Can SST Proxy Data Reconstruct Common Era AMOC variability?Casey Saenger1 Mike Evans2

1. JISAO, University of Washington 2.Dept. of Geology, University of Maryland

Summary:Proxy surrogate reconstructions (PSRs) combine the strengths of proxy data and climate simulations. We construct PSRs using the PAGES Ocean2K metadatabase of sea surface temperature annomalies (SSTa) and a subset of CMIP5 piControl, past1000 and historical simulations, with the goal of reconstructing Atlantic Meridional Overturning Circulation (AMOC) and quantifying reconstruction skill. Proxy data appear sufficient for centennial resolution, but greater variance than observed in simulations compromises skill. Empiri-cal scaling of proxy data improves skill, but further work is needed to determine if such a scaling is realistic. Preliminary results suggest subtle AMOC variations of +/- 0.5 Sv over the past 2000 yrs without clear trends during the Little Ice Age or Medieval Climate Anomaly.

Calibration/Validation:

Uk'37Mg/CaTEX86diatom transfer func.d18O1 sample/100 yrs20 samples/100 yrs100 samples/100 yrs

Proxy Surrogate Reconstructions (PSRs)• Simulations are physically realistic, but don’t necessarily track the climate’s true evolution• Proxy data record actual climate climate, but are spatially irregular and noisy.• PSRs are an analog approach to utilize the strengths of proxy and GCM data (Graham et al. 2007) 1. Compile a network of paleoclimate data. Here, Ocean2K SST reconstructions + 2. Construct a catalog of model-based estimates of the same variable. Here, CMIP5 historical, past1000 (CCSM4, MPI-ESM3) and piControl (CCSM4, MPI-ESM3, CNRM and INMC) 3. When simulations skillfully capture proxy SST, all model variables can be accessed

What temporal resolution is possible?

97% of proxy data has average sampling resolution of 1 sample/100 years or higher.100 year binned averages, shifted every 50 years yields 36-53 records per bin

Other important considerations and what we’ve learned

1 3 10 30 100 samples/100yr

n

calibration 50-75% of recordsEuclidian distance

√ ∑(SSTacal -SSTasim)2

pearson rp value

intercept with 95% ci

m best analogs

validation 25-50% of records

Euclidan distancepearson rp value

intercept with 95% ciRECE

AMOC, NAM, etc. in m best analogs

CE

Absolute SST vs. anomalies: • Absolute SST: Strong statistics just capture climatology • Best choice: Use anomalies

Time period for anomalies: •1200-1400CE: Maximum data density, but spurious statistics because proxy/model are “told” to equal each other • 0-2000CE: Fixes problem above, but excludes most records • Best choice: Calculate anomalies over a proxy record’s entire interval. This is unique for each record and requires individual ized catalogs of simulations with anomalies calculated in the same way (for past1000).

Seasonality: • Annual only: Assume seasonal anomalies will be very similar to mean annual at 100 year intervals • Annual, Spring and Summer: Include MAM and JJA in simulation catalog searched for PSR • Best choice: Individualized catalog uses seasonality re ported by original publication

Skill metrics: • calibration/validation r >0 • calibration/validation p < 0.1 • intercept within error of 0 • validation RE > 0 • validation CE > 0 • Best choice: We adopt 3 highest RE, but are open to ideas.

Calibration/validation ratio: • Appears largely insensitive to 50/50, 66/33 and 80/20

Weighting: • Appears largely insensitive to no weight vs. 1/s.d.

