Embed Size (px)
Citation preview
ICOS Open Science Conference, Helsinki, Sept 2016
CombiningOceanic And
Atmospheric Carbon Datato constrain CO2 fluxes
in Europe and its surrounding oceansChristian Rodenbeck
Max Planck Institute for Biogeochemistry, Jena
D. C. E. Bakker, B. Pfeil, G. Rehder, M. Glockzin,R. Keeling, and M. Heimann
In collaboration with
C. Le Quere, S. Zaehle
N. Gruber, Y. Iida, A.R. Jacobson, S. Jones, P. Landschutzer, N. Metzl, S. Nakaoka, A. Olsen,G.-H. Park, P. Peylin, K.B. Rodgers, T.P. Sasse, U. Schuster, J.D. Shutler, V. Valsala,
R. Wanninkhof, J. Zeng,
Many thanks to:Data contributors, DKRZ, CarboChange, IMBER / SOLAS
just-to-prevent-the-PDF-page-to-be-flipped
↓ ↓ ↓
just-to-prevent-the-PDF-page-to-be-flipped
↓ ↓ ↓
just-to-prevent-the-PDF-page-to-be-flipped
↓ ↓ ↓
←←←
Atmosphere
ZOT
NOAA/GMD
MLO
SPO
340
360
380
1980 1984 1988 1992 1996 2000 2004
CO2MLO
Atmosphere
◮l l l
Causality: Fluxes, Transport −→ Concentration gradientKnowledge:
∑Fluxes ←− Concentration gradient, Transport
cmeas ←→ cmod = Af + c0
“Atmospheric Inversion” = Multidimensional linear regression
Schauinsland - SCH (night time)
360
370
380
390
400
2002 2003 2004
Monte Cimone - CMN (day time)
360
370
380
390
400
2002 2003 2004
Schauinsland - SCH (night time)
360
370
380
390
400
2002 2003 2004
Monte Cimone - CMN (day time)
360
370
380
390
400
2002 2003 2004
W Europe
1990 1995 2000 2005 2010 2015year (A.D.)
-1.5-1.2-0.9-0.6-0.30.00.30.6
CO
2 Flu
x (P
gC/y
r)
s90 v3.8
L
CO2 uptake
KCO2 release
Schauinsland - SCH (night time)
360
370
380
390
400
2002 2003 2004
Monte Cimone - CMN (day time)
360
370
380
390
400
2002 2003 2004
W Europe
1990 1995 2000 2005 2010 2015year (A.D.)
-1.2
-0.9
-0.6
-0.3
0.0
CO
2 Flu
x (P
gC/y
r)
s90 v3.8 (summer only)
L
More CO2 uptake
K
More CO2 release
Schauinsland - SCH (night time)
360
370
380
390
400
2002 2003 2004
Monte Cimone - CMN (day time)
360
370
380
390
400
2002 2003 2004
W Europe
1990 1995 2000 2005 2010 2015year (A.D.)
-1.2
-0.9
-0.6
-0.3
0.0
CO
2 Flu
x (P
gC/y
r)
s90 v3.8 (summer only)
L
More CO2 uptake
K
More CO2 release
Schauinsland - SCH (night time)
360
370
380
390
400
2002 2003 2004
Monte Cimone - CMN (day time)
360
370
380
390
400
2002 2003 2004
W Europe
1990 1995 2000 2005 2010 2015year (A.D.)
-1.2
-0.9
-0.6
-0.3
0.0
CO
2 Flu
x (P
gC/y
r)
s90 v3.8 (summer only)
L
More CO2 uptake
K
More CO2 release
∆ Temperature: ∆ Precipitation: (Jul-Sep) [Ciais et al., Nature (2005)]
2003 European climate anomaly
K
just-to-prevent-the-PDF-page-to-be-flipped
Atmosphere
CO2
Sea-air gas exchange
Carbon
CO2 fluxinferred from CO2 inversion
Pacific 15S-15N
1990 1995 2000 2005 2010year (A.D.)
-0.50
-0.25
0.00
0.25
0.50
Sea
-air
CO
2 flu
x (P
gC/y
r)
�
just-to-prevent-the-PDF-page-to-be-flipped
Atmosphere
CO2
Sea-air gas exchange
Carbon
CO2 fluxinferred from CO2 inversion
Pacific 15S-15N
1990 1995 2000 2005 2010year (A.D.)
