Interpreting Past Hydroclimate Variability from Sedimentary Records: Challenges & Considerations

Interpreting past hydroclimate variability from sedimentary records: Challenges & considerations

Jessica Tierney The University of Arizona

a) Archive types

80oN

60oN borehole 40oN coral

20oN documents glacier ice o 0 hybrid

20oS lake sediment marine sediment o 40 S sclerosponge

60oS speleothem tree 80oS

b) Resolution

80oN 50 60oN 20 40oN 10 o 20 N 5

0o 2

20oS

1 Resolution (years) 40oS The farther back in time you go, the 1/2 60oS more important sediments become. 1/6 80oS 1/12 Marine Sediments Lake Sediments Tree Rings Corals c) Temporal Availability First Millennium 600 150 500 100 400 50 0 300

200 # proxies

100

0 200 400 600 800 1000 1200 1400 1600 1800 2000 Year (CE) PAGES2K, 2016, Scientiﬁc Data, in prep

Figure 1: Spatiotemporal data availability in the PAGES2k database. (a) Geo- graphical distribution, by archive type, coded by color and shape; (b) Temporal resolution in the PAGES2k database, deﬁned here as the median of the spacing between consecutive observation. Shapes as in (a), colors encode the resolution in years (see colorbar); (c) Temporal availability, coded by color as in (a). Some nice features of sediment archives

• Long, continuous records • Multiple sensors and proxies can be measured Sedimentary Observations (Proxies) with hydroclimate information

• Oxygen isotopes (δ18O) on foraminifera (marine) • Lake levels • Things sensitive to lake levels (Mg/Ca of minerals, δ18O of authigenic carbonate). • δD of leaf waxes • Runoff indicators (Ti concentration, varves) • Microfossil assemblages (pollen, diatoms) Some challenging features of sediment archives

• Age uncertainty is usually large • Bioturbation mixes signals • The archive itself adds red noise (e.g. lakes). Interpreting sedimentary archives: an overview

5 200 (a) Climate (b) Smoothed Climate 180 Data on true age model Data on age model #1 160 Data on age model #2 140 0 120

100

60 −5 Proxy values (offset) True age model

Time − years before present 40 Age model #1 Age model #2 20 Measured Ages Proxy Observations 0 −10 0 100 200 0 20 40 60 80 100 120 140 160 180 200 Depth Time − Years before present

Figure made by Martin Tingley Proper interpretation of hydroclimate data from sediments requires:

• Proxy (observation) forward model(s) • An archive model to account for smoothing/reddening • A way to deal with age uncertainty Archive issues: Age Uncertainty The time uncertainty continuum Cross-dated Archives (Tree rings)

Layer-counted archives (varves, ice cores, corals)

Radiometrically dated - Gaussian (U/Th on speleothems,210Pb in sediments)

Radiometric dating - non-Gaussian (14C on sediments)

100 75 50 25 0 Uncertainty in Years R EPORTS Fig. 1. Location and by the planktic foraminifer Globigerina bul- correlation map show- loides reflect the cooler, more nutrient-rich ing the spatial relation waters found along the coast and that season- between variations in al assemblage changes mirror the changes in the abundance of the planktic foraminifer G. upwelling intensity, and thus trade winds, bulloides in the Cariaco over the year (6). Basin (southern Carib- We sampled sediments from box core bean) and SST anoma- PL07-71 BC (395 m water depth), recovered lies in the Atlantic Ocean. from the gentle northern slope of the eastern Mapped correlation co- Cariaco Basin, at continuous 1-mm intervals efﬁcients (r)indicatethe correlationsbetweenbox- (7). The age model for core PL07–71 BC car decadal-averaged G. (Fig. 2) consists of two independently de- bulloides data and histor- rived, mutually supporting age data sets. Age ical SST anomaly data control for the upper 12.6 cm of the core (12)forthetimeperiod (1880 to 1990 A.D.) is based on robust faunal 1856–1990 A.D. Corre- correlations with a nearby box core already lations between G. bul- 210 loides abundance and possessing a well-constrained Pb and varve SST anomaly time series chronology (8). Below a depth of 12.6 cm, are highest (–0.8 Ͻ r Ͻ sediment laminations are less distinct and are Ϫ0.9, signiﬁcant at the difficult to count reliably. For this reason, 99.9% level) in the North below 12.6 cm (1165 to 1879 A.D.) the age Atlantic region between model is based on a series of accelerator mass 50o and 60oN, indicating that G. bulloides abun- spectrometry (AMS) dates from monospe- dance, and hence trade cific samples of G. bulloides. Radiocarbon- wind–induced upwelling based sedimentation rates for samples imme- on June 1, 2016 over the Cariaco Basin, diately at and below 12.7 cm are the same as are tightly coupled to the 210Pb/varve-based estimates above 12.7 SST anomalies in the cm. The resulting 825-year record has a sam- northern North Atlantic region. ple resolution of about one sample per year near the top of the record, decreasing to one sample per 2.5 years at the base. The recent (1880 to 1990 A.D.) abundance record for G. bulloides illustrates sub-

