Algorithms for the Prediction of Coastal CO2

1
Algorithms for the Prediction of Coastal
CO2 Burke Hales1 Dick Feely2 Ricardo
Letelier1 Chris Sabine2 Pete Strutton1 1
Oregon State University 2 NOAA Pacific Marine
Environmental Laboratory
2
Summer pCO2 sea surface variability off the
Oregon Coast
Illustration of the large near-shore variability
of surface XCO2, both as a function of latitude
(upper panel), and time at a fixed latitude
(Cascade Head, at 45N lower panel), and the
consistency of the low XCO2 observations covering
most of the shelf.
3
- Coastal CO2 data recently compiled.
LDEO, MBARI, OSU, AOML, UGA databases contain
106 coastal surface pCO2 measurements dating to
1979 that were excluded from global compilations.
These data were mapped into 1 x 1 pixels within
3 from the coastline and monthly-mean fluxes
were calculated for each pixel.
Integrating fluxes from coastal pixels, the
bottom line, in Tg C yr-1 Total 2
35 Mexico 45 14 US -21 18 Canada -22 27
4
So, is the coastal air-sea flux just a small
number with large uncertainties? Maybe. But we
still have an undersampling problem.
Even in well-sampled regions, we still suffer
from uncertainties related to large spatial and
temporal variability, primarily near shore.
5
Part of this is due to the inherent variability
of the coastal oceans. But even the historical
dataset has major flaws in coverage. For
example, the large flux regions at high and low
latitudes are almost completely unsampled.
And most pixels are undersampled. None have
measurements in over 25 of the months of record
lt10 have full 12 calendar-month over the
aggregate 25 years of the record. This is
because the CO2 measurements were primarily made
on vessels participating in open-ocean programs.
6
Can we alleviate this problem with algorithm
development?
Illustration of the co-dependence of XCO2 in the
central Oregon coastal ocean on temperature and
fluorescence. Cool temperatures indicate recently
upwelled, high-XCO2 water, while high
fluorescence indicates high photosynthetic
productivity and low XCO2. The bifurcation of
this relationship into two separate trends
indicates additional dependences.
7
Early Stages approach To date we have included
only the following ancillary data in the Coastal
CO2 database Lat, Lon,time, SST, SSS in an
equation of the form CO2_mod A BLat
CLon Dsin(J_Day/365 E) FSST
GSSS (This is essentially an MLR, but we are
using an algorithm that will allow non-linear
functionality) Complete dataset cannot be even
remotely approximated by a single function like
the above. But separating the data into regional
subsets holds some promise
8
Lat, SST
SST
Lat, Time
SST
9
Results show that different regions have
different dominant factors driving the
algorithms. But none of the results are very
pleasing. The algorithm in general underpredicts
the observed range of variability. Some regions
have clear subsets of data that have large
observed variability for a small range in
predicted CO2. Other regions, notably the
Pacific, are overall very poorly represented by
this algorithm. -Are higher-order dependences
in, er order? -Can subsets of the data give
better results? We examined the surface data
from the COAST project with an equation of the
form CO2_mod A BLat CLon
Dsin(J_Day/365 E) FSST GSSS HChl
Ibeam-C JLonSST Here, Lon is a good proxy
for cross-shelf distance. The LonT term
includes some strength of upwelling
characteristics-- e.g., cool water far from shore
represents strong upwelling. It also gives a
higher-order dependence on SST.
10
Coastal Oregon, upwelling season, 2001. Data
from cross-shelf transects between 44-45N.
COAST
SST, LonSST, are most important. Bioptical
factors make no improvement in algorithm
agreement with observations!
11

• Results summary
• Algorithms based on large composite database show
  some promise, but also some serious shortcomings.
  Are linear dependences enough?
• Algorithms for data subsets are encouraging, but
  seem to be insensitive to remotely-observible
  bio-optical parameters. Is this due to temporal
  decoupling between the persistence of bio-optical
  and CO2 signatures?
• Some obvious things to try are in progress
• Get historical remote-sensing bio-optical data
  included in algorithms for large data-base (data
  is merged, but not yet included in algorithm
  optimization).
• Add wind-speed and direction data to algorithms.
• Include distance-from-shore, and potentially
  water-column depth, as parameters in algorithms.
• Assess pCO2 as a function of the time derivative
  of SST and chl.

12

• Addressing the undersampling problem
• A small, low-cost, fast-response system for
  autonomous surface CO2 and bio-optical
  measurement has been developed and tested at OSU
  and will be deployed on the RV Wecoma when she
  returns from the shipyard in weeks.
• - pCO2 system is centered on a LI-COR 820 IR
  detector, with a  miniature Liqui-Cel membrane
  equilibrator and tangential flow filtration.
• - Optics include chlorophyll and CDOM
  fluorometry and, initially, beam attenuation.
• A high-precision traditional CO2 system has been
  built at PMEL and deployed on the Miller Freeman.

13
Instrument deployments are planned for NDBC buoys
46041 and 46050. Submersible bio-optical package
and PMEL surface PCO2 analyzer system are
prepared and ready to go on 46041 when it is
redeployed this Spring. Agreements with NDBC for
diver-deployment of bio-optical package on 46050
are being finalized now.
14
(No Transcript)
15
High-spatial resolution depth and
longitude-dependent observations of seawater
XCO2, at a) Cascade Head, and b) Cape Perpetua,
as in Figure 1, during August 2001. Subsurface
XCO2 has increased dramatically since the
sections were occupied in May, exceeding 1000 ppm
for much of the depth range sampled. Despite
this, photosynthetic productivity still occurs to
such an extent that the surface waters are
dominated by XCO2 values far below saturation
with respect to the atmosphere, and cross-shelf
averaged air-sea differences demand a large
area-specific uptake of atmospheric CO2.