How well do proxy system (data) models simulate real...
Transcript of How well do proxy system (data) models simulate real...
Thomas Laepple, Thomas Münch, Andrew Dolman, Maria Reschke Alfred Wegener Institute, Germany
DPG 2012, Berlin
How well do proxy system (data) models simulate real paleoclimate observations?
DAPS 29. May 2017, (apples, oranges and other figures are copied from Sylvia Dee‘s talk)
We need tests for proxy system models that are independent of climate models
simulated proxies observed proxies
Testing option: replicate proxy data
partly known climate
Compare the replicability of true and simulated data
EDML
Ice-core proxy system model test site
Münch et al., CP 2016
Signal of two shallow firn cores 500m distance
Observed replicability
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
-55
-45
-35
depth (m)
d18O
(‰)
R=0.3
500m
Current ice PSM (e.g. Dee et al.,2015)
Measurement error, Time-uncertainty
Diffusion burial densification
dO18=f(T, …), altitude correction Precipitation weighting
Münch et al., CP 2016
Signal of two shallow firn cores 500m distance
Observed replicability
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
-55
-45
-35
depth (m)
d18O
(‰)
R=0.3
500m
Horizontal structure
0 10 20 30 40
100
80
60
40
20
0
trench position [m]
dept
h [c
m]
−55
−50
−45
−40
−35
δ18O
[p
erm
il]
●● ● ● ●● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
Horizontal structure
Temporal structure of depositional noise, Münch et al., in prep
500m
Spatial structure of depositional noise, Münch et al., CP 2016
Stratigraphic noise model needed
Testing option: Design field experiments to test proxy system model components
Diffusion burial densification
Look at the archive again after several years
Münch et al., TCD 2017
Look at the archive again after several years
Münch et al., TCD 2017
• Observed changes match the simulated changes
• No evidence for other changes
unknown climate
PSM for proxy type 1 observed proxy type 1
PSM for proxy type 2 observed proxy type 2
Testing option: different proxies of the same physical parameter
corals and deep sediment cores
corals Uk37 Mg/Ca
Laepple&Huybers, PNAS 2014
proxy spectra in temperature units
period (years)
PS
D (S
ST
varia
nce/
df)
0.0001
0.001
0.01
0.1
3000 1000 500 100 50 10 5 2
Coral SSTUk37 SSTMg/Ca SST
model at proxy position
Period (years) Laepple&Huybers, PNAS 2014
Sediment PSM + inversion (spectral correction algorithm)
Validation Laepple and Huybers, EPSL 2013
https://bitbucket.org/ecus/sedproxy Dolman and Laepple in prep.
MgCa= f(T) Uk37 = f(T)
Seasonal export
bioturbation Sampling a finite #tests Cleaning Analytical error
Spectral filter Sclimate = (Sclimate_M/Sproxy_PSM) * Sproxy
^ ^
proxy spectra in temperature units
period (years)
PS
D (S
ST
varia
nce/
df)
0.0001
0.001
0.01
0.1
3000 1000 500 100 50 10 5 2
Coral SSTUk37 SSTMg/Ca SST
model at proxy position
Period (years)
corrected spectra
period (years)
PS
D (S
ST
varia
nce/
df)
0.0001
0.001
0.01
0.1
3000 1000 500 100 50 10 5 2
Coral SSTUk37 SST correctedMg/Ca SST corrected
model at proxy position
Uk37 SST rawMg/Ca SST raw
Period (years)
simulated proxies observed proxies
Testing option: same proxy from nearby sites
partly known climate (right spatial covariance structure)
Compare the spatial covariance structure in true and simulated data
empirical vs. proxy model based Signal/Noise ratio estimate - Correlate pairs <5000km distance in model and proxy and
compare their covariance
Empirical estimate: S/NUk37 =1.3 (0.25-5) S/NMg/Ca =1.5 (0.20-5)
Proxy forward model: S/NUk37 =3 S/NMg/Ca =0.5
pUk37 <0.01
pMg/Ca <0.02
Laepple&Huybers, PNAS 14, supplement Reschke et al., in prep.
Conclusion
n Essential to test and validate PSM’s to avoid interpreting or assimilating wormholes
n Possibilities include replicate data, multiproxy data, the spatial covariance structure…
n Likely that PSM are too optimistic (e.g. ice-cores)
n Aim are PSM’s that are consistent with the process understanding as well as observational (paleo-)evidence
Are corals able to quantify SST variability?