Fig. 2. The age model for Cariaco Basin box core PL07-71 BC is based on a 210 14 combination of Pb, varve, and AMS Cchronologies.Fortheupper 12.6 cm http://science.sciencemag.org/ of PL07-71 BC, the age model is based on detailed correlations to PL07-81 BC (A), a nearby box core that is already well dated by 210Pb dating and varve counts (8), through an intermediate box core PL07-64 BC (B); dpm, disintegra- tions per minute. Cores PL07-81 BC and PL07-64 BC lie deeper in the basin and contain identical interbedded turbidites. Varve counts and 210Pb values both indicate that the turbidites are not erosive and that a continuous sediment sequence is present in both cores. Sediments from PL07-64 BC (B) above and

between the turbidites were sampled at continuous 1-mm intervals and Downloaded from analyzed for their planktic foraminiferal content. These data were used to develop faunal correlations between PL07-64 BC (B) and PL07-71 BC (C)and to transfer the established chronology from PL07-81 BC (A) to the latter. An example of this type of faunal correlation is shown for the foraminifer Orbulina universa (data shown as a three-point smooth); multiple taxa were used to check and verify the correlations. For sediment in core PL07-71 BC deposited at a depth between 12.6 and 56.4 cm (D), an interval where laminae were too faint to count, the age model is based on a series of 12 AMS 14C Annually resolved ≠ time certain dates measured on monospeciﬁc samples of G. bulloides (Table 1). Radiocar- bon dates were converted to the calendar ages shown in (D) using a calibration model for marine samples (35); a local reservoir correction of 420 years, shown to be constant over periods of large climatic change, was used to account for radiocarbon Cariaco Basin Age Model differences between the surface waters and the atmosphere (4). Error bars represent the 1␴ deviation; the unevenness of the error bars is a function of the marine calibration curve. A third-order regression line passing through both sets of age control points yields an r 2 of 0.98 and was used to assign calendar year ages to all samples in the box core deeper than 12.6 cm.

Black et al., 1999, Science

1710 26 NOVEMBER 1999 VOL 286 SCIENCE www.sciencemag.org LAKE CHALLA, KENYA (VARVE-COUNTED) Wolff et al., 2011, Science; Verschuren et al., 2009, Nature, Tierney et al., 2011 QSR 4 published age model 68% uncertainty 3 90% uncertainty ~+/- 1-2 years 2 ~+/- 25-30 years

PROXY PC1 0 − 3

−1

−2

−3 1100 1200 1300 1400 1500 1600 1700 1800 1900 YEAR LAKE NAIVASHA, KENYA (14C DATED) Verschuren et al., 2000, Nature

35 dates 68% uncertainty 90% uncertainty 30 published age model

15 Lake Level (m)

5 1200 1400 1600 1800 Year AD OxCal v4.1.7 Bronk Ramsey (2010); r:5 Marine data from Reimer et al (2009);

P_Sequence P17815P 1000 Age Modeling Techniques

Boundary bottom Simplest approach uses MC R_Date 900.5 OxCal R_Date 840.5 iteration and the constraint of

800 Bronk Ramsey, 1995 R_Date 780.5 Bronk Ramsey,M. Blaauw2008 andsuperposition. J. A. Christen More complex 467

R_Date 700.5 approaches make assumptions R_Date 680.5 R_Date 640.5 about the sedimentation process 600 R_Date 600.5

R_Date 560.5

R_Date 520.5

Position R_Date 460.5 R_Date 420.5 BACON 400 Blaauw & Christen, 2011 R_Date 340.5

R_Date 260.5

R_Date 220.5 cal yr BP 200 R_Date 200.5

R_Date 150.5

R_Date 90.5

R_Date 41.5 0 2000 4000 6000 8000 0 Boundary top

0 50 100 150 200

60000 50000 40000 30000 20000 10000 0 Depth Modelled date (BP)

Figure 3: Posterior age–depth model of core RLGH3 (grey), overlaying the calibrated distributions of the individual dates (blue). Grey dots indicate the model’s 95% probability intervals.

