Estimation of Large-scale Net Community Production Patterns

Report
Estimation of Net Community Production (NCP)
Using O2/Ar Measurements and Satellite
Observations
Zuchuan Li, Nicolas Cassar
Division of Earth and Ocean Sciences
Nicholas School of the Environment
Duke University
Overall objective
• Develop an independent estimate of global Net
Community Production (NCP)
1.
A large independent training dataset : O2/Ar-derived NCP
2.
Satellite observations
3.
Statistical methods:
 Support Vector Regression
 Genetic Programming
• Compare to current algorithms of export
production
Examples of current export production
algorithms
• Laws et al. (2000)
ef-Ratio
Export production ~ NPP * Export ratio
• Dunne et al. (2005 & 2007)
pe−ratio = −0.0081 ∗  + 0.0806 ∗ ln ℎ + 0.426
0.04 < pe-ratio < 0.72
O2/Ar-derived NCP
Atmosphere
NCP ~ Δ[O2]biosat*gas exchange coefficient
Photosynthesis (GPP)
CO2
NCP
Organic matter + O2
Auto- & hetero- trophic
respiration
Base of the mixed layer
1. NCP
•
Gross Primary Production (GPP) – Community respiration
•
Net Primary Production (NPP) – Heterotrophic respiration
2. NCP estimation
•
O2/Ar measurements
•
Satellite observations (e.g. NPP and SST)
3. Uncertainties in O2/Ar measurements
•
See Reuer et al. 2007, Cassar et al. 2011, Jonsson et al. 2013
Total O2/Ar
Observations
N = 14795 (9km)
Satellite match
observations
N = 3874
1.
2.
SeaWiFS
1) NPP (from VGPM)
2) POC
3) Chl-a
4) phytoplankton size structure
(Li et al. 2013)
5) Rrs(λ)
6) PAR
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)
Support vector regression
(Vapnik 2000)
• Theory: Search for the form of
equations and their coefficients
• Theory: Search for a nonlinear
model within an error and as flat
as possible
• Input: NPP, Chl-a, POC, SST …
• Input: NPP, Chl-a, POC, SST
• Output: Equations
• Output: Implicit model
Model validation
• Equation from genetic programming:

12.6 + 1.5 ∗ 
Genetic Programming
Support Vector Regression
Predicted NCP
Predicted NCP
 =
Observed NCP
NCP has units of (mmol O2 m-2 day-1)
Observed NCP
Comparison
A.
B.
C.
D.
E.
F.
G.
H.
Eppley: Eppley and Peterson (1979)
Betzer: Betzer et al. (1984)
Baines: Baines et al. (1994)
Laws: Laws et al. (2000)
Dunne: Dunne et al. (2005 & 2007)
Westberry: Westberry et al. (2012)
This study (GP): genetic programming
This study (SVR): support vector regression
Differences between algorithms
• Consistent regions:
– North Atlantic
– North Pacific
– Region around 45o 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
• Transitional 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 O2/Ar collaborators for providing
the field observations
Thank you!
Dissolved O2/Ar-based NCP
• O2/Ar measurement
• [O2] contributed to biological process
• NCP
O2/Ar-based NCP measurement
Atmosphere
NCP = D[O2]sat*gas exchange coefficient
NCP=Photosynthesis-Respiration
NCP = Net (POC + DOC) change
Base of the mixed layer
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
Validation
• Genetic programming
– A:  =

12.6+1.5∗
– B:  = 16.5 + 0.0198 ∗
 − 0.617 ∗ 
– C:  =

0.117∗+1.61∗

similar documents