and the tropical Pacific drives a 223

northeasterly wind anomaly on top of the background northeasterly trades, increasing the 224

surface wind speed and evaporative cooling across central America to the eastern and 225

central Pacific (Fig. 4A; Fig. S4). The amplitude of the annual cycle of SSTs in the Niño3 226

region is reduced by 40% under hosing. In the annual mean, surface easterlies increase by 227

20-70% near along the equator and in the tropical north Pacific, SST cools by ~1°C 228

across the tropical Pacific, the thermocline warms by up to 2°C along the equator in the 229

Pacific (Fig. 4), and the ITCZ shifts southward in the eastern and central Pacific (Fig. 5; 230

Fig. S5). In CCSM4, this southward ITCZ shift is the result of the weakening of the NH 231

9

ITCZ branch and strengthening and southward shift of the SH ITCZ branch in the central 232

and eastern Pacific. A southward shift of precipitation in the eastern Pacific and in 233

central-to-South America occurs in both the model simulations and the proxy records of 234

the 8200 BP event (Fig. 5). Although there is not perfect point-to-point correspondence 235

between the simulated and reconstructed rainfall changes, considering the strengths and 236

limitations of both the model and data sets, there is general agreement in their large-scale 237

dynamical context. 238

Associated with this southward ITCZ shift is an increase in easterly zonal wind 239

stress that maximizes under the northern branch of ITCZ and along the equator in the 240

eastern and central Pacific (Fig. 4B). This pattern of anomalous climatological zonal 241

wind stress is a key contributor to the ENSO response in CCSM4, as described in the 242

subsequent section. ENSO weakens under the freshwater release in the North Atlantic, as 243

demonstrated by a 23% reduction in the variance of Niño 3 SSTAs in CCSM4 hosed 244

simulations, which is significant at the 95% confidence interval using an F test of 245

variances (see Methods; Fig. S6). The reduction in variance is even more pronounced 246

(32%) when evaluated using the 2-7 year bandpass filtered time series (data not shown). 247

The southward shift of the Pacific ITCZ and weakened ENSO variability simulated by 248

CCSM4 under abrupt North Atlantic cooling agrees with the hydroclimate changes in 249

Galápagos Islands and Borneo during the 8200 BP event. 250

251

RESPONSE OF ENSO TO AN EQUATORWARD SHIFT OF THE PACIFIC ITCZ 252

In order to evaluate the impact of the simulated changes in the tropical Pacific 253

climatological annual cycle on ENSO, we performed simulations with the Linearized 254

Ocean Atmosphere Model (LOAM). Two simulations were performed in which the 255

tropical Pacific mean state variables were prescribed in LOAM based on the CCSM4 256

preindustrial control run and hosed run, respectively. The first EOF of tropical Pacific 257

SST variability in the CCSM4 and LOAM simulations is shown in Fig. S7. In agreement 258

with the fully coupled CCSM4, ENSO variability (as defined by variance of Niño 3 259

SSTAs) in LOAM decreased when the prescribed mean states from the CCSM4 control 260

were replaced by the mean states from the hosed run (Table 1), although the simulated 261

reduction in Niño 3 variance is considerably larger in LOAM (58%) than that in CCSM4 262

10

(23%). It is possible that nonlinearities in CCSM4 (which are absent in LOAM) may act 263

to compensate for the large (mean state induced) changes in linear stability, thereby 264

tempering the sensitivity of ENSO to the mean state changes in CCSM4. A comparison 265

of the eigenmodes of the coupled system in the two LOAM simulations demonstrates that 266

the decreased ENSO variance under hosing is due to a decrease in the growth rate of the 267

ENSO mode (as opposed to changes in the lower-order modes of the system; Table 1). 268

Sensitivity tests performed in LOAM demonstrate that the reduced growth rate of 269

the ENSO mode under hosing is primarily due to a deeper thermocline and secondarily 270

due to a more diffuse thermocline (decreased ), in the equatorial Pacific (Table 1; Fig. 271

4C). The growth rate is further reduced by the large-scale surface cooling in the tropical 272

Pacific (Fig. 4A). Details of the sensitivity tests can be found in the Supplementary 273

Material. Physically, the influence of these mean state changes can be understood in the 274

following way. In the first case, the equatorial subsurface warming decreases the 275

efficiency with which wind stress anomalies produce SST anomalies (as a given 276

thermocline depth anomaly on top of a deeper and more diffuse thermocline yields a 277

smaller change in SST). In the second case, the colder tropical Pacific mean state SSTs 278

decrease the response of the atmosphere to a given SST anomaly28. The source of the 279

subsurface warming appears to be associated with a dipole pattern in the windstress curl 280

anomaly along the equator, with negative (positive) windstress curl anomalies north 281

(south) of the equator (Fig. 4B). Following the theory of Clarke, et al. 48, the anomalous 282

windstress curl integrated along the northern and southern boundaries of the equatorial 283

Pacific basin between 5ºS-5ºN modify the warm water volume in the equatorial upper 284

ocean through Sverdrup transport. The dipole pattern of anomalous windstress curl is 285

thus consistent with the meridional transport of (warm) water into the equatorial upper 286

ocean, deepening the equatorial thermocline. 287

In summary, simulations with CCSM4 and an intermediate complexity model 288

support the available reconstructions of tropical Pacific climate change during the 8200 289

BP event and demonstrate a mechanistic link between abrupt North Atlantic cooling, a 290

southward shift of the Pacific ITCZ, and reduced ENSO variability. The simulations 291

indicate that the global climate response to North Atlantic freshwater release includes 292

hemispherically asymmetric changes in surface energy fluxes (initiated by the weakening 293

11

of the AMOC and intensified by the positive feedback from Arctic sea ice growth) that 294

produce a southward shift of the ITCZ and drive increased easterly trade winds across 295

central America and into the eastern and central tropical Pacific. Equatorial subsurface 296

warming occurs under the southward shift of the Pacific ITCZ through the associated 297

changes in wind stress curl. The coupled ocean-atmosphere interactions that give rise to 298

ENSO are stabilized by both the equatorial Pacific subsurface warming and by the large-299

scale surface cooling. 300

These simulations outline a clear set of mechanisms by which abrupt North 301

Atlantic cooling gives rise to reduced ENSO variability in CCSM4. In contrast to the 302

proxy data and the results from CCSM4, a number of the models used to examine the 303

impact of hosing in the Paleoclimate Model Intercomparison Project Phase 2 (PMIP2) 304

simulated increased ENSO variability under hosing43,49. Importantly, many of these 305

PMIP2 models (CCSM2, CCSM3, HadCM3, ECHAM5-OM1, GFDL CM2.1) had very 306

large biases in the mean state, annual cycle and interannual variability of the tropical 307

Pacific43. The CMIP5/PMIP3 models generally demonstrate notable improvements in 308

their representation of the mean state of the tropical Pacific50. While tropical Pacific 309

mean state biases are still present in CCSM4 (Fig. S8; Supplementary Material) the 310

CCSM4 model used in the present study is recognized as a top-performing GCM with 311

regard to its simulation of tropical Pacific climate variability50,51. The overarching 312

agreement between the simulated response of the tropical Pacific in CCSM4 and that 313

found in proxy reconstructions represents a notable advancement in our understanding of 314

the response of the tropical Pacific to abrupt climate change. Further, the proposed 315

relationship between changes in the mean climate, including the Pacific ITCZ, and 316

ENSO, may help uncover new frameworks with which to constrain the response of 317