-0.50
-0.25
0.00
0.25
0.50
Sea
-air
CO
2 flu
x (P
gC/y
r)
�
[Peylin et al., RECCAP (2013)]
But:
just-to-prevent-the-PDF-page-to-be-flipped
Mixed
layer
pCO2,
[DIC]
Sea-air gas exchange
Carbon
just-to-prevent-the-PDF-page-to-be-flipped
Mixed
layer
pCO2,
[DIC]
Sea-air gas exchange
Carbon
just-to-prevent-the-PDF-page-to-be-flipped
[http://www.socat.info/]
just-to-prevent-the-PDF-page-to-be-flipped
Data density / distribution
[www.socat.info]
just-to-prevent-the-PDF-page-to-be-flipped
Data density / distribution
[www.socat.info]
just-to-prevent-the-PDF-page-to-be-flipped
Statistical
Interpolation
Linear
Regression
Non-linear
Regression
Model-based
Regr./Tuning
Mapping methods
just-to-prevent-the-PDF-page-to-be-flipped
Statistical
Interpolation
Linear
Regression
Non-linear
Regression
Model-based
Regr./Tuning
Role of driver data
Modelassump-
tions
Mapping methods
just-to-prevent-the-PDF-page-to-be-flipped
Statistical
Interpolation
Linear
Regression
Non-linear
Regression
Model-based
Regr./Tuning
Role of driver data
Modelassump-
tions
Direct data signals
Bridging data gaps
−→ Interesting complementarity−→ Extracting robust features
SOCOM: Collating 14 mapping methods
Mapping methods
just-to-prevent-the-PDF-page-to-be-flipped
UEA-SIOceanFlux-SIJena-MLSCU-SCSEAOML-EMPUEx-MLRJMA-MLRUNSW-SOMLOETH-SOMFFNCARBONES-NNNIES-SOMNIES-NNPU-MCMCNIES-OTTM
Monthly pCO2 (uatm)
2003 2004 2005 2006320
340
360
380
400
420
Seasonality:
Most methods roughly agreeon phasing and amplitude
(also to Takahashi et al., 2009)
→ Seasonality well constrained from data
First SOCOM results [Rodenbec k et al., BG (2015)]
just-to-prevent-the-PDF-page-to-be-flipped
UEA-SI
Jena-MLS
AOML-EMP
JMA-MLR
UNSW-SOMLO
ETH-SOMFFN
NIES-SOM
NIES-NN
Yearly CO2 flux (PgC/yr)
1990 1995 2000 2005 20100.0
0.2
0.4
0.6
0.8
Interannual Variations (IAV):– Tropical Pacific:
* Biome with largest IAV* Link to ENSO
Methods selected / weighted byrelative IAV mismatch to SOCATv2
Thicker lines:methods better matching the data
also mutually agree more closely
First SOCOM results [Rodenbec k et al., BG (2015)]
just-to-prevent-the-PDF-page-to-be-flipped
UEA-SI
Jena-MLS
AOML-EMP
JMA-MLR
UNSW-SOMLO
ETH-SOMFFN
NIES-SOM
NIES-NN
Yearly CO2 flux (PgC/yr)
1990 1995 2000 2005 20100.0
0.2
0.4
0.6
0.8
Interannual Variations (IAV):– Tropical Pacific:
* Biome with largest IAV* Link to ENSO
UEA-SIJena-MLSAOML-EMPJMA-MLRUNSW-SOMLOETH-SOMFFNCARBONES-NNNIES-SOMNIES-NNPU-MCMC
Yearly CO2 flux (PgC/yr)
1990 1995 2000 2005 2010-3.0
-2.5
-2.0
-1.5
-1.0
-0.5
– Global Ocean:* Larger spread
due to poorly constrained areas
First SOCOM results [Rodenbec k et al., BG (2015)]
just-to-prevent-the-PDF-page-to-be-flipped
UEA-SI
Jena-MLS
AOML-EMP
JMA-MLR
UNSW-SOMLO
ETH-SOMFFN
NIES-SOM
NIES-NN
Yearly CO2 flux (PgC/yr)
1990 1995 2000 2005 20100.0
0.2
0.4
0.6
0.8
Interannual Variations (IAV):– Tropical Pacific:
* Biome with largest IAV* Link to ENSO
UEA-SIJena-MLSAOML-EMPJMA-MLRUNSW-SOMLOETH-SOMFFNCARBONES-NNNIES-SOMNIES-NNPU-MCMC
Yearly CO2 flux (PgC/yr)
1990 1995 2000 2005 2010-3.0
-2.5
-2.0
-1.5
-1.0
-0.5
︸ ︷︷ ︸
Little decadal change
︸ ︷︷ ︸
Increasing sink
First SOCOM results [Rodenbec k et al., BG (2015)]
just-to-prevent-the-PDF-page-to-be-flipped
Statistical
Interpolation
Linear
Regression
Non-linear
Regression
Model-based
Regr./Tuning
Role of driver data
Modelassump-
tions
Direct data signals
Bridging data gaps
−→ Interesting complementarity−→ Extracting robust features
SOCOM: Collating 14 mapping methods
Jena-MLS �
[Rodenbeck et al.]