history, with some ﬂuctuations in accumulation rate at sub-millennial scale (Figure 3). The posteriors for accumulation rate and its variability are comparable to their priors, although the posterior indicates more memory than assumed apriori. As in the previous example of the peat core, the prior used for accumulation rates is su±ciently strong and data cannot provide further information. Little more can be learned from radiocarbon data than what we already know about sediment accumulation in this lake core. What constitutes a reasonable (non-decreasing) age-depth model between 0 and 50 cm, for example? This information is not contained within the radiocarbon data and must be an external input, indeed formally provided by our prior for accumulation rates. There is a clear disagreement between the dates c. 130 to 40 cm depth, which can be attributed to erosion of older organic material from the catchment (Jones et al. 1989). Our model seems unbothered by the discording dates and eÆectively bypasses them. As detailed in Section 1.1, our method uses a Student-t model to calibrate the dates, instead of the usual normal model. The wider tails of our calibration model reduce the need for detecting and removing outliers. For comparison purposes, we ran the RLGH3 data through BChron (Parnell et al. Monte Carlo EOF East Africa MCEOF1; 10,000 simulations

4 MEDIAN 68% UNCERTAINTY EOF1 Loadings 95% UNCERTAINTY 3 sizes of circles represents loading value colors represent median and uncertainty bounds 2 4oN

0 0o AMPLITUDE + Wet −1

−2 4oS −3 32% ± 6% (2σ) Explained Variance

1300 1400 1500 1600 1700 1800 1900 - Dry LOWER 1σ YEAR MEDIAN 8oS UPPER 1σ 1 10oN 0.8 Tierney, Smerdon, Anchukaitis & Seager 12oS Nature, 2013 0.6 28oE 32oE 36oE 40oE 44oE 5oN 0.4

EDWARD 0.2 VICTORIA 0o NAIVASHA 0

−0.2 CHALLA o 5 S TANGANYIKA −0.4 LOWER 1σ MASOKO MEDIAN −0.6 UPPER 1σ 10oS −0.8 MALAWI

o o o o −1 30 E 35 E 40 E 45 E Archive issues: Bioturbation You need to account Unless you have this: for bioturbation!