-4.5 -4.0 -3.5 -3.0
-4.5
-4.0
-3.5
-3.0
2-10yr raw
log10 var coral SST
log1
0 va
r HA
DS
ST3
-4.5 -4.0 -3.5 -3.0
-4.5
-4.0
-3.5
-3.0
2-10yr both corrected
log10 var coral SST
log1
0 va
r HA
DS
ST
-5.5 -5.0 -4.5 -4.0 -3.5 -3.0
-5.5
-5.0
-4.5
-4.0
-3.5
-3.0
10-50yr raw
log10 var coral SST
log1
0 va
r HA
DS
ST3
-5.5 -5.0 -4.5 -4.0 -3.5 -3.0
-5.5
-5.0
-4.5
-4.0
-3.5
-3.0
10-50yr both corrected
log10 var coral SSTlo
g10
var H
AD
SS
T
Quinn 1998 dO18Dunbar 1994 dO18Asami 2005 dO18Hendy 2002 Sr/CaLinsley 2000 Sr/CaGoodkin 2008 Sr/CaCalvo et al., 2007 Sr/CaKilbourne et al., 2009 Sr/CaLinsley et al., 2006 Sr/CaSaenger 2009 growth rate
Calibration provided by the authors HadSST3 as provided
0.084 (mmol/mol SrCa)/C -0.23 permil δ18O/C, Mean replicate/uncertainty variability substracted from corals /ship data var(SST_coral)=var(coral)-var(replicate) var(SST_obs)=var(HadSST)-var(obsError)
Power spectra and sensitivity on proxy type
5 10 20 50 100 500
5e-05
5e-04
5e-03
5e-02
f (1/kyr)
PSD
all corals (10)only Sr/Ca (6)only dO18 (3)
5 10 20 50 100 500
5e-05
5e-04
5e-03
5e-02
f (1/kyr)
PSD
all corals, recalibratedall corals, authors calib
200 50 20 10 5 2 200 50 20 10 5 2
Period (years) Period (years)
known climate (e.g. instrumental observations)
simulated proxies observed proxies
Testing option 3: known climate, observed proxy
Compare time-series or power spectra
EDML Site, only densification
0.01 0.02 0.05 0.10
1e-02
1e+00
1e+02
f (1/cm)
PSD
EDML snow profilesAutomatic weather station AWS
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
-10
05
1015
input timeseries
depth (m)
dO18
ano
m
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
-10
-50
510
simulated
depth (m)
dO18
ano
m
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
-10
-50
510
EDML data
depth (m)
dO18
ano
m
0.01 0.02 0.05 0.10
1e-02
1e+00
1e+02
f (1/cm)
PSD
EDML snow profilesAutomatic weather station AWSAWS+Diffusion
EDML Site… adding diffusion
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
-10
05
1015
input timeseries
depth (m)
dO18
ano
m
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
-10
-50
510
simulated
depth (m)
dO18
ano
m
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
-10
-50
510
EDML data
depth (m)
dO18
ano
m
0.01 0.02 0.05 0.10
1e-02
1e+00
1e+02
f (1/cm)
PSD
EDML snow profilesAutomatic weather station AWSAWS+DiffusionAWS+aliasing/redistribution + Diffusion
EDML Site... Adding 80% conversion to white noise
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
-15
-50
515
input timeseries
depth (m)
dO18
ano
m
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
-10
-50
510
simulated
depth (m)
dO18
ano
m
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
-10
-50
510
EDML data
depth (m)
dO18
ano
m
Stratigraphic noise model needed
Test for the archive model: 2015 vs. 2013 trench data 1.) Estimated changes from
observations match the simulated changes 2.) No evidence for other changes
Münch et al., TCD 2017
Spectral filter Sxmodel/Symodel
Spectral correction algorithm (or sediment PSM + inversion)
synthetic climate records xmodel
S(f)=f-β,f<1/50yr
Bioturbation (2,10,20cm)
Sampling (as core)
+ measurement intrasample noise
σ
Synthetic proxy records ymodel
compare synthetic and real spectra
Estimate σ,β
Validation Laepple and Huybers, EPSL 2013
Validation https://bitbucket.org/ecus/sedproxy