� ETH-SOMFFN[Landschutzer et al.]
� UEx-MLR[Talk Ute Schuster]
Mapping methods
just-to-prevent-the-PDF-page-to-be-flipped
Mixed
layer
pCO2,
[DIC]
Sea-air gas exchange
Carbon
CO2 fluxinferred from pCO2 interpolation
�
Pacific 15S-15N
1990 1995 2000 2005 2010year (A.D.)
-0.50
-0.25
0.00
0.25
0.50
Sea
-air
CO
2 flu
x (P
gC/y
r)
just-to-prevent-the-PDF-page-to-be-flipped
Atmosphere
CO2
Mixed
layer
pCO2,
[DIC]
Sea-air gas exchange
Carbon
CO2 fluxinferred from CO2 inversion
Pacific 15S-15N
1990 1995 2000 2005 2010year (A.D.)
-0.50
-0.25
0.00
0.25
0.50
Sea
-air
CO
2 flu
x (P
gC/y
r)
�
inferred from pCO2 interpolation
�
just-to-prevent-the-PDF-page-to-be-flipped
Atmosphere
CO2
Mixed
layer
pCO2,
[DIC]
Sea-air gas exchange
Carbon
Pacific 15S-15N
1990 1995 2000 2005 2010 2015year (A.D.)
-0.50
-0.25
0.00
0.25
0.50
CO
2 Flu
x (P
gC/y
r)
CO2 fluxinferred from CO2 inversion
Pacific 15S-15N
1990 1995 2000 2005 2010 2015year (A.D.)
-0.50
-0.25
0.00
0.25
0.50
CO
2 Flu
x (P
gC/y
r)
�
inferred from pCO2 interpolation
�
just-to-prevent-the-PDF-page-to-be-flipped
Atmosphere
CO2
Mixed
layer
pCO2,
[DIC]
Sea-air gas exchange
Carbon
Pacific 15S-15N
1990 1995 2000 2005 2010 2015year (A.D.)
-0.50
-0.25
0.00
0.25
0.50
CO
2 Flu
x (P
gC/y
r)
CO2 fluxinferred from CO2 inversion
Pacific 15S-15N
1990 1995 2000 2005 2010 2015year (A.D.)
-0.50
-0.25
0.00
0.25
0.50
CO
2 Flu
x (P
gC/y
r)
�
inferred from pCO2 interpolation
�
South American Trop.
1990 1995 2000 2005 2010 2015year (A.D.)
-1.5-1.0
-0.5
0.0
0.5
1.01.5
CO
2 Flu
x (P
gC/y
r)
just-to-prevent-the-PDF-page-to-be-flipped
Atmosphere
CO2
Mixed
layer
pCO2,
[DIC]
Sea-air gas exchange
Carbon
North Atlantic Temp.
1990 1995 2000 2005 2010 2015year (A.D.)
-0.50
-0.25
0.00
0.25
0.50
CO
2 Flu
x (P
gC/y
r)
CO2 fluxinferred from CO2 inversion
North Atlantic Temp.
1990 1995 2000 2005 2010 2015year (A.D.)
-0.50
-0.25
0.00
0.25
0.50
CO
2 Flu
x (P
gC/y
r)
�
inferred from pCO2 interpolation
�
just-to-prevent-the-PDF-page-to-be-flipped
Atmosphere
CO2
Mixed
layer
pCO2,
[DIC]
Sea-air gas exchange
Carbon
North Atlantic Temp.
1990 1995 2000 2005 2010 2015year (A.D.)
-0.50
-0.25
0.00
0.25
0.50
CO
2 Flu
x (P
gC/y
r)
CO2 fluxinferred from CO2 inversion
North Atlantic Temp.
1990 1995 2000 2005 2010 2015year (A.D.)
-0.50
-0.25
0.00
0.25
0.50
CO
2 Flu
x (P
gC/y
r)
�
inferred from pCO2 interpolation
�
Europe
1990 1995 2000 2005 2010 2015year (A.D.)
-0.50
-0.25
0.00
0.25
0.50
CO
2 Flu
x (P
gC/y
r)
just-to-prevent-the-PDF-page-to-be-flipped
Atmosphere
CO2
Mixed
layer
pCO2,
[DIC]
Sea-air gas exchange
Carbon
Southern Ocean
1990 1995 2000 2005 2010 2015year (A.D.)