Photo by Guillaume Leduc RESEARCH | REPORTS

events in these data is possible only for very strong specific events is beyond the method’scapability mean have a >87% probability of occurring dur- negative (El Niño) and positive (La Niña) outliers. (19). For more modest anomalies, however, we can ing El Niño and a 100% probability of occurring For example, our data identified strong negative only assign probabilities that they represent ENSO during La Niña (Fig. 2A) (19). The frequency of outliers suggestive of the historic El Niños of 1997– events. To this end, we used historical ENSO events these high-probability ENSO outliers shows clear 18 1998, 1982–1983, and 1876–1877 (Fig. 2A), although and monthly reanalysis data to calculate that d Oc modulation through time. For example, warm one-to-one correspondence of any of our data with values exceeding a ±1‰ departure from the and cold outliers from the period 1500–1850 CE occurred 2.3 times more frequently than those from the period 1150–1500 CE (normalized to sample size). Bootstrap analysis with varying sample size confirms significant differences in outlier frequency between these two periods (fig. S7). Such high-probability ENSO outliers are the largest 18 Example RESEARCH | REPORTS driver of the integrated d OcV signal (Fig. 2B), justifying its interpretation as an ENSO index. 18 in the North Atlantic may represent a global trend. PALEOCEANOGRAPHY The pattern of d OcVduringthemillennium Contrary to the generalized assumption of negative (Fig. 2B) shows marked similarity with Mg/Ca effects of ocean acidification on calcifiers, coccoli- SSTs (correlation coefficient R = –0.43, P <0.05) thophores may be capable of adapting to a high-CO2 world (24), especially given evidence of highly cal- Dynamical excitation of the tropical (Fig. 2C), considering that these data sets are en- cified coccolithophores in areas with seasonally tirely independent and based on different geo- high pCO2 or low pH (7, 8). Coccolithophores show Pacific Ocean and ENSO variability by outstanding competitive abilities under the strati- chemical proxies and separate foraminifera samples. fied, warm, nutrient-depleted conditions projected Little Ice Age cooling Aconspicuousfeatureofbothisamid-millennium for the future ocean (6). Nevertheless, with increas- (~1500 CE) change in the proxy character. When ing pCO ,wemightexpectchangesincommunity 2 Gerald T. Rustic,1,2*† Athanasios Koutavas,1,2,3 18 composition and a decrease in calcification, leading d OcViscomparedwiththereconstructedSST Thomas M. Marchitto,4 Braddock K. Linsley3 to changes in rain ratio, export efficiency, and gradient (Fig. 3, F and G), an inverse correlation is trophic effects higher in the marine food web. Tropical Pacific Ocean dynamics during the Medieval Climate Anomaly (MCA) and the Little evident. Taking into account our age-model un- Ice Age (LIA) are poorly characterized due to a lack of evidence from the eastern equatorial RESEARCHREFERENCES| REPORTS AND NOTES certainty, Monte Carlo analysis shows an average 1. J. C. Orr et al., Nature 437,681–686 (2005). RusticPacific. We reconstructedet al., 2015 sea surface temperature, El Niño–Southern Oscillation (ENSO) 2. B. Rost, U. Riebesell, S. Burkhardt, D. Sultemeyer, activity, and the tropical Pacific zonal gradient for the past millennium from Galápagos correlation of R = –0.55 with P <0.05in>95%of Limnol. Oceanogr. 48,55–67 (2003). ocean sediments. We document a mid-millennium shift (MMS) in ocean-atmosphere realizations (fig. S8). This result implies a funda- ENSO events are3. consistentlyR. Francois, S. Honjo, captured R. Krishfield, by S. isotopic Manganini, Global Biogeochem. Cycles 16,34-1–34-20 (2001). circulation around 1500–1650 CE, from a state with dampened ENSO and strong zonal mental association between a weaker zonal gra- outliers (19). 4. R. A. Armstrong, C. Lee, J. I. Hedges, S. Honjo, S. G. Wakeham, gradient to one with amplified ENSO and weak gradient. The MMS coincided with the Our Mg/Ca recordDeep Sea reveals Res. Part systematic II Top. Stud. Oceanogr. changes49,219–236 deepest LIA cooling and was probably caused by a southward shift of the intertropical dient and strong ENSO variability, and vice versa. ineastern Pacific(2002). SST of ~1°C (Figs. 2C and 3D). convergence zone. The peak of the MCA (900 1150 CE) was a warm period in the eastern 5. P. Matrai, M. Vernet, P. Wassmann, J. Mar. Syst. 67,83–101 (2007). – The LIA (~1500–1850 CE) stands out as a period Pacific, contradicting the paradigm of a persistent La Niña pattern. 18 Warm SSTs characterize6. B. Rost, U. Riebesell, thepeak Coccolithophores of the and MCA the biological with high d OcVand,byinference,greaterENSO (~900–1150 CE),pump: coeval Responses with toNH environmental warmth changes, (Fig. in Coccolithophores: From Molecular Processes to Global Impact, activity. The LIA also shows unusual general vola- 3C). The timingH. of R. Thierstein,this warm J. R. Young, period Eds. (Springer, is firmly 2004), pp. 99–125. he tropical Pacific Ocean exerts a major in- El Niño–like state during the LIA (6). This hy- 14 tility with large century-scale swings in mean EEP constrained by7. a L. calibrated Beaufort et al., NatureCageat1139CE476,80–83 (2011). fluence on global climate through the inter- pothesis has found some support in central Pacific 8. H. E. K. Smith et al., Proc. Natl. Acad. Sci. U.S.A. 109, 18 annual El Niño–Southern Oscillation (ENSO) corals (15), North American tree rings document- SST (Fig. 3D) and d OcV(Fig.3G).Incontrast,the (1108–1172 CE, 1s8845range)–8849 (Fig. (2012). 2). This finding con- trasts with a previously9. M. D. Iglesias-Rodriguez hypothesizedet al., Science dynamical320,336–340 (2008). (1)andPacificDecadalOscillation(PDO)(2). ing medieval megadroughts (16), and multiproxy preceding period (1150–1500 CE) had the lowest La Niña state for10. U.the Riebesell, MCA A. Körtzinger,(6). To A.further Oschlies, Proc. test Natl. this Acad. Sci. U.S.A. However, its role on centennial to millennial climate field reconstructions (6). In each of these 106,20602–20609 (2009). T ENSO activity of the millennium and was less time scales remains unclear. Recent evidence that cases, however, direct confirmation from the EEP