ENSO in a future of global warming. 318

319

ACKNOWLEDGEMENTS 320

This material is based upon work supported by the U.S. National Science Foundation 321

under Grants ESH-0639640, EAR-0823503 and the U.S. National Oceanic and 322

Atmospheric Administration under Grant No. NOAA-NA08OAR4310685 to J. Sachs. A. 323

Atwood was supported by the National Science Foundation Graduate Research 324

12

Fellowship Program (DGE-1256082) and the Department of Energy Global Change 325

Education Program. D.S. Battisti was supported by a grant from the Tamaki Foundation. 326

We would like to acknowledge high-performance computing support from Yellowstone 327

(ark:/85065/d7wd3xhc) provided by NCAR's Computational and Information Systems 328

Laboratory, sponsored by the National Science Foundation. The authors would like to 329

thank Z. Zhang for the botryococcene data and for sample preparation and J. Conroy as 330

well as Z. Zhang for the age model. We thank the following for assistance in the field and 331

logistical support: P. Colinvaux, J. Overpeck, M. Steinitz-Kannan, J. Conroy, R. 332

Smittenberg, the Galápagos National Park, and the Charles Darwin Research Station. 333

Finally, we thank O. Kawka, D. Nelson, J. Gregersen, S. Wakeham, and J. Volkman for 334

laboratory support and useful discussions that improved this manuscript. 335

336

AUTHOR CONTRIBUTIONS 337

A.R.A. and J.P.S. developed the concept and design for the geochemical measurements. 338

A.R.A. carried out the geochemical analyses. A.R.A, C.M.B and D.S.B designed and 339

performed the model simulations. 340

341

METHODS 342

343

Data availability 344

The El Junco Lake biomarker data are available online at NOAA National Centers for 345

Environmental Information (https://www.ncdc.noaa.gov/data-access/paleoclimatology-346

data/datasets). 347

348

Field methods 349

Overlapping sediment cores to 372 cm depth were collected from El Junco Lake in 350

September 2004 from 6 m water-depth using a Nesje piston corer (9 cm dia.). An adapted 351

Livingston-type corer (7 cm dia.) was used to recover a 75 cm-long core with an undisturbed 352

sediment-water interface. Unconsolidated near-surface sediment was sectioned at 1 cm intervals 353

on-site and was stored frozen prior to analysis. Whole cores were transported to J. Overpeck’s 354

lab at the University of Arizona where they were split, imaged, and placed on a common depth 355

13

scale. The resulting composite sediment sequence was subsampled in 0.5 or 1 cm increments at 5 356

cm intervals. 357

358

Sediment age model 359

Details of the age model can be found in Zhang, et al. 32. Briefly, the upper 15 cm of the 360

interface core was dated with 210Pb, using a constant rate of supply model. 17 samples from the 361

Nesje cores and 4 samples from the interface core were selected for radiocarbon measurements. 362

Due to lack of sufficient macrofossils, bulk sediment (treated with acid and base to remove 363

carbonates and humics, resulting in “humin”) was used. Overlapping radiocarbon dates were used 364

to build the composite sedimentary sequence. Age model uncertainty was calculated following 365

the method of Anchukaitis and Tierney 52, in which a Monte Carlo approach was used to re-366

sample the probability distributions of the dated samples 10,000 times to create an ensemble of 367

age models. The probability distributions of the 210Pb-dated samples were taken to be Gaussian 368

functions while those of the 14C-dated samples were obtained from OxCal 4.2 calibrations (Bronk 369

Ramsey, 2009), using the ShCal04 curve. The 68% confidence interval from the 10,000-member 370

ensemble was used as the age uncertainty in Fig. 3. This age error ranged from a few years in the 371

upper layers of the sediment to 150 years in the deepest layers. 372

373

Lipid extraction and purification 374

Details of the lipid extraction and purification procedures can be found in the 375

Supplementary Material of Atwood and Sachs 29. Sediments were freeze-dried and heneicosanol 376

(n-C21 alcohol) was added as a recovery standard prior to accelerated solvent extraction (Dionex 377

ASE 200) using three cycles of dichloromethane (DCM) and methanol (MeOH) (9:1) at 100 °C 378

and 1500 psi. Neutral and polar lipids were separated from the total lipid extract on aminopropyl 379

SPE cartridges (Burdick & Jackson, 500 mg/4 mL) with DCM and isopropyl alcohol (IPA) (3:1 380

v/v), followed by 4% acetic acid (HOAc) in diethyl ether (Et2O). The neutral fraction was applied 381

to a column of 5% water-deactivated Si gel and separated into hydrocarbon, wax ester, 382

sterol/alcohol, and polar fractions with hexane, 10% ethyl acetate (EtOAc) in hexane, DCM, and 383

MeOH, respectively. Botryococcenes were further separated from the hydrocarbon fraction by 384

column chromatography as described in Zhang, et al. 32. Dinosterol was purified from the 385

sterol/alcohol fractions via normal phase (NP)- and/or reverse phase (RP)- high performance 386

liquid chromatography (HPLC) based on the methods presented in Atwood and Sachs (2012). 387

Biomarker identification and quantification methods can be found in the Supplementary Material 388

of Atwood and Sachs 29 and in Zhang, et al. 32. 389

14

390

δ2H measurements 391

Hydrogen isotope measurements of dinosterol were performed on a Finnigan Delta V 392

Plus Isotope Ratio Mass Spectrometer (IRMS) coupled to a Thermo Trace GC Ultra with a 393

Varian VF-17ms FactorFour capillary column (60 m x 0.32 mm x 0.25 μm) and a pyrolysis 394

reactor according to the methods described in Atwood and Sachs 31. Isotope values, expressed as 395

δ2H values, were calculated in Isodat software relative to VSMOW using a co-injection standard 396

containing n-C32, n-C40, and n-C44 of known δ2H values (obtained from A. Schimmelmann, 397

Indiana University, Bloomington, IN, USA). The measured isotope values of dinosterol were 398

corrected for the addition of hydrogen atoms (of known δ2H value) that occurred during 399

acetylation. The mean values and standard errors are reported from 3-6 analyses that were 400

typically performed on each sample (up to 11 analyses on samples that received different HPLC 401

purification procedures; see Atwood and Sachs 31). Instrument performance was evaluated using 402

H2 reference gas and a prepared standard of n-alkanes of known δ2H values. 403

404

CCSM4 configuration 405

We perform simulations with version 1 of the CCSM4 global atmosphere/ocean/land/ice 406

model, which employs the CAM4 atmospheric model (resolution of 0.9° latitude and 407

1.25°�longitude) and the POP2 ocean model (resolution is uniform in longitude at 1.13°�and 408

varies in latitude from 0.27°-0.65°). The model configuration matches the 1,000-year CESM1 409

(also known as CCSM4) preindustrial control run, which has been well studied and existing 410

biases of the tropical Pacific climatology and variability are well known51. While some ENSO 411

characteristics in CCSM4 compare well with observations, substantial ENSO biases in CCSM4 412

exist, including overly strong variance and periodicity in CCSM4 relative to observations. 413

However, mean state biases and ENSO biases are substantially smaller in CCSM4 at 1°� 414

resolution (employed here) as compared to 2°�resolution51. 415

The hosing runs are branched from the 1,000-year pre-industrial control run of CCSM4. 416

In these runs, a prescribed freshwater flux of 1.0 Sv was applied for a period of 100 years across 417

the surface of the northern North Atlantic (50°-70°N). This freshwater flux was used in the 418

PMIP2 hosing simulations (https://pmip2.lsce.ipsl.fr/). Although this method likely overestimates 419

the total freshwater flux to the North Atlantic as compared to that during the 8200 BP event 420

(recent estimates suggest 5.2 Sv over approximately 1 year5), our target is a large reduction in the 421