-0.50
-0.25
0.00
0.25
0.50
CO
2 Flu
x (P
gC/y
r)
CO2 fluxinferred from CO2 inversion
Southern Ocean
1990 1995 2000 2005 2010 2015year (A.D.)
-0.50
-0.25
0.00
0.25
0.50
CO
2 Flu
x (P
gC/y
r)
�
inferred from pCO2 interpolation
�
just-to-prevent-the-PDF-page-to-be-flipped
Mixed
layer
pCO2,
[DIC]
Ocean-internal sources/sinks
Sea-air gas exchange
Carbon
CO2 fluxinferred from pCO2 interpolation
�
Pacific 15S-15N
1990 1995 2000 2005 2010year (A.D.)
-0.50
-0.25
0.00
0.25
0.50
Sea
-air
CO
2 flu
x (P
gC/y
r)
[www.socat.info]
But:
just-to-prevent-the-PDF-page-to-be-flipped
Mixed
layer
pCO2,
[DIC]
Ocean-internal sources/sinks
Sea-air gas exchange
Carbon
CO2 fluxinferred from pCO2 interpolation
�
Pacific 15S-15N
1990 1995 2000 2005 2010year (A.D.)
-0.50
-0.25
0.00
0.25
0.50
Sea
-air
CO
2 flu
x (P
gC/y
r)
0
1
“Reduction of Uncertainty” of flux IAV
But:
just-to-prevent-the-PDF-page-to-be-flipped
Mixed
layer
pCO2,
[DIC]
Sea-air gas exchange
Carbon
just-to-prevent-the-PDF-page-to-be-flipped
Mixed
layer
pCO2,
[DIC]
Sea-air gas exchange
Ocean-internal sources/sinks
Carbon
just-to-prevent-the-PDF-page-to-be-flipped
Mixed
layer
pCO2,
[DIC]
Ocean-internal sources/sinks
, MLD, wind speed, XCO2, SLP,
Alk, SSS
Sea-air gas exchange
hd
dtC = fma(C) + finth
d
dtC = fma(C) + fint
Using parameterizations of– Solubility– Sea–air gas exchange– Carbonate chemistry– Mixed-layer tracer budget
Carbon
just-to-prevent-the-PDF-page-to-be-flipped
Mixed
layer
pCO2,
[DIC]
Sea-air gas exchange
Ocean-internal sources/sinks
Carbon Oxygen
• Biology:Redfield stoichiometry
RO:C ≈ −1.4
• Transport+Mixing:Common pathways
just-to-prevent-the-PDF-page-to-be-flipped
Mixed
layer
pCO2,
[DIC][O2]
Ocean-in
Sea-air gas exchange
Carbon Oxygen
• Biology:Redfield stoichiometry
RO:C ≈ −1.4
• Transport+Mixing:Common pathways
just-to-prevent-the-PDF-page-to-be-flipped
Mixed
layer
pCO2,
[DIC][O2]
Ocean-in
Sea-air gas exchange
Carbon Oxygen
• Biology:Redfield stoichiometry
RO:C ≈ −1.4
• Transport+Mixing:Common pathways
just-to-prevent-the-PDF-page-to-be-flipped
Atmosphere
O2
Mixed
layer
pCO2,
[DIC][O2]
Ocean-internal sources/sinks
Sea-air gas exchange
Carbon Oxygen
• Biology:Redfield stoichiometry
RO:C ≈ −1.4
• Transport+Mixing:Common pathways
just-to-prevent-the-PDF-page-to-be-flipped
Atmosphere
O2
Mixed
layer
pCO2,
[DIC][O2]
Ocean-internal sources/sinks
Sea-air gas exchange
Carbon Oxygen
CO2 fluxinferred from O2/N2
Pacific 15S-15N
1990 1995 2000 2005 2010year (A.D.)
-0.50
-0.25
0.00
0.25
0.50
Sea
-air
CO
2 flu
x (P
gC/y
r)
? �
just-to-prevent-the-PDF-page-to-be-flipped
Atmosphere
O2
Sea-air gas exchange
Carbon Oxygen
�
Pacific 15S-15N
1990 1995 2000 2005 2010year (A.D.)