hypothesis, we11. used A. J. Richardson reconstructedet al., Prog. Oceanogr.WPWP68 SSTs,27–74 (2006). on March 12, 2016 a cool eastern equatorial Pacific (EEP) played a has been lacking. Insights into past ENSO var- volatile. Compared with the LIA, this state had (Fig. 3E) and our12. G. Galápagos C. Hays, A. J. Warner, record A. W. to G. John, compute D. S. Harbour, P. M. Holligan, J. Mar. Biol. Assoc. U. K. 75,503–506 (1995). key role in the global warming slowdown of the iability have also remained elusive and are mostly 18 20% less d Oc variance (F test: P <0.05;Brown- the zonal SST13. gradient. A. N. Antia et Duringal., Global Biogeochem. ~900– Cycles115015 CE,,845–862 (2001). past ~15 years (3, 4)emphasizestheneedtoun- based on discontinuous corals and remote tree- the zonal gradient14. A. wasWinter, similar J. Henderiks, to L. that Beaufort, of the R. E. 20th M. Rickaby, derstand tropical Pacific climate in the recent past. ring records (15 18). Here we address these data Forsythe test: P <0.05)anda1.0°to1.5°Cstrong- C. W. Brown, J. Plankton Res. 36,316–325 (2014). – century. This finding15. E. K. Hovland, does H. not M. Dierssen, support A. S. an Ferreira, anom- G. Johnsen, During the last millennium, Northern Hemisphere limitations with climate reconstructions from the er zonal gradient. Because the ENSO variance in alous La Niña–likeMar. state; Ecol. Prog. instead, Ser. 484 warm,17–32 (2013). SSTs were (NH) climate evolved from a warm Medieval heart of the EEP cold tongue. We present con- our data is superimposed on a large annual cycle, 16. T. J. Smyth, T. Tyrrell, B. Tarrant, Geophys. Res. Lett. 31, Climate Anomaly (MCA) (~900–1450 CE) into a tinuous, multidecadally resolved estimates of local generally presentL11302 during (2004). this period in both the this 20% change in d18O Vpotentiallysignifiesa east and west,17. in J. phase C. Cubillos withet al., NHMar. Ecol. warmth. Prog. Ser. 348,47–54 (2007). substantially colder Little Ice Age (LIA) (~1450– SST, basin-wide zonal SST gradient, and ENSO c 18. L. Breiman, Mach. Learn. 45,5–32 (2001). 1850 CE), followed by modern warming (5–7). In activity from Galápagos marine sediments over Downloaded from After peak MCA warmth, Galápagos SSTs cooled much greater change in ENSO variability. To 19. T. Takahashi et al., Deep Sea Res. Part I Oceanogr. Res. Pap. the tropics, the MCA-to-LIA transition exhibited the period ~1000–2009 CE. by ~1°C (±0.3°C, 156s)between~1150and~1500CE.,2075–2076 (2009). further quantify this, we modeled the effect of 20. M. Edwards, G. Beaugrand, P. Helaouët, J. Alheit, S. Coombs, aweakeningoftheEastAsiansummermonsoon Our primary data sets were developed from an 18 An anomalous strongPLOS ONE zonal8,e57212(2013). SST gradient became (8, 9), a southward shift of the Atlantic and Pacific ocean multicore (KNR195-5 MC42C) retrieved in ENSO on d OcV. By holding the seasonal cycle established during21. A. Gnanadesikan, this period J. P. (Fig.Dunne, R. 3F), Msadek, largelyJ. Mar. Syst. 133, intertropical convergence zones (ITCZs) (10, 11), 2009 near Española Island in the Galápagos constant to its modern value and varying the due to more rapid39– cooling54 (2014). in the east versus the and sea surface cooling in the western Pacific warm (01°15.18′S, 89°41.13′W, 615 m depth) (Fig. 1). Undis- 22. A. McIntyre, A. W. H. Be, Deep-Sea Res. 14,561(1967). amplitude of ENSO anomalies in the instrumental west. This enhanced23. B. Hannisdal, zonal J. gradient Henderiks, L. is H. evidence Liow, Glob. Change of Biol. 18, pool (WPWP) (12, 13). Notably absent, however, turbed recovery of the sediment-water interface dynamical changes3504 in–3516 Pacific (2012). circulation and sup- are contemporaneous records of sea surface tem- and detection of bomb radiocarbon confirm that era, we calculated that a 20% decrease in total 24. K. T. Lohbeck, U. Riebesell, T. B. H. Reusch, Nat. Geosci. 5, perature (SST) from the ENSO-sensitive and dyna- the core top is modern (19). Sedimentation rates ports the notion346 of–351 a cool,(2012). La Niña–like mean variance required a ~55% decrease in ENSO amp- mically important cold tongue of the EEP. Without averaged 13 cm/ky over the past millennium. state. However, the timing of this anomalous trop- litude. We conclude from these comparisons that ACKNOWLEDGMENTS such records, any dynamical connections between Detection of a modern 0.3–per mil (‰)carbonm/s 13 ical Pacific meanWe thank state D. Johns,(~1150 the– Sir1500 Alister CE, Hardy ±30 for Ocean years, Science (SAHFOS) basin-scale tropical Pacific processes and NH cli- isotope (d C) depletion, or Suess effect (fig. S9), the ~350-year periods preceding and following 1s)postdatedpeakNHMCAwarmthandwasAssociated External Researcher fund, and everyone involved in the CPR mate cannot be properly diagnosed. further corroborates the presence of a modern ~1500 CE represent fundamentally different dy- instead associatedsurvey. We with thank t theransitional Johns Hopkins University NH cooling and the Applied Physics It has been hypothesized that an “ocean dy- core top and core integrity and argues against Laboratory for financial support and B. Zaitchik, D. Waugh, J. Dunne, namical states. The basin-wide tropical Pacific into the LIA. Theand origin three anonymous of this reviewers dynamical for helpful anomaly comments. Funding for namical thermostat” (14)responseoftheEEPto substantial signal attenuation by bioturbation remains unclear.S.D.G. Volcanic was from NSF-1149460. and solar Support forcing for W.B. came(Fig. from NSF solar and volcanic forcing induced a cool, La Niña– (19). The core was slabbed continuously in sub- Fig. 2. Climatic proxies measured on Galápagos multicore KNR195-5 MC42C. (A)IndividualG. ruber appears to have toggled between these states dur- 3, A and B) indicate(Division of an Ocean increase Sciences) and in NASA volcanism (Ocean Biology in and like mean state during the MCA and a warmer, centimeter slices representing ~20- to 80-year d18O in 24 contiguous samples after removing the sample means [shown in (D)]. Red circles denote ing a mid-millennium shift (MMS) that unfolded Biogeochemistry Program). CPR data used in this study can be periods. We analyzed Mg/Ca ratios of the mixed- c the late 12th centuryobtained from followed SAHFOS upon by request. a decrease in sensu stricto (ss)andbluesquaressensulato(sl) morphotypes, which together capture the correct over the 16th and early 17th centuries. 