0 500 1000 1500 2000 CE

CariacoBasin.Black.2007CariacoBasin.Lea.2003GarrisonBasi.Richey.2009FiskBasinGul.Richey.2009EM9606M200.Richter.2009Laurent.Fan.Keigwin.2005EmeraldBasin.Sachs.2007VirginiaSlope.Sachs.2007ODP984.Came.2007RAPiD121KSou.Thornalley.2009NorthIceland.Sicre.2011SouthIceland.Sicre.2011CH0798MC22Ca.Saenger.2011KNR140-2.2010.Saenger.2011GulfofGuinea.Weldeab.2007GreatBahamaA.Lund.2006GreatBahamaB.Lund.2006DryTortugasA.Lund.2006DryTortugasB.Lund.2006EmeraldBasin.Keigwin.2003PigmyBasinGu.Richey.2007SouthAtlanti.Leduc.2010CapeGhirNWAf.McGregor.2007East.trop.Atl.Kuhnert.2011Tagusmudpatc.Abrantes.2005CapeHatteras.Cleroux.2012MD952011.Calvo.32CariacoBasin.Goni.2006CapeGhirNWAf.Kim.2006RAPiD-35-25B_POC_Moffa2014RAPiD-17-5P_NGS_Moffa14RAPiD-35-25B NGS Moffa14AI07-03G_AI07-04BC_SST_Sicre14AI07-12G_AI07-11BC_SST_Sicre14Kanger_SST_Miettinen15MD99-2275_DiatomSST_Jiang1572GGC_73BC_CAR25-1_SST-Mg/Ca72GGC_73BC_CAR25-1_SST-Mg/CaHU91-045-093 Hoogakker15WEqPacific.Stott.2007MakassarStra.Linsley.2010EasternTropicalPacific.Rustic.2015Philippines.Stott.2007MakassarStra.Newton.2011MakassarStra.Newton.32NorthwestPac.Harada.2004MakassarStra.Oppo.2009PanamaBasin(.Pahnke.2007PanamaBasin(.Pahnke.2007SouthernChil.Mohtadi.2007Chileanmargi.Lamy.2002KuroshioCurr.Isono.2009PiscoCentral.Gutierrez.2011OkinawaTroug.Wu.2012ArabianSea.Doose-Rolinski.2001SWcoastofInd.Saraswat.2013JacafFjord.Sepulveda.2009WesternAntar.Shevenell.2011EM9606M200.Richter.2009Laurent.Fan.Keigwin.2005Laurent.Fan.Keigwin.2005ODP984.Came.2007RAPiD121KSou.Thorlley.2009RAPiD121KSou.Thorlley.2009CH0798MC22Ca.Saenger.2011KNR140-2.2010.Saenger.2011GulfofGuinea.Weldeab.2007DryTortugasA.Lund.2006DryTortugasB.Lund.2006GreatBahamaB.Lund.2006GreatBahamaA.Lund.2006RAPiD-35-25B_POC_Moffa2014RAPiD-35-25B_POC_Moffa2014RAPiD-17-5P_NGS_Moffa14RAPiD-35-25B NGS Moffa14GeoB13862-1 Voigt15GeoB6211-2 Voigt15GeoB6308-3 Voigt15HU91-045-093 Hoogakker15Emerald Basin MC-29D Keigwin03Emerald Basin MC-29D Keigwin03N.Iceland.Arctica Wanamaker12PRB-PRP12 Nyberg02P1-003McSC Sejrup10Cariaco.bulloides Black04Cariaco.ruber Black04Ce96 HaaseSchramm03Pb19 HaaseSchramm03MD95.2011 Risebrobakken03MD95.2011 Risebrobakken03WEqPacific.Stott.2007MakassarStra.Linsley.2010EasternTropicalPacific18OMg/CaSSTPhilippines.Stott.2007MakassarStra.Newton.2011MakassarStra.Newton.32MakassarStra.Oppo.2009ArabianSea.Doose-Rolinski.2001ArabianSea.Doose-Rolinski.2001SWcoastofInd.Saraswat.2013SouthernChil.Mohtadi.2007

SST

δ18

O

SSTa (ºC)

n

-3 -2 -1 0 1 2 3

0

2000

4000

6000

8000

10000

12000 | | 1950| | 1900| | 1850| | 1800| | 1750|| | 1700| | 1650| | 1600| | 1550| | 1500| | 1450|| | 1400| | 1350| | 1300| | 1250| | 1200| | 1150| | 1100|| | 1050|| | 1000| | | 950| | 900| | 850|| | 800| | 750| | 700| | 650| | 600| | 550| | 500| | 450|| | 400| | 350| | 300| | 250| | 200| | 150|| | 100|| | 50

2σ proxy 2σ model

• No strong model bias• Proxy age similar to time in past 1000 simulations.• Historical simulations favored by proxy data before ~500 CE

• Significant calibrations (median r>0, median p<0.1), but proxy cool-ing trend not seen in simulations.• Mixed validation statistics (r>0, but no p<0.1) and similar bias• If median r>0, median RE is often >0• No positive median CE

0 500 1000 1500 2000

-1.0

-0.5

0.0

0.5

1.0

Index

AM

OC

ano

mal

y (S

v)

----------------------

----------------- ------------------------ ------ ------ ------ ---------- -------------- ----------------- ---------------------------- ------ ------ ---------- -------

0 0.25 0.5 0.75 1RE

0 500 1000 1500 2000

0

1

2

3

4

Index

calib

ratio

n st

atis

tics

---------------------------------------

-------

-------------

------

----

----

-

----

pearson r -log (p)

0 500 1000 1500 2000

0

1

2

3

4

valid

atio

n st

atis

tics

------------

-------------------------

-- ------------

----------------

------

-----

-------------------------------------

--pearson r -log (p) CE

0 500 1000 1500 2000

-1.0

-0.5

0.0

0.5

1.0

AM

OC

ano

mal

y (S

v)

------------------------

--------------- -

------0 0.25 0.5 0.75 1

RE

0 500 1000 1500 2000

0

1

2

3

4

calib

ratio

n st

atis

tics

---------------------------------------

---

----

-----

--

-------

-

-

----

---

-------

pearson r -log (p)

0 500 1000 1500 2000

0

1

2

3

4

valid

atio

n st

atis

tics

--------------------------------------- --

--

--

-

-----

--

-

-

--

-

--

--------

-----

---

----------------------

-----------------pearson r -log (p) CE

• Median proxy SSTa data in each time bin (red circles) overlaps with SSTa variance in simulation catalog (black histogram).