AMOC over a long enough period of time to obtain robust ENSO statistics. All other boundary 422

15

conditions are fixed at 1850 AD values. An ensemble of four hosing experiments were 423

performed, each integrated for 100 years in order to evaluate the robustness of the ENSO 424

response to changes in initial conditions and allow for a robust representation of the ensemble 425

mean and spread. 426

427

Description of the Linearized Ocean Atmosphere Model 428

The Linearized Ocean Atmosphere Model (LOAM)44,45 is a linearized variant of the 429

Zebiak and Cane 24 intermediate complexity model of the tropical Pacific, updated to include 430

observationally constrained parameter values and observed climatological mean fields. LOAM is 431

comprised of a linearized two-layer atmosphere model (assumed to be in equilibrium with respect 432

to underlying ocean) coupled to a 1.5-layer ocean model governed by the linear shallow water 433

equations on an equatorial β-plane with an active upper layer that governs the evolution of SST. 434

The heating of the atmosphere is a function of SST and surface wind convergence53; the forcing 435

of the ocean is by the wind stress. 436

The dynamics are governed by: 437

/ = ( ) + (1) 438

where x is the state of the system (consisting of the dependent variables for the ocean: meridional 439

and zonal current, upwelling, thermocline depth, and SST perturbations), M(t) is the monthly 440

(and spatially) varying dynamical system matrix, and F is a stochastic forcing. The equations that 441

define the dynamical system matrix are linearized around the seasonally-varying climatological 442

mean state of the atmosphere and ocean. The coupled modes of the system are found by 443

decomposing the annual propagator matrix (R = eMt) into its Floquet modes (eigenmodes of the 444

cyclo-stationary annual propagator matrix). Each mode has an associated period and decay rate 445

that is derived from the eigenvalues of the mode. Thompson and Battisti 54 and Roberts and 446

Battisti 45 demonstrated that LOAM with observed background states supports a leading mode of 447

the coupled system that has a similar spatial structure, decay rate, and period to that observed. 448

The leading (slowest-decaying) Floquet mode in LOAM is thus referred to as the ENSO mode. 449

Further details on LOAM can be found in Roberts 55, Roberts and Battisti 45, and Atwood, et al. 46. 450

A summary of the constants and tuning parameters used in LOAM are provided in the 451

Supplementary material (Table S1; Fig. S9). 452

In the present study, the tropical Pacific mean states from 100 years of the CCSM4 pre-453

industrial control and all 100 years of one of the four hosed simulations are prescribed in LOAM 454

to evaluate the influence of the mean state changes on ENSO dynamics. Further, sensitivity tests 455

are performed to determine which of the mean state changes are responsible for the changes in 456

16

ENSO, by comparing ENSO properties in LOAM run with climatological mean states from the 457

control run to those in LOAM run with individually substituted mean states from the hosed runs. 458

459

Statistical Information 460

Following Russon, et al. 56 and von Storch and Zwiers 57, the significance of the 461

relative change in the variance of Niño3 SSTAs under North Atlantic hosing was 462

assessed by comparing the ratio of the sample variances (2 2) to the F distribution 463

with degrees of freedom given by k1/τ1 -1 and k2/τ2 -1, where k1 and k2 are the number of 464

time steps in 320 years of the CESM preindustrial control run and 320 years of the 465

compiled hosed run, respectively, and τ1 and τ2 represent the characteristic time scales, 466

which were obtained from the empirical estimate of the autocorrelation function of each 467

time series ( ), summed over the number of lag intervals equivalent to the first two 468

changes in sign (L). 469

= ; = − 1, − 1 ; = 1 + 2 (1) Using this framework: k1 = k2 = 3840 mo, τ1 = 15 mo, τ2 = 13 mo, F =

2 2 470

1.11° 0.87° = 1.27 (Fig. S6), and ( − 1, − 1) = 1.22at the 95% quantile. 471

Thus, the null hypothesis that the two sample variances are equal is rejected at the 95% 472

confidence level. 473

474

475

17

Tables and Figures 476

477 Table 1. Period and growth rate of the ENSO mode, and variance of Niño 3 SSTAs (3-478

month running mean) from the LOAM simulations with prescribed mean states from 479

100 years of the CESM preindustrial control run (Control) and a 100-year CESM 480

hosed run (Hosed). Control + X indicates LOAM run with all mean state fields from 481

the CESM preindustrial control run except X, and instead X is taken from the CESM 482

hosed run. u,v-vel represent the mean horizontal ocean currents averaged over the top 483

50m of the ocean, w-vel represents the mean upwelling at 50m, and U,V-winds 484

represent the zonal and meridional near-surface wind fields (i.e. lowest vertical level in 485

CAM). “subsfc temp” denotes a suite of mean state fields and parameter values that 486

affect the rate of surface warming due to entrainment of water below the surface mixed 487

layer; these include the depth of the climatological thermocline (ℎ), the mean vertical 488

temperature gradient at the base of the surface mixed layer ( ), and a set of parameter 489

values (Γ) that relate variations in temperature at the base of the mixed layer to 490

variations in thermocline depth (see Eqn. 12 of Roberts and Battisti 45). The variance 491

of Niño 3 SSTAs in 320 years of the CESM preindustrial control run and the last 80 492

years of the hosed CESM run (the ensemble member from which mean states were 493

prescribed in LOAM) is indicated in the top two rows. 494

Model Mean States Mode Period (yr)

Mode Growth (yr-1)

Variance (°C2)

CESM Control -- -- -- 1.11 CESM Hosed -- -- -- 0.72

LOAM

CESM Control 3.97 0.57 1.11 CESM Hosed 3.35 0.41 0.46 Control + SST

(atm only) (ocn only)

3.70 (3.87) (3.78)

0.52 (0.51) (0.58)

0.80 (0.84) (1.07)

Control + u,v,w-vel 4.69 0.57 1.54 Control + U,V-winds 4.00 0.71 2.20

Control + subsfc temp ( only)

(ℎ + Γ only)

3.43 (4.05) (3.35)

0.34 (0.52) (0.37)

0.32 (0.91) (0.37)

18

495

19

Figure 1. El Junco Lake biomarker records in comparison to North Atlantic and 496

tropical hydroclimate records encompassing the 8200 BP event. The highlighted 497

color bar and the diamond with error bars represent the timing of the terminal 498

Lake Agassiz outburst at 8.33 kyr with error ranges from 8.04 – 8.49 kyr 8. A) 499

Mean sortable silt size from a deep-sea sediment core in the subpolar North 500

Atlantic (core MD99-2251) indicating flow speed of the Iceland Scotland 501

Overflow Water (ISOW)6; B) bidecadal δ18O values of ice from the GISP2 ice 502

core58; C) subpolar North Atlantic SST reconstruction based on the relative 503

abundance of the polar foraminifer Neogloboquadrina pachyderma sinistral 504

coiling from core MD99-22516; D) mean atmospheric CH4 concentrations from 505

the Greenland ice core GRIP59; E) botryococcene accumulation rate from this 506

study (plotted on a log scale; modified from Zhang, et al. 32); F) variance of 2-7 yr 507

bandpass filtered speleothem δ18O (sub-annually resolved) in 60-yr sliding 508

windows from Bukit Assam Cave in Malaysian Borneo36 (red). The yellow circles 509

represent the variance across each entire high-resolution window and the yellow 510

square represents the variance across the 60-yr period following the large δ18O 511

enrichment at 8230 yr BP (see Fig. S1 for an enlarged view of these data); (G) 512

percent sand in El Junco Lake sediments 35; H) speleothem δ18O from Dongge 513

Cave, China20; I) speleothem δ18O from Qunf Cave, Oman60; J) speleothem δ18O 514

from Venado Cave, Costa Rica33; K) speleothem δ18O from Lapa Grande, 515

Brazil16; L) Cariaco Basin grey scale record34; M) dinosterol δ2H from El Junco 516