1Department of Engineering Science and Physics, College of layer foraminifera Globigerinoides ruber (white) solar activity in the 13th century, but overall cor- 18 SUPPLEMENTARY MATERIALS Staten Island, City University of New York, Staten Island, NY as a proxy for mean SST. We also analyzed d Oc mean, variance, and skewness of historical SST data (19). Age control points are represented by black The MMS occurred as the NH descended into relations between these forcings and SSTs are 10314, USA. 2Doctoral Program in Earth and Environmental www.sciencemag.org/content/350/6267/1533/suppl/DC1 (calcite) of individual G. ruber shells (~60 indi- 14 Sciences, Graduate Center of the City University of New diamonds (top), with black whiskers denoting 1s age uncertainty for Cages.Themulticoretopis the coldest portions of the LIA, leading us to con- poor. In light ofMaterials decreasing and Methods external forcing, the vidual G. ruber analyses per sample horizon) to York, New York, NY 10016, USA. 3Lamont–Doherty Earth Supplementary Text 18 observed EEP cooling and enhanced zonal gra- Observatory of Columbia University, 61 Route 9W, Palisades, quantify population-level d Oc variance related assumed to be modern (2009). Horizontal dashed lines at ±1.0‰ from the mean denote high- sider a NH trigger through atmospheric mech- Figs. S1 to S7 NY 10964, USA. 4Department of Geological Sciences and to ENSO activity (20, 21). Benchmark tests com- 18 dient are not consistentTable S1 with an ocean dynamical probability El Niño and La Niña events outside this range. d Oc values consistent with historical strong anisms. The tropical atmosphere is dominated by Institute of Arctic and Alpine Research, University of thermostat mechanismReferences (25 (6–47, 14) ). paring modern samples with instrumental data 18 El Niño events at 1876–1877, 1982–1983, and 1997–1998 are indicated. VPDB, Vienna Pee Dee belemnite. the seasonally migrating ITCZ. Evidence linking Colorado, Boulder, CO 80309, USA. support the use of single-shell d O variance After ~150030 CE, January the 2015; EEP accepted cooling 11 November trend 2015 ended *Corresponding author. E-mail: [email protected] c 18 (B) G. ruber d18O variance (d18O V) for each of the 24 sample intervals. Vertical error bars show SE of the the ITCZ to hemispheric warming and cooling and local SSTsPublished began online to 26 increase November 2015 and become †Present address: Lamont–Doherty Earth Observatory of Columbia (d OcV) as a proxy of the monthly SST variance c c 10.1126/science.aaa8026 University, 61 Route 9W, Palisades, NY 10964, USA. of the real ocean and further show that large more volatile, signaling a shift in the dynamical variance. (C) G. ruber Mg/Ca SST. Each datum is the average of triplicate analyses. Vertical error bars cycles is widespread throughout the last glacial 18 character of the EEP (Fig. 3D). Beginning in the indicate SD of the replicates. (D)MeanG. ruber d Oc.VerticalbarsindicateSEM.For(B),(C),and(D), period (22)andimplicatessouthwardITCZdis- 16th century, anSCIENCE anomaloussciencemag.org weak zonal gradient 18 DECEMBER 2015 • VOL 350 ISSUE 6267 1537 the horizontal bars denote the duration of each sample interval and horizontal whiskers show the 1s age placements during NH cold periods, consistent was established as the WPWP continued to cool uncertainty. Modern values computed from SODA 2.1.6 are indicated by arrows on the right. with theory and models (22, 23). We propose but the EEP trend reversed from cooling to warming (Fig. 3, D to F). This reversal occurred as the NH descended into the coldest portion of the LIA SCIENCE sciencemag.org 18 DECEMBER 2015 • VOL 350 ISSUE 6267 1539 in the 16th and 17th centuries (Fig. 3C) and lasted into the mid-19th century. Throughout most of the LIA, the zonal gradient remained reduced, evoking an extended El Niño–like mean state (6). This state appears to have ended with renewed WPWP warming in the early 20th century. The record of d18O from single-shell G. ruber c m/s allows us to test for ENSO modulation in association with the above mean state shifts. Figure 18 2A shows d Oc measurements of individual shells (n =1325shells)from24samplesspanningthe entire past millennium. By virtue of the short Fig. 1. Average tropical Pacific SST and 1000-mbar winds over the period 1981–2008. (Top)Data foraminiferal life cycle (1 month or less), single- for March; (bottom)dataforSeptember.TheKNR195-5MC42corelocationisdenotedbyawhitestar.Mean 18 shell d Oc responds to seasonal and interannual SST values are from the Simple Ocean Data Assimilation (SODA) analysis, release 2.1.6 (1981–2008) (30); signals, the latter of which dominate the tails of 1000-mbar winds are from the National Oceanic and Atmospheric Administration (NOAA) National Centers for 18 the d Oc distribution through the presence or Environmental Prediction–National Center for Atmospheric Research Climate Data Assimilation System absence of outliers. Explicit identification of ENSO (1981–2009) (31).The position of the ITCZ is indicated by the gray dashed lines where wind vectors converge.