-100
-50
0
50
100
Sea
-air
AP
O fl
ux (
Tm
ol/y
r)
APO fluxinferred from APO inversion
just-to-prevent-the-PDF-page-to-be-flipped
Atmosphere
O2
Mixed
layer
pCO2,
[DIC][O2]
Ocean-internal sources/sinks
Sea-air gas exchange
inferred from pCO2
�
Assumption: Linked a la Redfield
Carbon Oxygen
�
Pacific 15S-15N
1990 1995 2000 2005 2010year (A.D.)
-100
-50
0
50
100
Sea
-air
AP
O fl
ux (
Tm
ol/y
r)
APO fluxinferred from APO inversion
just-to-prevent-the-PDF-page-to-be-flipped
Atmosphere
O2
Mixed
layer
pCO2,
[DIC][O2]
Ocean-internal sources/sinks
Sea-air gas exchange
inferred from pCO2
�
Assumption: Linked a la Redfield
Carbon Oxygen
Pacific 15S-15N
1990 1995 2000 2005 2010year (A.D.)
-100
-50
0
50
100
Sea
-air
AP
O fl
ux (
Tm
ol/y
r)
APO fluxinferred from APO inversion
just-to-prevent-the-PDF-page-to-be-flipped
Atmosphere
O2
Mixed
layer
pCO2,
[DIC][O2]
Ocean-internal sources/sinks
Sea-air gas exchange
inferred from pCO2
�
Assumption: Linked a la Redfield
Carbon Oxygen
Pacific 15S-15N
1990 1995 2000 2005 2010year (A.D.)
-100
-50
0
50
100
Sea
-air
AP
O fl
ux (
Tm
ol/y
r)
APO fluxinferred from APO inversion
Ocean 15N-90N
-2000
-1000
0
1000
2000
Ocean 15S-15N
-2000
-1000
0
1000
2000
Sea
-air
AP
O fl
ux (
Tm
ol/y
r)
Ocean 90S-15S
Jan Jul Jan Jul
-2000
-1000
0
1000
2000
just-to-prevent-the-PDF-page-to-be-flipped
Atmosphere
O2
Mixed
layer
pCO2,
[DIC][O2]
Ocean-internal sources/sinks
Sea-air gas exchange
inferred from pCO2
�
Assumption: Linked a la Redfield
Carbon Oxygen
Pacific 15S-15N
1990 1995 2000 2005 2010year (A.D.)
-100
-50
0
50
100
Sea
-air
AP
O fl
ux (
Tm
ol/y
r)
APO fluxinferred from APO inversion
Ocean 15N-90N
-2000
-1000
0
1000
2000
Ocean 15S-15N
-2000
-1000
0
1000
2000
Sea
-air
AP
O fl
ux (
Tm
ol/y
r)
Ocean 90S-15S
Jan Jul Jan Jul
-2000
-1000
0
1000
2000
!
!
just-to-prevent-the-PDF-page-to-be-flipped
What do the atmospheric stations “see”?Testing existing and potential CO2 observations (RINGO)
just-to-prevent-the-PDF-page-to-be-flipped
W Europe
2012 2013 2014 2015year (A.D.)
-0.6
-0.3
0.0
0.3
0.6
CO
2 Flu
x (P
gC/y
r)
– “Known truth”(OCN, Zaehle et al., 2010)
– Retrieved from “synthetic data”
(s04 v3.8 sites)
(3-monthly anomalies)
What do the atmospheric stations “see”?Testing existing and potential CO2 observations (RINGO)
just-to-prevent-the-PDF-page-to-be-flipped
W Europe
2012 2013 2014 2015year (A.D.)
-0.6
-0.3
0.0
0.3
0.6
CO
2 Flu
x (P
gC/y
r)
– “Known truth”(OCN, Zaehle et al., 2010)
– Retrieved from “synthetic data”
(s04 v3.8 sites)
(3-monthly anomalies)
Northern Europe
2012 2013 2014 2015year (A.D.)
-0.3
-0.2
-0.1
0.0
CO
2 Flu
x (P
gC/y
r)
What do the atmospheric stations “see”?Testing existing and potential CO2 observations (RINGO)
just-to-prevent-the-PDF-page-to-be-flipped
W Europe
2012 2013 2014 2015year (A.D.)
-0.6
-0.3
0.0
0.3
0.6
CO
2 Flu
x (P
gC/y
r)
– “Known truth”(OCN, Zaehle et al., 2010)
– Retrieved from “synthetic data”
(s04 v3.8 sites)
(3-monthly anomalies)
Northern Europe
2012 2013 2014 2015year (A.D.)