Lake, Galápagos (this study). 517

20

518

519

520

521

Figure 2. Climatological precipitation averaged over A) Aug–Oct and B) Feb–522

April for the period 1980–2010. Precipitation data are from GPCP Version 2.2 523

Combined Precipitation Data Set for the period 1979-2010 524

(http://www.esrl.noaa.gov/psd/data/gridded/data.gpcp.html). The proxy records 525

referred to in this study are indicated by the colored symbols, which consist of the 526

dinosterol δD record from El Junco Lake, Galápagos (1°S, 89°W; this study), 527

greyscale record from the Cariaco Basin34, speleothem δ18O and Mg/Ca from 528

Heshang Cave, China61 and speleothem δ18O from Venado Cave, Costa Rica33, 529

Dongge Cave, China20, Paixão Cave and Padre Cave, eastern Brazil15, Lapa 530

Grande Cave, Brazil16, Cueva del Tigre Perdido, Peru62, and Bukit Assam Cave 531

21

in Malaysian Borneo36. The changes in climatological precipitation and upwelling 532

associated with the 8200 BP event (compared to prior and after the event) are 533

noted by the color coding. 534

535 536

537

Figure 3. El Junco Lake biomarker records with the timing of Lake Agassiz 538

drainage highlighted. A) dinosterol δ2H (∆ ), indicative of changes in total 539

precipitation; B) botryococcene accumulation rate (plotted on a log scale; (∆ ), 540

indicative of changes in precipitation due to changes in the frequency and/or 541

amplitude of El Niño events adapted from Zhang, et al. 32; C) the sign of the 542

inferred change in climatological rainfall due to processes not associated with El 543

Niño (∆ ). Circles at the top of the figure represent the age control points of the 544

sediment record; unfilled circles indicate 14C dates, filled circles indicate 210Pb 545

dates. The highlighted color bar (and the diamond with error bars near the top of 546

the figure) represent the timing of the terminal Lake Agassiz outburst at 8.33 kyr 547

with error ranges from 8.04–8.49 kyr 8. 548

549

550

22

551 552 553 554

555 556

Figure 4. Tropical Pacific upper ocean and surface wind response to hosing in the 557

CESM. A) SST (colors) and near-surface wind velocity (vectors); B) wind stress 558

curl (colors) and change in zonal wind stress (unfilled contours; contour spacing is 559

0.005 N/m2 and the negative contours are dashed) with the region from 5ºS to 5ºN 560

highlighted (colors poleward of this region have been muted for clarity); C) mean 561

subsurface temperature (colors) averaged across the equatorial region (2ºS to 2ºN) 562

and mean subsurface temperature from the control run (unfilled contours; contour 563

spacing is 2ºC) where the bold line represents the 20ºC isotherm. Changes were 564

23

calculated by subtracting the average over last 80 years of the four hosed ensemble 565

members from the average of 320 years of the preindustrial control run. 566

567 568

569 570 Figure 5. Tropical precipitation response to hosing in the CESM (colored contours) 571

compared to the response inferred from hydroclimate proxy records during the 8200 BP 572

event (colored symbols; see Fig. 2 for references). Climatological precipitation in the 573

CESM control run is indicated by the unfilled contours (contour interval is 4 mm/day; the 574

4 mm/day contour is solid and the 8 mm/day contour is dashed). 575

24

References (including Methods) 576

577

1 Alley, R. B. et al. Holocene climatic instability: A prominent, widespread event 578 8200 yr ago. Geology 25, 483-486, doi:10.1130/0091-579 7613(1997)025<0483:hciapw>2.3.co;2 (1997). 580

2 Barber, D. C. et al. Forcing of the cold event of 8,200 years ago by catastrophic 581 drainage of Laurentide lakes. Nature 400, 344-348, doi:10.1038/22504 (1999). 582

3 LeGrande, A. N. et al. Consistent simulations of multiple proxy responses to an 583 abrupt climate change event. Proc. Nat. Acad. Sci. U.S.A. 103, 837-842, 584 doi:10.1073/pnas.0510095103 (2006). 585

4 Rohling, E. J. & Palike, H. Centennial-scale climate cooling with a sudden cold 586 event around 8,200 years ago. Nature 434, 975-979, doi:10.1038/nature03421 587 (2005). 588

5 Teller, J. T., Leverington, D. W. & Mann, J. D. Freshwater outbursts to the oceans 589 from glacial Lake Agassiz and their role in climate change during the last 590 deglaciation. Quat. Sci. Rev. 21, 879-887, doi:10.1016/s0277-3791(01)00145-7 591 (2002). 592

6 Ellison, C. R. W., Chapman, M. R. & Hall, I. R. Surface and deep ocean 593 interactions during the cold climate event 8200 years ago. Science 312, 1929-594 1932, doi:10.1126/science.1127213 (2006). 595

7 Kleiven, H. F. et al. Reduced North Atlantic Deep Water coeval with the glacial 596 Lake Agassiz freshwater outburst. Science 319, 60-64, 597 doi:10.1126/science.1148924 (2008). 598

8 Lewis, C. F. M., Miller, A. A. L., Levac, E., Piper, D. J. W. & Sonnichsen, G. V. 599 Lake Agassiz outburst age and routing by Labrador Current and the 8.2 cal ka 600 cold event. Quat. Int. 260, 83-97, doi:10.1016/j.quaint.2011.08.023 (2012). 601

9 Kobashi, T., Severinghaus, J. P., Brook, E. J., Barnola, J. M. & Grachev, A. M. 602 Precise timing and characterization of abrupt climate change 8200 years ago from 603 air trapped in polar ice. Quat. Sci. Rev. 26, 1212-1222, 604 doi:10.1016/j.quascirev.2007.01.009 (2007). 605

10 Hoffman, J. S. et al. Linking the 8.2 ka event and its freshwater forcing in the 606 Labrador Sea. Geophys. Res. Lett. 39, doi:10.1029/2012gl053047 (2012). 607

11 Morrill, C. & Jacobsen, R. M. How widespread were climate anomalies 8200 608 years ago? Geophys. Res. Lett. 32, doi:10.1029/2005gl023536 (2005). 609

12 Daley, T. J. et al. The 8200 yr BP cold event in stable isotope records from the 610 North Atlantic region. Global Planet. Change 79, 288-302, 611 doi:10.1016/j.gloplacha.2011.03.006 (2011). 612

13 Lea, D. W., Pak, D. K., Peterson, L. C. & Hughen, K. A. Synchroneity of tropical 613 and high-latitude Atlantic temperatures over the last glacial termination. Science 614 301, 1361-1364, doi:10.1126/science.1088470 (2003). 615

14 Chappellaz, J. et al. Synchronous changes in atmospheric CH4 and Greenland 616 climate between 40 kyr and 8 kyr BP. Nature 366, 443-445, 617 doi:10.1038/366443a0 (1993). 618

25

15 Cheng, H. et al. Timing and structure of the 8.2 kyr BP event inferred from delta 619 O-18 records of stalagmites from China, Oman, and Brazil. Geology 37, 1007-620 1010, doi:10.1130/g30126a.1 (2009). 621

16 Strikis, N. M. et al. Abrupt variations in South American monsoon rainfall during 622 the Holocene based on a speleothem record from central-eastern Brazil. Geology 623 39, 1075-1078, doi:10.1130/g32098.1 (2011). 624