1538 18 DECEMBER 2015 • VOL 350 ISSUE 6267 sciencemag.org SCIENCE The Problem:

RecordFig. is S1: only ca. 15 cm long, and is not laminated.

Rustic et al., 2015

Figure S1. MC42C age model with 1-sigma and 2-sigma age uncertainty generated by

Bchron v. 4.1.2 (36). Age control points are 14C ages (Table S1), except the mutlicore top

which is assigned the year of collection (2009) with zero uncertainty.

14 Using the Suess effect to assess bioturbation 1.8 −6.4

1.6 −6.6 MC42 DATA TURBO2 model 1.4 −6.8 5.5 cm mixing depth C C foram foram

13 1.2

13 1.2 C C δ δ −7

−7 13 13 δ δ

1 −7.2

Suess effect signal (Rubino et al., 2013) 0.8 −7.4 ModeledSuess effect bioturbated signal (Rubino signal inet MC42Cal., 2013) Actual core data 0.6 1100 1200 1300 1400 1500 1600 1700 1800 1900 2000 Year AD 104 Bioturbation models Bard et al.: Reconstruction of the last deglaciation be interpolated,Typically without modeled losing sight as ofan the impulse fact that response the ""ih ...... - ...... V/////////, I an paleoclimatologicalfunction significance that describes of interpolated instantaneous points is nil. mixing of initial deposition in the b. When the deconvolved abundance of the carrier E k tends rapidly“bioturbation towards zero, the layer” relative (H).error on the restored (5 signal becomes greater and the value of deconvolved isotopes IEk maye.g., be erroneous. Berger and Heath (1968) and Bard et 3 ....."1 ...... c. The signalsal., used (1987) are functions model: of depth and it is assumed that proportionality exists between this variant and time. This hypothesis remains valid during periods of sedimentation rate stability,dC/dt but = becomes1/H * (C falsedh-C Hwhen) this parameter varies. In order or,to illustratethe diffusion the signal model attenuation of Guinasso effect, theand system is fed with a sinusoidal signal e (t) whose amplitude and period areSchink, e ~ and 1975, L, respectively. JGR: In this case, the response s (t) will also be sinusoidal with the same period 2 2 L, an amplitudedC/dt s ~ =and D ad phaseC/dx shift - v*dc/dx Ywith respect to the o, .... input e (0: Fig. 1. Parameters usedBard for theet al.,theoretical 1987, Clim. calculations Dyn. (see text) e (t)=e where~ X sin D[(27r/L) is diffusivity X t] and v is sed rate. s(t)=s ~ X sin[(21r/L) X (t+Y)l