-0.3
-0.2
-0.1
0.0
CO
2 Flu
x (P
gC/y
r)
– Retrieved from “synthetic data”
(s04 v3.8 sites +assumend FINNMAID data)
Travemunde↔ Helsinki
What do the atmospheric stations “see”?Testing existing and potential CO2 observations (RINGO)
just-to-prevent-the-PDF-page-to-be-flipped
Conclusions W Europe
1990 1995 2000 2005 2010-1.2
-0.9
-0.6
-0.3
0.0
CO
2 Flu
x (P
gC/y
ear)
Atmospheric CO2 data & inversion:• Constraint on land variability• Southern Ocean trends
just-to-prevent-the-PDF-page-to-be-flipped
Conclusions W Europe
1990 1995 2000 2005 2010-1.2
-0.9
-0.6
-0.3
0.0
CO
2 Flu
x (P
gC/y
ear)
Atmospheric CO2 data & inversion:• Constraint on land variability• Southern Ocean trends
Surface-ocean pCO2 data & mapping:Well-constrained ocean seasonality •
Ocean IAV constrained e.g. in Eq. Pac. •
Yearly CO2 flux (PgC/yr)
1990 1995 2000 2005 20100.0
0.2
0.4
0.6
0.8
just-to-prevent-the-PDF-page-to-be-flipped
Conclusions W Europe
1990 1995 2000 2005 2010-1.2
-0.9
-0.6
-0.3
0.0
CO
2 Flu
x (P
gC/y
ear)
Atmospheric CO2 data & inversion:• Constraint on land variability• Southern Ocean trends
Surface-ocean pCO2 data & mapping:Well-constrained ocean seasonality •
Ocean IAV constrained e.g. in Eq. Pac. •
Yearly CO2 flux (PgC/yr)
1990 1995 2000 2005 20100.0
0.2
0.4
0.6
0.8
Ship-based atmospheric CO2 meas.:• Testing impact of additional data• Potential for regional flux estimates
just-to-prevent-the-PDF-page-to-be-flipped
Conclusions W Europe
1990 1995 2000 2005 2010-1.2
-0.9
-0.6
-0.3
0.0
CO
2 Flu
x (P
gC/y
ear)
Atmospheric CO2 data & inversion:• Constraint on land variability• Southern Ocean trends
Surface-ocean pCO2 data & mapping:Well-constrained ocean seasonality •
Ocean IAV constrained e.g. in Eq. Pac. •
Yearly CO2 flux (PgC/yr)
1990 1995 2000 2005 20100.0
0.2
0.4
0.6
0.8
Ship-based atmospheric CO2 meas.:• Testing impact of additional data• Potential for regional flux estimates
Products available for download:Atmospheric CO2 inversion •
pCO2-based mixed-layer scheme •Combined products, sensitivity cases, atm. fields •
www.BGC-Jena.mpg.de/CarboScope/
JenaCarboScope
just-to-prevent-the-PDF-page-to-be-flipped
BACK-UP SLIDES
just-to-prevent-the-PDF-page-to-be-flipped
Statistical
Interpolation
Linear
Regression
Non-linear
Regression
Model-based
Regr./Tuning
Role of driver data
Modelassump-
tions
Direct data signals
Bridging data gaps
−→ Interesting complementarity−→ Extracting robust features
SOCOM: Collating 14 mapping methods
Mapping methods
just-to-prevent-the-PDF-page-to-be-flipped
Statistical
Interpolation
Linear
Regression
Non-linear
Regression
Model-based
Regr./Tuning
Role of driver data
Modelassump-
tions
Direct data signals
Bridging data gaps
−→ Interesting complementarity−→ Extracting robust features
SOCOM: Collating 14 mapping methods
Jena-MLS �
[Rodenbeck et al.]
� ETH-SOMFFN[Landschutzer
et al.]