17 Fleitmann, D. et al. Holocene ITCZ and Indian monsoon dynamics recorded in 625 stalagmites from Oman and Yemen (Socotra). Quat. Sci. Rev. 26, 170-188, 626 doi:10.1016/j.quascirev.2006.04.012 (2007). 627

18 Hong, Y. T. et al. Correlation between Indian Ocean summer monsoon and North 628 Atlantic climate during the Holocene. Earth Planet. Sci. Lett. 211, 371-380, 629 doi:10.1016/s0012-821x(03)00207-3 (2003). 630

19 Pausata, F. S. R., Battisti, D. S., Nisancioglu, K. H. & Bitz, C. M. Chinese 631 stalagmite delta O-18 controlled by changes in the Indian monsoon during a 632 simulated Heinrich event. Nat. Geosci. 4, 474-480, doi:10.1038/ngeo1169 (2011). 633

20 Wang, Y. J. et al. The Holocene Asian monsoon: Links to solar changes and 634 North Atlantic climate. Science 308, 854-857, doi:10.1126/science.1106296 635 (2005). 636

21 Gasse, F. Hydrological changes in the African tropics since the Last Glacial 637 Maximum. Quat. Sci. Rev. 19, 189-211, doi:10.1016/s0277-3791(99)00061-x 638 (2000). 639

22 Shanahan, T. M. et al. Paleoclimatic variations in West Africa from a record of 640 late Pleistocene and Holocene lake level stands of Lake Bosumtwi, Ghana. 641 Palaeogeog. Palaeoclimatol. Palaeoecol. 242, 287-302, 642 doi:10.1016/j.palaeo.2006.06.007 (2006). 643

23 Thompson, L. G. et al. Kilimanjaro ice core records: Evidence of Holocene 644 climate change in tropical Africa. Science 298, 589-593, 645 doi:10.1126/science.1073198 (2002). 646

24 Zebiak, S. E. & Cane, M. A. A model El Nino-Southern Oscillation. Monthly 647 Weather Review 115, 2262-2278, doi:10.1175/1520-648 0493(1987)115<2262:ameno>2.0.co;2 (1987). 649

25 Battisti, D. S. & Hirst, A. C. Interannual variability in a tropical atmosphere ocean 650 model- influence of the basic state, ocean geometry and nonlinearity. J. Atmos. 651 Sci. 46, 1687-1712, doi:10.1175/1520-0469(1989)046<1687:iviata>2.0.co;2 652 (1989). 653

26 Clement, A. C., Seager, R., Cane, M. A. & Zebiak, S. E. An ocean dynamical 654 thermostat. J. Clim. 9, 2190-2196, doi:10.1175/1520-655 0442(1996)009<2190:aodt>2.0.co;2 (1996). 656

27 Fedorov, A. V. & Philander, S. G. A stability analysis of tropical ocean-657 atmosphere interactions: Bridging measurements and theory for El Nino. J. Clim. 658 14, 3086-3101, doi:10.1175/1520-0442(2001)014<3086:asaoto>2.0.co;2 (2001). 659

28 Roberts, W. H. G., Battisti, D. S. & Tudhope, A. W. ENSO in the Mid-Holocene 660 according to CSM and HadCM3. J. Clim. 27, 1223-1242, doi:10.1175/jcli-d-13-661 00251.1 (2014). 662

26

29 Atwood, A. R. & Sachs, J. P. Separating ITCZ- and ENSO-related rainfall 663 changes in the Galápagos over the last 3 kyr using D/H ratios of multiple lipid 664 biomarkers. Earth Planet. Sci. Lett. 404, 408-419 (2014). 665

30 Atwood, A. R. Mechanisms of Tropical Pacific Climate Change During the 666 Holocene.” University of Washington, (2015). 667

31 Atwood, A. R. & Sachs, J. P. Purification of dinosterol from complex mixtures of 668 sedimentary lipids for hydrogen isotope analysis. Org. Geochem. 48, 37-46, 669 doi:10.1016/j.orggeochem.2012.04.006 (2012). 670

32 Zhang, Z. H., Leduc, G. & Sachs, J. P. El Nino evolution during the Holocene 671 revealed by a biomarker rain gauge in the Galapagos Islands. Earth Planet. Sci. 672 Lett. 404, 420-434, doi:10.1016/j.epsl.2014.07.013 (2014). 673

33 Lachniet, M. S. et al. Tropical response to the 8200 yr BP cold event? Speleothem 674 isotopes indicate a weakened early Holocene monsoon in Costa Rica. Geology 32, 675 957-960, doi:10.1130/g20797.1 (2004). 676

34 Hughen, K. A., Overpeck, J. T., Peterson, L. C. & Trumbore, S. Rapid climate 677 changes in the tropical Atlantic region during the last deglaciation. Nature 380, 678 51-54, doi:10.1038/380051a0 (1996). 679

35 Conroy, J. L., Overpeck, J. T., Cole, J. E., Shanahan, T. M. & Steinitz-Kannan, 680 M. Holocene changes in eastern tropical Pacific climate inferred from a 681 Galapagos lake sediment record. Quat. Sci. Rev. 27, 1166-1180, 682 doi:10.1016/j.quascirev.2008.02.015 (2008). 683

36 Chen, S., Hoffmann, S. S., Lund, D. C., Cobb, K. M. & Emile-Geay, J. A high-684 resolution speleothem record of western equatorial Pacific rainfall: Implications 685 for Holocene ENSO evolution. Earth Planet. Sci. Lett. 442, 61-71 (2016). 686

37 Moerman, J. W. et al. Diurnal to interannual rainfall δ18O variations in northern 687 Borneo driven by regional hydrology. Earth Planet. Sci. Lett. 369-370 (2013). 688

38 Frierson, D. M. W. & Hwang, Y. T. Extratropical Influence on ITCZ Shifts in 689 Slab Ocean Simulations of Global Warming. J. Clim. 25, 720-733, 690 doi:10.1175/jcli-d-11-00116.1 (2012). 691

39 Kang, S. M., Held, I. M., Frierson, D. M. W. & Zhao, M. The response of the 692 ITCZ to extratropical thermal forcing: Idealized slab-ocean experiments with a 693 GCM. J. Clim. 21, 3521-3532, doi:10.1175/2007jcli2146.1 (2008). 694

40 Zhang, R. & Delworth, T. L. Simulated tropical response to a substantial 695 weakening of the Atlantic thermohaline circulation. J. Clim. 18, 1853-1860, 696 doi:10.1175/jcli3460.1 (2005). 697

41 Cheng, W., Bitz, C. M. & Chiang, J. C. H. Adjustment of the global climate 698 system to abrupt changes in the Atlantic meridional overturning circulation. 699 Ocean Circulation: Mechanisms and Impacts, AGU Geophysical Monograph 700 Series 173, 295-313, doi:10.1029/173GM19 (2007). 701

42 Otto-Bliesner, B. L. & Brady, E. C. The sensitivity of the climate response to the 702 magnitude and location of freshwater forcing: last glacial maximum experiments. 703 Quat. Sci. Rev. 29, 56-73, doi:10.1016/j.quascirev.2009.07.004 (2010). 704

43 Timmermann, A. et al. The influence of a weakening of the Atlantic meridional 705 overturning circulation on ENSO. J. Clim. 20, 4899-4919, doi:10.1175/jcli4283.1 706 (2007). 707

27

44 Thompson, C. J. & Battisti, D. S. A linear stochastic dynamical model of ENSO. 708 Part I: Model development. J. Clim. 13, 2818-2832, doi:10.1175/1520-709 0442(2000)013<2818:alsdmo>2.0.co;2 (2000). 710