• Individual proxy records (open grey circles) can show appreciably larger variance, with SSTa values that don’t have an analog in simu-lation catalog. Marginal validation likely related to much larger variance

in proxy data than in simulations.

• Local scale processes not captured by simulations• Changes in proxy seasonality• Non-temperature effects on proxy data

What if proxy data were dampped so each proxy record had the same variance as its simulation grid point?

See below...

inter

cept

95%

ci

log

(p)

pe

arson

r

dis

tance

(ºC)

inter

cept

95%

ci

log

(p)

pe

arson

r

dis

tance

(ºC)

inter

cept

95%

ci

pears

on r

CE

inter

cept

95%

ci

pears

on r

CE

log (

p)

d

istan

ce (º

C)

RE

log (

p)

d

istan

ce (º

C)

RE

• Generally similar results (still no strong model bias, proxy age sim-ilar to time in past 1000 simulations, etc).• More significant p values and smaller cooling bias (probably a result of damping proxy data) • Very significant calibrations (median r>0, median p<0.1), and more significant validation statistics (r>0, more p<0.1)

• Dots mark iterations with the highest 3 RE in each time step• High RE usually means significant validation, with modest reduc-tion in calibration statistics

66/33, weighted, scaled

66/33, weighted, unscaled

Simulation SSTa

n

-3 -2 -1 0 1 2 3

0

2000

4000

6000

8000

10000

12000 | | 1950| | 1900| 1850180017501700| 1650| 1600| 15501500145014001350| 1300| 1250| 1200| 1150| | 1100||||| | 1050| 1000| 950|| 900| 850| 800|| | 750| 700| 650|| 600|| 550| | 500| 450| | 400| | 350|| | 300|||| 250|| 200|| | 150| | 100|| | 50

AMOC reconstructions Conclusionsa

b

c

a

b

c

AMOC reconstruction for unscaled PSR. a) median AMOC anomaly (+/- 1s.d.) for 3 highest RE iterations (color bar). Grey bars note time steps where validation r>0, p < 0.1. Intervals where cali-bration r <0 and p>0.1 are masked. b) calibration r (black) and -log p (blue). Dashed lines note r>0 and p<0.1. c) as in b for vali-dation r (black) and -log p (orange). CE (brown) is also shown.

AMOC reconstruction for scaled PSR. As in panel at left. • Results suggest a relatively stable AMOC with centennial scale anomalies of +/- 0.5 Sv that are smaller than the ~3 Sv suggested from Florida Straits transport (i.e. Lund et al., 2006).• 1900-2000 AMOC does not appear anomalous, but does not vali-date well.

• As in histogram above, but with proxy data scaled to have the same variance as simulation in the closest 1ºx1º gridbox

• (Unsurprising) better agreement with few (if any) no analog cases.

TOWARD AMOC RECONSTRUCTIONS

• Use the 3 highest RE in a timestep to derive AMOC• Balances over calibrating and under validating• Gives similar distances for calibration and validation• Each of the 3 highest RE values is based on the mean of 3 iterations.• Median AMOC in 9 simulations (replicates are possible)

• PAGES, Ocean2K proxy data has spatial and temporal coverage that seems suitable for centennial scale PSRs based on SST anomalies during the past two millennia• Calculating simulation anomalies relative to the same time frame and season appears reasonable.• Higher variance in proxy data relative to simulations cre-ates no-analog cases that may reflect local scale dynamics not captured by coarse simulations, proxy vital effects, or other processes• Empirical scaling of proxy data improves PSR skill, but is it realistic?• Reconstructed AMOC shows subtle anomalies of +/- 0.5 Sv. Unscaled reconstruction hints at a MCA to LIA reduction, but is less obvious in scaled data.

Future work• Construct PSRs from higher spatial resolution simula-tions that capture local scale dynamics (e.g. eddies)• Add additional CMIP simulations• Incorporate oxygen isotope data and isotope enabled simulations• Compare piControl, single forcing and multiple forcing simulations for detection/attribuition• Pseudoproxy experiments of how proxy data variance af-fects AMOC, and where new proxy data has largest impact.

Acknowledgments: This work was supported by NSF-OCE award 1536418 to CPS and MNE. We appreciate fruitful discussions with Greg Hakim, Nathan Steiger, Wei Cheng and Nick GrahamReferences: Graham, N.E. et al. 2007: Tropical Pacific - mid-latitude teleconnections in medieval times. Climatic Change 83, 241-285.Lund, D.C. et al. 2006: Gulf Stream density structure and transport during the past millennium. Nature. 444, 30, 601-604.