Mapping methods
just-to-prevent-the-PDF-page-to-be-flipped
(18.%) Jena oc_v1.4S
(28.%) ETH-SOMFFN2016
Yearly CO2 flux (PgC/yr)
1985 1990 1995 2000 2005 2010 20150.1
0.2
0.3
0.4
0.5
Yearly pCO2 mismatch (uatm)
1985 1990 1995 2000 2005 2010 2015-50
-30
-10
10
30
50
Cross Validation
just-to-prevent-the-PDF-page-to-be-flipped
(18.%) Jena oc_v1.4S
(28.%) ETH-SOMFFN2016
Yearly CO2 flux (PgC/yr)
1985 1990 1995 2000 2005 2010 20150.1
0.2
0.3
0.4
0.5
Yearly pCO2 mismatch (uatm)
1985 1990 1995 2000 2005 2010 2015-50
-30
-10
10
30
50
(18.%) Jena oc_v1.4S
Yearly pCO2 (uatm)
1985 1990 1995 2000 2005 2010 2015360
380
400
420
440
460
Cross Validation
just-to-prevent-the-PDF-page-to-be-flipped
(18.%) Jena oc_v1.4S
(28.%) ETH-SOMFFN2016
Yearly CO2 flux (PgC/yr)
1985 1990 1995 2000 2005 2010 20150.1
0.2
0.3
0.4
0.5
Yearly pCO2 mismatch (uatm)
1985 1990 1995 2000 2005 2010 2015-50
-30
-10
10
30
50
(18.%) Jena oc_v1.4S
(100.%) Jena oc_v1.4S Benchmark
Yearly pCO2 (uatm)
1985 1990 1995 2000 2005 2010 2015360
380
400
420
440
460
“Benchmark”:Keep seasonality+trend, but no IAV→ Mismatch ≈ signal size→ “100% error”
Cross Validation
just-to-prevent-the-PDF-page-to-be-flipped
(18.%) Jena oc_v1.4S
(28.%) ETH-SOMFFN2016
Yearly CO2 flux (PgC/yr)
1985 1990 1995 2000 2005 2010 20150.1
0.2
0.3
0.4
0.5
Yearly pCO2 mismatch (uatm)
1985 1990 1995 2000 2005 2010 2015-50
-30
-10
10
30
50
Interpolation: Regression:Time-dep. DoF’s Constant DoF’s→ Any IAV possible → IAV from drivers
Cross Validation
just-to-prevent-the-PDF-page-to-be-flipped
(18.%) Jena oc_v1.4S
Yearly CO2 flux (PgC/yr)
1985 1990 1995 2000 2005 2010 20150.1
0.2
0.3
0.4
0.5
Yearly pCO2 mismatch (uatm)
1985 1990 1995 2000 2005 2010 2015-50
-30
-10
10
30
50
Cross Validation
just-to-prevent-the-PDF-page-to-be-flipped
(18.%) Jena oc_v1.4S(121.%) Jena oc_v1.4S (CrossVal5yr0)
Yearly CO2 flux (PgC/yr)
1985 1990 1995 2000 2005 2010 20150.1
0.2
0.3
0.4
0.5
Yearly pCO2 mismatch (uatm)
1985 1990 1995 2000 2005 2010 2015-50
-30
-10
10
30
50
No dataData
No dataData
No dataData
→ Data-only interpolationcannot bridge multi-year gaps
Cross Validation
just-to-prevent-the-PDF-page-to-be-flipped
(28.%) ETH-SOMFFN2016
(48.%) ETH-SOMFFN2016 (CrossVal5yr0)
Yearly CO2 flux (PgC/yr)
1985 1990 1995 2000 2005 2010 20150.1
0.2
0.3
0.4
0.5
Yearly pCO2 mismatch (uatm)
1985 1990 1995 2000 2005 2010 2015-50
-30
-10
10
30
50
No dataData
No dataData
No dataData
→ Regression against drivers(SST, SSS, Chl-a, atm. CO2)offers some bridging capacity
Cross Validation
just-to-prevent-the-PDF-page-to-be-flipped
(28.%) ETH-SOMFFN2016
(48.%) ETH-SOMFFN2016 (CrossVal5yr0)
Yearly CO2 flux (PgC/yr)
1985 1990 1995 2000 2005 2010 20150.1
0.2
0.3
0.4
0.5
Yearly pCO2 mismatch (uatm)
1985 1990 1995 2000 2005 2010 2015-50
-30
-10
10
30
50
No dataData
No dataData
No dataData
Yearly pCO2 mismatch (uatm)
1985 1990 1995 2000 2005 2010 2015-50
-30
-10
10
30
50
DataNo data
DataNo data
DataNo data
Cross Validation
just-to-prevent-the-PDF-page-to-be-flipped
(28.%) ETH-SOMFFN2016
(62.%) ETH-SOMFFN2016 (Unconstrained periods)
Yearly CO2 flux (PgC/yr)
1985 1990 1995 2000 2005 2010 20150.1
0.2
0.3
0.4
0.5
Yearly pCO2 mismatch (uatm)
1985 1990 1995 2000 2005 2010 2015-50
-30
-10
10
30
50
→ Regression against drivers
(SST, SSS, Chl-a, atm. CO2)offers some bridging capacity
Cross Validation
just-to-prevent-the-PDF-page-to-be-flipped
(28.%) ETH-SOMFFN2016
(35.%) ETH-SOMFFN2016, regr. SST & SSS
Yearly CO2 flux (PgC/yr)
1985 1990 1995 2000 2005 2010 20150.1
0.2
0.3
0.4
0.5
Yearly pCO2 mismatch (uatm)
1985 1990 1995 2000 2005 2010 2015-50
-30
-10
10
30
50
Chl-a data only available since 1998
– do SST and SSS suffice?