45 Roberts, W. H. G. & Battisti, D. S. A new tool for evaluating the physics of 711 coupled atmosphere-ocean variability in nature and in general circulation models. 712 Clim. Dyn. 36, 907-923, doi:10.1007/s00382-010-0762-x (2011). 713

46 Atwood, A. R., Battisti, D. S., Wittenberg, A. T., Roberts, W. H. G. & Vimont, D. 714 J. Characterizing unforced multi-decadal variability of ENSO: A case study with 715 the GFDL CM2.1 coupled GCM. J. Clim. 29, 1161-1178, doi:10.1007/s00382-716 016-3477-9 (2016). 717

47 Chiang, J. C. H. & Bitz, C. M. Influence of high latitude ice cover on the marine 718 Intertropical Convergence Zone. Clim. Dyn. 25, 477-496, doi:10.1007/s00382-719 005-0040-5 (2005). 720

48 Clarke, A. J., Van Gorder, S. & Colantuono, G. Wind stress curl and ENSO 721 discharge/recharge in the equatorial Pacific. J. Phys. Oceanogr. 37, 1077-1091, 722 doi:10.1175/jpo3035.1 (2007). 723

49 Dong, B. & Sutton, R. T. Enhancement of ENSO variability by a weakened 724 Atlantic thermohaline circulation in a coupled GCM. J. Clim. 20, 4920-4939, 725 doi:10.1175/jcli4284.1 (2007). 726

50 Bellenger, H., Guilyardi, E., Leloup, J., Lengaigne, M. & Vialard, J. ENSO 727 representation in climate models: from CMIP3 to CMIP5. Clim. Dyn. 42, 1999-728 2018, doi:10.1007/s00382-013-1783-z (2014). 729

51 Deser, C. et al. ENSO and Pacific Decadal Variability in the Community Climate 730 System Model Version 4. J. Clim. 25, 2622-2651, doi:10.1175/jcli-d-11-00301.1 731 (2012). 732

52 Anchukaitis, K. J. & Tierney, J. E. Identifying coherent spatiotemporal modes in 733 time-uncertain proxy paleoclimate records. Clim. Dyn. 41, 1291-1306 (2013). 734

53 Gill, A. E. Some simple solutions for heat-induced tropical circulation. Q. J. Roy. 735 Meteorol. Soc. 106, 447-462, doi:10.1256/smsqj.44904 (1980). 736

54 Thompson, C. J. & Battisti, D. S. A linear stochastic dynamical model of ENSO. 737 Part II: Analysis. J. Clim. 14, 445-466, doi:10.1175/1520-738 0442(2001)014<0445:alsdmo>2.0.co;2 (2001). 739

55 Roberts, W. An Investigation into the Causes for the Reduction in the Variability 740 of the El Niño-Southern Oscillation in the Early Holocene in a Global Climate 741 Model Ph.D. thesis, University of Washington, (2007). 742

56 Russon, T., Tudhope, A. W., Hegerl, G. C., Schurer, A. & Collins, M. Assessing 743 the Significance of Changes in ENSO Amplitude Using Variance Metrics. J. 744 Clim. 27, 4911-4922, doi:10.1175/jcli-d-13-00077.1 (2014). 745

57 von Storch, H. & Zwiers, F. W. Statistical Analysis in Climate Research. 746 (Cambridge Univ. Press, 2003). 747

58 Stuiver, M. & Grootes, P. M. GISP2 oxygen isotope ratios. Quat. Res. 53, 277-748 283, doi:10.1006/qres.2000.2127 (2000). 749

59 Blunier, T., Chappellaz, J., Schwander, J., Stauffer, B. & Raynaud, D. Variations 750 in atmospheric methane concentration during the Holocene epoch. Nature 374, 751 46-49, doi:10.1038/374046a0 (1995). 752

28

60 Fleitmann, D. et al. Holocene forcing of the Indian monsoon recorded in a 753 stalagmite from Southern Oman. Science 300, 1737-1739, 754 doi:10.1126/science.1083130 (2003). 755

61 Liu, Y.-H. et al. Links between the East Asian monsoon and North Atlantic 756 climate during the 8,200 year event. Nat. Geosci. 6, 117-120, 757 doi:10.1038/NGEO1708 (2013). 758

62 van Breukelen, M. R., Vonhof, H. B., Hellstrom, J. C., Wester, W. C. G. & 759 Kroon, D. Fossil dripwater in stalagmites reveals Holocene temperature and 760 rainfall variation in Amazonia. Earth Planet. Sci. Lett. 275, 54-60, 761 doi:10.1016/j.epsl.2008.07.060 (2008). 762

763

Response of the tropical Pacific to abrupt climate change 8,200 years ago

Alyssa R. Atwood, David S. Battisti, C. M. Bitz, Julian P. Sachs

Extended Data (supplementary figures)

Figure S1. A) Speleothem δ18O (sub-annually resolved) from Bukit Assam Cave in

Malaysian Borneo 1; B) the variance of the 2-7 yr bandpass filtered time series in 60-yr

sliding windows (centered on the mid-point of each window; as in Fig. 1 of the

manuscript). The yellow square represents the variance in the 60-yr period following the

large δ18O enrichment at 8230 yr BP and the yellow circle represents the variance across

the entire 180-yr high resolution window (as in Fig. 1).

Figure S2. Mean meridional streamfunction in the Atlantic in the CESM control and hosed

simulations: A) averaged over 320 years of the control simulation; B) averaged over the

last 80 years of the four hosed ensemble members.

Figure S3. Time evolution of the change in: A-C) sea ice concentration; D-F) SST (colors)

and wind stress (vectors); G-I) surface pressure (colors) and surface wind velocity (vectors)

in the CESM hosed runs with respect to 320 years of the control run. Time snapshots are

averages over: Year 0-1 (top), Year 10-11 (middle) and Year 50-51 (bottom). Each panel

shows the average of the four ensemble members.

Figure S4. Surface pressure (contours) from the control CESM integration and the

ensemble mean changes in the surface pressure (colors) and near-surface wind velocity

(vectors) in the CESM due to hosing. Changes were calculated by subtracting the average

over last 80 years of the four hosed ensemble members from the average of 320 years of

the preindustrial control run.

Figure S5. Tropical Pacific precipitation (colors) and zonal wind speed (unfilled

contours; m/s) in observations (top row), the CESM preindustrial control run (middle

row) and the hosed runs (bottom row) averaged over: A) the annual cycle; B) boreal fall

(Aug-Oct); and C) boreal spring (Feb-Apr). The observed climatological precipitation is

from GPCP Version 2.2 Combined Precipitation Data Set (averaged over 1979-2010) and

the observed surface winds are from ERA40 Reanalysis data. The control climatology is

the average of the 320-yr control PI CESM integration. The hosing climatology is the

ensemble average of the four 80-year hosing simulations.

Figure S6. Time series of 3-month running mean Niño 3 SSTAs in: A) 320 years of the

CESM preindustrial control run; B) the last 80 years of the four CESM hosed ensemble

members; C) 320 years of the LOAM with mean states from the preindustrial control run;

D) 320 years of the LOAM with mean states from one of the CESM hosed run.

Figure S7. EOF1 of tropical Pacific SSTAs in: A) 80 years of the preindustrial control

run of CESM; B) the last 80 years of one of the CESM hosed ensemble members; C) 80

years of the LOAM with mean states from the CESM preindustrial control run; D) 80

years of the LOAM with mean states from one of the CESM hosed ensemble members.