Following a Laplace transform of the transfer function L signal period in temporal coordinates, we obtain M the gain of the system. (The greater the attenua- 1 tion, the less the gain of the system.) G (p') = exp ((H/R) x p') x 1 + (H/R) x p" This equation clearly emphasizes the output attenuation effect for high frequencies. In order to illustrate this biotur- H bioturbation thickness bation effect, a figure displaying the variations of M versus R sedimentation rate L has been drawn for different values of the parameters H where H/R represents the sedimentation time ofa bioturba- and R (cf. Fig. 2). It confirms that, in order to observe a tion layer. clear stratigraphic signal, it is necessary to study deep-sea After calculating the modulus and argument of the cores with high sedimentation rate R and/or low bioturba- transmittance, the following is obtained: tion thickness H. 1 s~176 I G[(2~-/L) xj]l =

l+(27rxH)2-RxL ~aln 2000 4000 6000 8000 51gn~l period years db I I I I I I I r I / ~

Y= (L/2 70 x Arg [G (2 7r/L) x j] =H/R- (L/2 70

X (27r xH~ -10 tan-I \ R->~-L )

with p' = (2 7r/L) x j wherej 2= - 1 / -20 The phase-shift equation illustrates clearty the "ad- vance" effect linked to the bioturbation process (Y>0). Subsequently, the equation which gives the gain in decibels will be -30

M=20 • logi0 (s ~ o) Fig. 2. Attenuation effect ofbioturbation for different values of sedimentation rate R and mixing thickness H. (1) H=5 cm and R=10 cm/kyr; (2) [ x"Y7 H=10 cm and R=15 cm/kyr; (3) H=10 cm and R=10 cnVkyr; (4) H=15 =-10xlogw I+\RX-~/] J cm and R=10 crrdkyr; (5) H=10 cm and R=l.5 cm/kyr Archive issues: Reddening of the signal (lake level proxies) Jinja Example: Lake Victoria

500 Rainfall, Jinja Station 1962 400

300

200 mm/month

100

0 13.5

13 Lake Levels, Jinja Station

12.5

11.5

Meters above gauge zero 10.5

10 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 Year Spectral comparison

2 10 PRECIPITATION LAKE LEVELS 1 10 90% CI

0 10

−1 10

−2 POWER 10

−3 10

−4 10

−5 10 −1 0 10 10 FREQUENCY (cycles/year) Ways forward

• Ad-hoc approach: only use/interpret lowest frequencies (drawbacks: qualitative only, plus archive creates low frequency, so is this variability even “real”?) • Better idea: start forward modeling sedimentary records from climate models. Finally, a simple model for UK37:

K′ logit(U37 )=α + β · SST + ϵ, (16) ϵ ∼ N (0, τ 2)IID. (17)

Avisionforasedimenthierarchicalmodel: Level 1: Spatiotemporal process model for the climate variable of interest (e.g., SST):

Envisioning a SSTt+1 − µ = α · (SSTt − µ)+ϵt (18) ϵ ∼ N (0, Σ)(19) Level 1: Spatiotemporal t hierarchical 2 process model Σi,j = σ exp(−φ|xi − xj|)(20) model for Level 2: Proxy forward models, e.g., for UK: sediments K′ logit(U37 )|SST = α + β · SST + ϵ, (21) 2 Level 2: Proxy forward model, can be ϵ ∼ N (0, τ )IID. (22) updated to accountor, tofor account age model for age uncertainties: uncertainty (Werner & Tingley, 2015, Clim. Past)

K′ T logit(U37 )|T ,SST = α + β · Λ · SST + ϵ, (23) ϵ ∼ N (0, τ 2)IID. (24)

Level 3: ArchiveLevel 3: model Archive (bioturbation model to account or for bioturbation: other sedimentary features) K′ K′ K′ U37obs|U37 = U37t1+tn · g(tn)+ϵ(t1), (25) !tn 2 ϵ(t1) ∼ N (0, τ )IID. (26)

Jessica Tierney [email protected] 3rd May, 2015

3 Shameless Plug: Awesome Postdoctoral position available in my lab!

• Part of the “Data Assimilation for Deep Time” Project. We’re doing DA from the LGM to present…and also for the PETM! • Looking for someone with good quant skills, and expertise in paleoclimate/climate dynamics/climate modeling.