Using O/Ar Measurements and Satellite Observations

hostitchAI and Robotics

Oct 23, 2013 (3 years and 9 months ago)

98 views

Zuchuan

Li, Nicolas
Cassar


Division of Earth and Ocean Sciences

Nicholas School of the Environment

Duke University


Estimation of Net Community Production (NCP)
Using O
2
/
Ar

Measurements and Satellite
Observations

Overall objective


Develop an independent
estimate of
global Net
Community Production (NCP
)


1.
A large independent training dataset : O
2
/
Ar
-
derived NCP


2.
Satellite observations


3.
Statistical methods:


Support Vector
R
egression


Genetic Programming



Compare to current algorithms of export
production

Examples of current export production
algorithms


Laws et al. (2000)







Dunne et al. (2005 & 2007)

pe

ratio
=

0
.
0081


+
0
.
0806

ln
𝐶ℎ𝑙
+
0
.
426

0.04 <
pe
-
ratio
< 0.72

e
f
-
Ratio

Export production ~
NPP
* Export ratio






Base of the mixed layer

Atmosphere

O
2
/
Ar
-
derived NCP

NCP
~
Δ
[O
2
]
bio
sat
*gas
exchange coefficient







1.
NCP


Gross

Primary

Production

(GPP)



C
ommunity

respiration


Net

Primary

Production

(NPP)



Heterotrophic

respiration

2.
NCP

estimation


O
2
/
Ar

measurements


Satellite

observations

(e
.
g
.

NPP

and

SST)

3.
Uncertainties

in

O
2
/
Ar

measurements


S
ee

Reuer

et

al
.

2007
,

Cassar

et

al
.

2011
,

Jonsson

et

al
.

2013

Photosynthesis

(GPP)

Auto
-

&
hetero
-

trophic


respiration

NCP

CO
2

Organic matter + O
2

Total O
2
/
Ar

Observations

N = 14795 (9km)

S
atellite match
observations

N = 3874

1.
SeaWiFS

1)
NPP (from VGPM)

2)
POC

3)
Chl
-
a

4)
phytoplankton
size structure
(Li et al.
2013)

5)
Rrs
(
λ
)

6)
PAR

2.
Others

1)
SST

2)
Mixed
-
layer
depth (
Hosoda

et al. 2010)

Filter with
Rossby

Radius

N = 722

NCP vs. satellite observations


Increases with productivity
and biomass:


NPP


POC


Chl
-
a







Decreases trend with:


SST










Displays nonlinearity and
scatter

Statistical
algorithms

Genetic programming

(Schmidt
and Lipson
2009
)



Theory:
Search for the form of
equations and their
coefficients




Input:
NPP,
Chl
-
a, POC,
SST …



Output:
Equations


Support vector regression

(
Vapnik

2000
)



Theory:
Search
for
a nonlinear
model
within an error and as flat
as
possible



Input:
NPP,
Chl
-
a, POC, SST



Output:
Implicit model


Model validation


Equation from genetic programming
:



Observed NCP

Predicted NCP

𝑁𝐶𝑃
=
𝑁𝑃𝑃
12
.
6
+
1
.
5



Genetic Programming

Observed NCP

Predicted NCP

Support Vector Regression

Observed NCP

Predicted NCP

NCP has units of (
mmol

O
2

m
-
2

day
-
1
)

Comparison

A.
Eppley
:
Eppley

and Peterson (1979)

B.
Betzer
:
Betzer

et al. (1984)

C.
Baines: Baines et al. (1994)

D.
Laws: Laws et al. (2000)

E.
Dunne: Dunne et al. (2005 & 2007)

F.
Westberry
:
Westberry

et al. (2012)

G.
This study (GP): genetic programming

H.
This study (SVR): support vector regression

Differences between algorithms


Consistent regions:


North Atlantic


North Pacific


R
egion around 45
o

S



Regions with large
discrepancy:


Oligotrophic gyres


Southern Ocean


Arctic Ocean



Possible reasons:


Limited observations


Different


Field methods


Measured properties


Uncertainties in satellite
products ([
Chla
], NPP
(VGPM), etc.)

(CV
: coefficient of
variation)

Comparison with Laws et al. 2000


GP(this study)
/
Laws


Consistent in most regions


Our algorithm predicts higher NCP in:


Southern Ocean


T
ransitional regions

GP(this study)/Laws

Conclusions


Our
method shows a relatively good agreement to other
models


With a completely independent training dataset and scaling methods



However:


Our algorithms predict more uniform carbon fluxes in the world’s oceans


Discrepancies are observed in some regions, such as Southern Ocean where
our algorithms generally predict higher NCP



Work in progress…


Develop region
specific
algorithms


Test consistency of the genetic programming solutions
and
transferability


Test with additional datasets

Acknowledgements


All of our O
2
/
Ar

collaborators for providing
the field observations

Thank you!

Dissolved O
2
/
Ar
-
based NCP


O
2
/
Ar

measurement




[O
2
] contributed to biological process




NCP








Base of the mixed layer

NCP =
D
[O
2
]
sat
*gas exchange coefficient




NCP
=
Net (POC + DOC) change


Atmosphere

NCP=Photosynthesis
-
Respiration


Assumptions, Limitations, Uncertainties:


No
mixing across base of mixed
layer


Steady
-
state (see Hamme et al. 2012)


Restricted
to the
whole mixed
layer


Gas exchange
parameterized
in terms of
windspeed

Argon
:
Inert gas which has similar solubility properties as oxygen


O
2
/
Ar
-
based NCP measurement

Validation


Genetic programming


A:
𝑁𝐶𝑃
=

12
.
6
+
1
.
5




B:
𝑁𝐶𝑃
=
16
.
5
+
0
.
0198

𝑁𝑃𝑃

0
.
617




C:
𝑁𝐶𝑃
=

0
.
117

𝐶
+
1
.
61