The fraction of total variance captured by each pattern is indicated in bold. The unfilled

contours in panels B and D are EOF1 from the respective control run. The LOAM SST

output was interpolated to the CESM ocean grid and smoothed with a 1–2–1 filter to

reduce the noise, as in Zebiak and Cane 2, Thompson 3 and Thompson 4.

Figure S8. Tropical Pacific SST and surface wind biases in the CESM preindustrial

control run. A) Mean annual SST and surface winds in observations; B) SST and surface

wind difference in 100 years of the CESM preindustrial control run relative to

observations. Observed SST data is from NOAA ERSST v3b (1971-2010) and observed

surface winds are from ERA40 Reanalysis data.

Figure S9. Ocean efficiency factors δT (left) and δW (right) in LOAM.

References

1 Chen, S., Hoffmann, S. S., Lund, D. C., Cobb, K. M. & Emile-Geay, J. A high-resolution speleothem record of western equatorial Pacific rainfall: Implications for Holocene ENSO evolution. Earth Planet. Sci. Lett. 442, 61-71 (2016).

2 Zebiak, S. E. & Cane, M. A. A model El Nino-Southern Oscillation. Monthly Weather Review 115, 2262-2278, doi:10.1175/1520-0493(1987)115<2262:ameno>2.0.co;2 (1987).

3 Thompson, C. J. Initial conditions for optimal growth in a coupled ocean-atmosphere model of ENSO. J. Atmos. Sci. 55, 537-557, doi:10.1175/1520-0469(1998)055<0537:icfogi>2.0.co;2 (1998).

4 Thompson, C. J. A linear, stochastic, dynamical model of El Niño /Southern Oscillation PhD thesis, University of Washington, (1998).

Response of the tropical Pacific to abrupt climate change 8,200 years ago 1

Alyssa R. Atwood1 David S. Battisti2, C. M. Bitz2, Julian P. Sachs1 2

3

Supplementary Information 4

5

1. Inferring Climatological Rainfall Changes and El Niño Rainfall Changes from El Junco 6

Lake Biomarker Records 7

From the available evidence, we infer that sedimentary dinosterol δ2H (Fig. 2A) reflects the 8

long-term mean (climatological) lake δ2H and thus changes in the climatology of mean annual 9

rainfall owing to the “amount effect”. From the C34 botryococcene data, we infer that 10

the sedimentary accumulation rate of C34 botryococcene (Fig. 2B) scales with the 11

climatological rainfall associated with El Niño events (i.e. the total rain that fell during El Niño 12

events divided by the total amount of time in the sedimentary interval). This inference is 13

equivalent to assuming that, during a given interval, the total accumulation of botryococcenes in 14

the sediment scales with the total rain that fell during El Niño events. Under this assumption, 15

changes in the botryococcene accumulation rate reflect changes in climatological rainfall due to 16

changes in El Niño rainfall. Changes in the climatological rainfall associated with El Niño events 17

(and therefore botryococcene accumulation rate) should be sensitive to changes in the frequency, 18

amplitude, and/or duration of El Niño rainfall events. The botryococcene accumulation rate thus 19

provides a broad (statistical) measure of the variability of El Niño rainfall events. Due to the 20

temporal resolution of the biomarker data, all rainfall reconstructions represent averages over 21

decadal and longer timescales. 22

Based on these inferences, we propose that rainfall at El Junco Lake can be reconstructed in 23

the following way: 24

∆ = − ∙ ∆( ) (2.1) 25

∆ = ∙ ∆( ( )) (2.2) 26

where ∆ is the total change in climatological rainfall, ∆ is the change in climatological 27

rainfall attributable to El Niño events (i.e. the change in El Niño mean annual rainfall), and 28

is the sedimentary botryococcene accumulation rate (in μg cm-2yr-1). The botryococcene 29

accumulation rate is related to the botryococcene concentration in the following way: 30

= ( ∗ ∗ )/ (2.3) 31

where [botryo] is the botryococcene concentration (in μg/g), DBD is the dry bulk density of the 32

sediment (in g/cm3), d is the thickness of the sediment sample (in cm), and t is the duration of 33

time represented by the sediment sample (in yr). In this study, an average DBD has been 34

estimated from the top 15 cm of sediment. Future work will include incorporating improved 35

estimates of DBD based on down-core gamma ray density measurements. 36

The El Junco Lake biomarker records are shown in Fig. 2. Positive (negative) values of ∆ 37

indicate an increase (decrease) in climatological rainfall and positive (negative) values of ∆ 38

indicate an increase (decrease) in climatological rainfall due to an increase (decrease) in El Niño 39

rainfall. It is important to note the distinction between El Niño rainfall reconstructions (as 40

presented here) and El Niño SSTA reconstructions, since the frequency and amplitude of rainfall 41

anomalies in the eastern Pacific during El Niño events can change without a corresponding 42

change in the amplitude of SSTAs. For instance, 1 demonstrated, using future climate simulations 43

from the Coupled Model Intercomparison Project phase 3 and phase 5, that extreme rainfall in 44

the Niño 3 region increases in response to a weakened mean zonal SST gradient in the tropical 45

Pacific due to enhanced probability of anomalous deep convection in the eastern Pacific under a 46

given SSTA. Possible mechanisms of increased (decreased) El Niño rainfall as recorded by the 47

botryococcene records therefore include: (1) decreased (increased) mean zonal SST gradient in 48

the tropical Pacific, (2) increased (decreased) variance of ENSO SSTAs, and/or (3) a shift in the 49

distribution of ENSO SSTAs to more eastern Pacific (central Pacific) El Niño events. ∆ should 50

be more sensitive to changes in Eastern Pacific (EP) El Niño events as compared to Central 51

Pacific (CP) events due to the location of El Junco Lake in the eastern equatorial Pacific 2,3. 52

This interpretation of the El Junco Lake biomarker records matches that presented in 53

Atwood 4 and is similar, though not identical, to that presented in Atwood and Sachs (2014) and 54

Zhang et al. (2014), based on the addition of new data and recent advances in our understanding 55

of the proxy records and their climatic implications. An overview of the important updates to the 56

biomarker data and their interpretations is provided in Section 3, below. 57

2. Inferring changes in residual rainfall from El Junco Lake biomarker records 58

Due to the sensitivity of rainfall at El Junco Lake to the climatological position of the eastern 59

Pacific ITCZ, we propose that further climatic information can be gained by comparing the 60

changes in inferred total climatological rainfall to the changes in climatological rainfall 61

associated with El Niño events. We first assume that changes in climatological rainfall (∆ ) at 62

El Junco Lake is comprised of two contributions: (1) changes in climatological rainfall 63

associated with changes in El Niño rainfall (∆ ) and (2) changes in (residual) climatological 64

rainfall due to processes unassociated with El Niño (∆ ; i.e. changes in the climatological 65

position of the ITCZ, changes in the Hadley circulation, or changes in localized convective 66

processes). This is represented in the following equation: 67

∆ = ∆ +∆ . (2.4) 68

We therefore propose that changes in residual rainfall (i.e. changes in climatological rainfall 69

unassociated with changes in El Niño rainfall) can be inferred in the following way. From Eq. 70