→ Less fitting capacity(but main modes similar)
Cross Validation
just-to-prevent-the-PDF-page-to-be-flipped
(28.%) ETH-SOMFFN2016(62.%) ETH-SOMFFN2016 (Unconstrained periods)(35.%) ETH-SOMFFN2016, regr. SST & SSS(77.%) ETH-SOMFFN2016, regr. SST & SSS (Unconstr.)
Yearly CO2 flux (PgC/yr)
1985 1990 1995 2000 2005 2010 20150.1
0.2
0.3
0.4
0.5
Yearly pCO2 mismatch (uatm)
1985 1990 1995 2000 2005 2010 2015-50
-30
-10
10
30
50
Chl-a data only available since 1998
– do SST and SSS suffice?
→ Less fitting capacity(but main modes similar)
→ Less bridging capacity
Cross Validation
just-to-prevent-the-PDF-page-to-be-flipped
(18.%) Jena oc_v1.4S(16.%) Jena oc_v1.4S, regr. SST & SSS(131.%) Jena oc_v1.4S, regr. SST & SSS (Unconstr.)
Yearly CO2 flux (PgC/yr)
1985 1990 1995 2000 2005 2010 20150.1
0.2
0.3
0.4
0.5
Yearly pCO2 mismatch (uatm)
1985 1990 1995 2000 2005 2010 2015-50
-30
-10
10
30
50
Cross Validation
just-to-prevent-the-PDF-page-to-be-flipped
(47.%) ETH-SOMFFN2016
Yearly CO2 flux (PgC/yr)
1985 1990 1995 2000 2005 2010 2015-0.4-0.3
-0.2
-0.1
0.0
0.10.2
Yearly pCO2 mismatch (uatm)
1985 1990 1995 2000 2005 2010 2015-50
-30
-10
10
30
50
Southern Ocean – sparse data
Cross Validation
just-to-prevent-the-PDF-page-to-be-flipped
(47.%) ETH-SOMFFN2016
(106.%) ETH-SOMFFN2016 (Unconstrained periods)
Yearly CO2 flux (PgC/yr)
1985 1990 1995 2000 2005 2010 2015-0.4-0.3
-0.2
-0.1
0.0
0.10.2
Yearly pCO2 mismatch (uatm)
1985 1990 1995 2000 2005 2010 2015-50
-30
-10
10
30
50
Southern Ocean – sparse data
→ Bridging difficult &difficult to test
Cross Validation
just-to-prevent-the-PDF-page-to-be-flipped
(47.%) ETH-SOMFFN2016(106.%) ETH-SOMFFN2016 (Unconstrained periods)(53.%) ETH-SOMFFN2016, regr. SST & SSS
Yearly CO2 flux (PgC/yr)
1985 1990 1995 2000 2005 2010 2015-0.4-0.3
-0.2
-0.1
0.0
0.10.2
Yearly pCO2 mismatch (uatm)
1985 1990 1995 2000 2005 2010 2015-50
-30
-10
10
30
50
Southern Ocean – sparse data
→ Bridging difficult &difficult to test
→ again main modes similar w/o Chl-a
Cross Validation
just-to-prevent-the-PDF-page-to-be-flipped
(37.%) Jena oc_v1.4S
(47.%) ETH-SOMFFN2016
Yearly CO2 flux (PgC/yr)
1985 1990 1995 2000 2005 2010 2015-0.4-0.3
-0.2
-0.1
0.0
0.10.2
Yearly pCO2 mismatch (uatm)
1985 1990 1995 2000 2005 2010 2015-50
-30
-10
10
30
50
Southern Ocean – sparse data
→ Bridging difficult &difficult to test
→ again main modes similar w/o Chl-a
→ Decadal trends also from data directly
Cross Validation
just-to-prevent-the-PDF-page-to-be-flipped
(25.%) ETH-SOMFFN2016
(18.%) Jena-MLS14SSS
Yearly CO2 flux (PgC/yr)
1985 1990 1995 2000 2005 2010 2015-2.50-2.25-2.00-1.75-1.50-1.25-1.00-0.75
Yearly pCO2 mismatch (uatm)
1985 1990 1995 2000 2005 2010 2015-50
-30
-10
10
30
50
Global Ocean flux– affected by data-sparse regions
→ Complementary mapping methods(interpolation, regression)help to assess robustness
Cross Validation
