Welcome to the

1
Welcome to the Ocean Color Bio-optical Algorithm
Mini Workshop Goals, Motivation, and
Guidance Janet W. Campbell University of New
Hampshire Durham, New Hampshire September 27, 2005
2
Goals
This workshop is aimed at evaluating ocean color
algorithms that produce chlorophyll retrievals.
It is expected that the algorithms tested may
also retrieve other constituents and / or related
inherent optical properties. Our goal is to
determine how new algorithms perform compared to
the operational empirical algorithms (OC4, OC3M)
and to evaluate whether improved accuracy is
achieved by accounting for other optically active
constituents.
3
Motivation

• NOMAD. We have a new data set to use in
  evaluating algorithms.

Jeremy Werdell will present overview of NOMAD.
4
Motivation

• NOMAD. We have a new data set to use in
  evaluating algorithms.
• The operational algorithms used for SeaWiFS (OC4)
  and MODIS (OC3M) are not mutually consistent.

5
MODIS is currently producing the SeaWiFS-analog
chlorophyll product. It employs the OC3M
algorithm parameterized with the same data set
used for the SeaWiFS OC4.v4 algorithm (n
2,804).
OC3M
Both are described in NASA TM 2000-206892, Vol.
11 (OReilly et al., 2000).
6
Approach Test algorithms with in situ data and
later with satellite match-ups. Four in situ
data sets of reflectance and chlorophyll data
shown here (n 1,119).
RMSE 0.293 (SeaBAM 0.184 AMT 0.256 W. Fla.
Shelf 0.175 Ches. 0.787)
7
(No Transcript)
8
The algorithms are different
The SeaWiFS algorithm (OC4.v4) is log10(CHL)
0.366 3.067R 1.930R2 0.649R3
1.532R4 where R log10max(Rrs(443),
Rrs(490), Rrs(510))/ Rrs(555)
The MODIS algorithm (OC3M) is log10(CHL)
0.283 2.753R 1.457R2 0.659R3
1.403R4 where R log10max(Rrs(443),
Rrs(488))/ Rrs(551)
9
There are systematic differences between the OC3M
and OC4 algorithms even when applied to the same
data set (assuming 448 490, 551 555)
The MODIS chlorophylls will be slightly less over
most of the ocean, i.e., where Chl lt 3 mg m-3.
The algorithms are not the same even when using
the 443550 ratio. Differences were intentional
to account for differences in spectral responses
of the MODIS and SeaWiFS bands and also fact that
488 ? 490 and 551 ? 555.
10
Motivation

• NOMAD. We have a new data set to use in
  evaluating algorithms.
• The operational algorithms used for SeaWiFS (OC4)
  and MODIS (OC3M) are not mutually consistent. It
  is desireable to have an algorithm that can be
  applied to different sensors to facilitate the
  generation of Climate Data Records (CDRs).

11
Motivation

• NOMAD. We have a new data set to use in
  evaluating algorithms.
• The operational algorithms used for SeaWiFS (OC4)
  and MODIS (OC3M) are not mutually consistent. It
  is desireable to have an algorithm that can be
  applied to different sensors to facilitate the
  generation of Climate Data Records (CDRs).
• A paper is about to appear in Geophysical
  Research Letters arguing for the importance of
  accounting for the effects of colored dissolved
  organic matter.

12
This paper, entitled Colored dissolved organic
matter and its influence on the satellite-based
characterization of the ocean biosphere by D.
Siegel, S. Maritorena, N. Nelson, M. Behrenfeld,
and C. McClain, is in press in Geophysical
Research Letters. The authors applied the GSM01
algorithm to global SeaWiFS data and compared the
derived chlorophyll distributions with OC4
chlorophyll maps. Differences are quite
significant and this paper will predictably
cause a stir. We should be prepared to respond
Stephane Maritorean will present overview of this
paper.
13
Guidance (Rules of Engagement)
Any algorithm approach may be considered. This
is not a workshop to look only at
semi-analytical algorithms. Algorithms will
be evaluated using a common data set (a subset of
NOMAD) and common performance criteria. Dont
present results for someone elses algorithm
unless youre sure it is implemented
correctly. We should regard the OC4 and OC3M as
the algorithms to beat. Compare your results
with these operational algorithms, not with other
non-operational ones.
14
If we can resolve other optical properties or
constituents (e.g., CDOM, POC) all the better
but our focus should remain on chlorophyll. Ideall
y, we should eliminate systematic differences
between SeaWiFS and MODIS chlorophyll
algorithms. The challenge is to explain and
reduce the errors in Chl with an algorithm that
can be implemented practically. Issues related
to its practical application include its speed,
sensitivity to errors in Lwn, and ability to
converge on a solution.
15
(No Transcript)
16

• Model-based Algorithms
• Forward model
• L(l ) f(C,Q(l))
  L(l) An ocean color spectrum
  (reflectance, radiance)
• f A semi-analytical forward model
• C Optically-active constituent vector
• Q(l) A model parameter vector related to
  IOP models
• Inverse model the retrieval of C
• C f -1(L(l ), Q(l) )
  
• f -1 An inverse of f ( an approach )

17

• Model-based Algorithms
• The forward model can actually be a simulation
  model (e.g., Hydrolight). Whatever it is, it
  should be tested with empirical data. How
  accurate should the forward model be?
• The inverse model is what we call the algorithm.
  We tend to think of semi-analytic algorithms
  inverted by linear or non-linear optimization
  techniques. But the inversion approach can be
  highly statistical (e.g., neural network). If it
  is trained with model-generated data, then this
  type of algorithm is also a model-based
  algorithm.

18

• Ocean Color Bio-optical Algorithm Mini-Workshop
  (OCBAM)
• Penobscot Room, New England Center
• University of New Hampshire
• Durham, New Hampshire
• Tuesday, September 27, 2005
• 900 Welcoming remarks
• 915 Overview talks
• Goals, motivation, and guidance Janet Campbell
• CDOM its influence on satellite chlorophyll
  Stephane Maritorena
• Discussion
• 1030 Break
• 1100 Framework
• Brief background on SeaBAM methods Jay OReilly
  
• Performance criteria for algorithms Janet
  Campbell
• The NOMAD dataset Jeremy Werdell
• 12 noon Lunch (NEC dining room)

19
(No Transcript)