(2.4), we have: 71

∆ = ∆ − ∆ . 72

Substituting Eq. (2.2) and (2.3) into (2.4), we therefore obtain 73

∆ = − ∙ ∆( ) − ∙ ∆( ( )) . (2.5) 74

Since the proportionality constants β and γ are unknown (although both are greater than 75

zero), ∆ cannot be fully reconstructed. However, the sign of ∆ is constrained for periods in 76

which −∆(dinosterol δD) and ∆( ( )) are of opposite sign – that is, for periods in which 77

the inferred change in total climatological rainfall opposes the change in climatological rainfall 78

associated with El Niño events. We thus limit our interpretation of ∆ to its sign during such 79

periods. Positive (negative) values of ∆ indicate an increase (decrease) in climatological 80

rainfall due to processes unassociated with El Niño. 81

The primary utility of the ∆ reconstructions is to help interpret the source of the most 82

prominent mean rainfall changes at El Junco Lake. E.g. a positive value of ∆ indicates that the 83

increase in ∆ was due an increase in ∆ that was greater than the decrease in ∆ (i.e. that 84

the source of the increased climatological rainfall was from non-El Niño contributions). For 85

instance, a positive ∆ would result if the increase in climatological rainfall associated with a 86

southward shifted ITCZ in the eastern Pacific was greater than a coincident decrease in El Niño 87

contributions to climatological rainfall. In contrast, a negative value of ∆ indicates that the 88

decrease in ∆ was due a decrease in ∆ that was greater than the increase in ∆ (i.e. that the 89

source of the decreased climatological rainfall was from non-El Niño contributions, such as a 90

northward shift of the ITCZ). The El Junco Lake rainfall reconstructions are shown in Fig. 2 for 91

the last 9,100 years. 92

By design, the periods of inferred ∆ highlight periods in which ∆ was opposed by ∆ 93

(or, equivalently, ∆ opposed ∆ ). Such opposing changes in ∆ and ∆ appear to account 94

for many of the largest hydrologic changes in El Junco Lake over the last 9,100 years (Fig. 2). 95

3. Revisions to Previous Biomarker-Based El Junco Lake Rainfall Reconstructions 96

Interpretation of the biomarker data presented here is consistent with Atwood 4 and follows that 97

presented in Atwood and Sachs 5 and Zhang, et al. 6 with the following exceptions: 98

a. Zhang, et al. 6 propose that changes in botryococcene concentration scale with changes in 99

the frequency of El Niño events. In this study, we adopt what we believe to be a more 100

conservative interpretation of the data. Here, as in Atwood (2015), we infer that the 101

sedimentary botryococcene accumulation rate is proportional to climatological rainfall 102

associated with El Niño events. In adopting this interpretation, we allow for the 103

possibility that the sedimentary botryococcene accumulation rate is sensitive to changes 104

in the frequency, amplitude, and duration of El Niño rainfall events (as determined 105

through their combined influence on climatological rainfall associated with El Niño 106

events). 107

b. Atwood 4 uses the botryococcene δ2H to infer the mean isotopic composition of El Junco 108

Lake water during El Niño events and assumes that, to first order, changes in lake water 109

δ2H during El Niño events are driven by changes in the mean monthly rainfall during 110

those events. Changes in botryococcene δ2H thus provide a measure of changes in the 111

average amplitude of El Niño rainfall events. This interpretation is an update from that 112

presented in Atwood and Sachs 5, in which botryococcene δ2H was interpreted in terms 113

of a change in “El Niño activity”, defined as a change in the average frequency and/or 114

intensity of El Niño events. Atwood and Sachs 5 further used the botryococcene 115

concentration data to fill in missing values of botryococcene δ2H during periods in which 116

the concentration was too low for isotopic analysis (justified by the correlation between 117

the down-core botryococcene δ2H and concentration values). 118

c. Another distinction between the results presented in Atwood and Sachs 5 and this work is 119

the interpretation of changes in residual climatological rainfall (i.e. climatological rainfall 120

that is unassociated with El Niño). Here, as in Atwood (2015), we reconstruct changes in 121

residual climatological rainfall (∆ ) based on differences in the dinosterol δ2H and 122

botryococcene accumulation rate data (versus the dinosterol δ2H and botryococcene δ2H 123

data as in Atwood and Sachs 5. However, due to the correlation between the down-core 124

botryococcene δ2H and concentration, our conclusions concerning changes in residual 125

climatological rainfall have not been substantially modified. In addition, in an effort to 126

adopt a more conservative approach to the reconstruction of ∆ , in this study we define 127

only the sign of ∆ (as relative changes in its magnitude are dependent on the 128

proportionality constants β and γ in Eq. 2.5, which are not yet empirically constrained). 129

4. Sensitivity tests in LOAM 130

Sensitivity tests were performed in LOAM to determine which components of the tropical 131

Pacific background state were responsible for the decreased growth rate of the ENSO mode 132

under hosing. In these experiments, individual components of the mean fields from the hosing 133

run were prescribed in LOAM with all other mean fields provided from the control run. Table 1 134

shows the configuration and results of these experiments. 135

5. Mean state biases in CESM 136

CESM with CAM4 is recognized as a top-performing GCM with regard to its simulation of 137

tropical Pacific climate variability (Deser et al., 2012; Bellenger et al., 2014). However, tropical 138

Pacific mean state biases are still present in CESM, including a cold SST bias in the NE tropical 139

Pacific that extends along the equator, a warm SST bias along the coast of South America and 140

the northern subtropical Pacific, and equatorial easterlies that are too strong across the central 141

Pacific and in the far eastern equatorial Pacific (Fig. S8). 142

143

Supplementary Tables 144

Table S.1. Summary of the constants and tuning parameters used in LOAM (the governing 145

equations in LOAM are provided in the Supplementary material of Atwood et al., 2016). 146

Symbol Value Parameter

A Rayleigh damping constant 1.57607 ∙ 10

β Meridional gradient of Coriolis force 2.2915 ∙ 10 Coupling coefficient 1.4

, b Reference temperatures 303K, 5400 K

Thermal damping coefficient 125 days

Mean upper layer depth 150 m

Surface layer depth 50 m

Subsurface layer depth 100 m ( ) Ocean efficiency factor See Fig. S1 ( ) Ocean efficiency factor See Fig. S1 ′ Reduced gravity 5.6 cm

Ocean mechanical damping 8.7 months

Surface layer damping 2 days

, Density of the atmosphere, ocean 1.275 , 1.026 ∙ 10 ,

Gravity wave speed in atmosphere, ocean 60 , 2.9

Atmospheric drag coefficient 1.5 × 10

WBR Western boundary reflection 0.7

Diffusion Applied to meridional advection term

in ocean thermodynamic equation -0.014

147

References 148 149 1 Cai, W. J. et al. Increasing frequency of extreme El Nino events due to greenhouse 150

warming. Nature Climate Change 4, 111-116, doi:10.1038/nclimate2100 (2014). 151 2 Ashok, K., Behera, S. K., Rao, S. A., Weng, H. & Yamagata, T. El Niño Modoki and its 152

possible teleconnection. Journal of Geophysical Research 112, C11007 (2007). 153 3 Murphy, B. F., Ye, H. & Delage, F. Impacts of variations in the strength and structure of 154

El Nino events on Pacific rainfall in CMIP5 models. Clim. Dyn. 44, 3171-3186, 155 doi:10.1007/s00382-014-2389-9 (2015). 156

4 Atwood, A. R. Mechanisms of Tropical Pacific Climate Change During the Holocene.” 157 University of Washington, (2015). 158

5 Atwood, A. R. & Sachs, J. P. Separating ITCZ- and ENSO-related rainfall changes in the 159 Galápagos over the last 3 kyr using D/H ratios of multiple lipid biomarkers. Earth Planet. 160 Sci. Lett. 404, 408-419 (2014). 161

6 Zhang, Z. H., Leduc, G. & Sachs, J. P. El Nino evolution during the Holocene revealed 162 by a biomarker rain gauge in the Galapagos Islands. Earth Planet. Sci. Lett. 404, 420-163 434, doi:10.1016/j.epsl.2014.07.013 (2014). 164

165