Inferred δ 13 C and δ 18 O distributions in the modern and Last Glacial Maximum deep Atlantic...

Post on 21-Jan-2016

216 views 0 download

Transcript of Inferred δ 13 C and δ 18 O distributions in the modern and Last Glacial Maximum deep Atlantic...

Inferred δ13C and δ18O distributions in the modern and Last Glacial Maximum deep Atlantic

Holly Dail

hdail@fas.harvard.edu

ECCO Meeting

November 1, 2012.

blah blah blah……

Last Glacial Maximum

Artist’s renditionImbrie & Imbrie 1979

Last Glacial Maximum climate– Generally colder– Sea level lower by ~120 m– CO2 ~190 ppm

Standard approaches + limitations– Compilations of proxy records– Model intercomparison

projects

After Waelbroeck et al. (2002)

Sea level

450,000years ago

LGM

Ocean circulation model+ model adjoint

δ18O171 records

δ13C171 records

Alkenones55 records

Forams181records

Mg/Ca32 records

Deep ocean isotope dataNear sea surfacetemperatures

Dinocysts53 records

Seek estimate of Atlantic Ocean state at the LGM that is consistent with

available data ocean general circulation model

Key questions in this talk

Is it possible to find a model state close to the available LGM data?

Are LGM δ13C proxies consistent with an AMOC shallower than today’s?

75°N

35°S

100°W 20°E

LGM bathymetry based on ICE-5G (Peltier, 2004)

MIT GCM + adjoint

Model– MIT GCM ocean / sea ice

model– Prescribed atmosphere– 1° degree, 50 vertical levels– Open southern boundary

Cost function– Misfit to MARGO SSTs

• uncertainties: as published by MARGO NGS’09

– Misfit to deep ocean isotope data

• uncertainties: 0.2 ‰

– Penalties on controls

mean

State estimate fits the SST proxies better than PMIP2 coupled model simulations do

PMIP2: Braconnot et al., 2007

Sta

te e

stim

ate

Incorporating a simple model for isotopes[draws on Marchal and Curry, JPO 2008]

δ13C– δ13Ccalcite = 0.13 + 0.90 δ13CDIC

– δ13CDIC modeled as a passive tracer + remineralization

– Uncertainty = 0.2‰

– Data excluded above 1000 m

– δ13CDIC model tested against modern GEOSECS δ13C

δ18O – not discussed here

New controls for initial and boundary conditions on tracers

A gap in available modeling methods

Paleoclimate simulations w/ coupled models (e.g. PMIP): – run to equilbrium, but model state may be far from

data

Standard ocean state estimation (e.g. ECCO): – close to data, but deep ocean not in equilibrium

Our solution: long-running state estimates– goal: build connection between deep ocean and model

forcing– results: reach maximum 80 years -- still not long

enough

Modern: GEOSECS δ13CDIC distribution at 2730 m

(‰)

10 year state estimate 80 year state estimate

1000 m 1900 m 2500 m

3000 m 3900 m 4600 m

LGM: 93% of modeled δ13C values are within 2σ of the proxy data

(‰)

Benthic δ13C records: shifts in water mass boundaries?

Observation:– weak gradients with

depth in modern ocean– strong gradients with

depth at LGM

Inference:– deep Atlantic dominated

by AABW?– changes in end-member

concentrations?Curry & Oppo (2005)

Curry & Oppo (2005)

State estimate δ13C at 28°W

State estimate δ13C at 28°W

Modern AMOC streamfunction (Sv)(OCCA, Forget JPO 2010)

LGM AMOC streamfunction (Sv)(Inferred with state estimation)

• Results are preliminary: - model partially equilibrated- tendency towards weak MOC in modern tests- other AMOC arrangements have not been excluded

Key questions

Is it possible to find a model state close to the available LGM data?– Unconstrained coupled models: not yet– With state estimation: yes

Are LGM δ13C proxies consistent with an AMOC shallower than today’s?– Yes