the role of optical water type classification in the context of giop timothy s. moore university of...

The role of Optical Water Type classification in the context of

GIOP

Timothy S. Moore University of New Hampshire, Durham NH

Mark D. DowellJoint Research Centre, Ispra Italy

September 25, 2010

Rationale

• There is necessity to describe a considerable amount of variability in Inherent Optical Property (IOP) subcomponent models.

• This is particularly true, if inversion algorithms are to be applicable at global scale yet remain quantitatively accurate in coastal & shelf seas.

• This is unlikely to be achieved in the foreseeable future, with a single representation of IOP subcomponents.– BEAM – Case2R, GIOP

• The proposed approach is an algorithm framework more than a specific algorithm.

Practical uses of a classification approach based on Optical Water Types (OWT)

• Describe variance and co-variance of optically active constituents

• Parameterizing IOP subcomponent models (or fit coefficients for empirical relationships)

• Selecting different inversions methods for different optical waters

• Avenue to spatial uncertainty estimates for remotely-sensed products

• Value-added products (directing new Cal/Val field work, data collection)

Advantages of fuzzy logic defined provinces

• They allow for spatial and temporal dynamics both seasonal and inter-annual in the optical properties of a given region.

• They address the issue of transitions at the boundaries of provinces (through the fuzzy membership function of each class) thus resulting finally in the seamless reconstruction of a single geophysical product.

Our OWT method uses a fuzzy logic approach for optical classification of in situ and satellite data based on remote sensing reflectance.

In-situ Database(NOMAD)Rrs()

IOPsSgd, aph*,…….

Station data sorted by class

Class based relationships

8 classes

Class Mi, Σi

Satellite Measurements

Individual classderived products

Merged Product

Calculatemembership

Rrs()

Conceptual Framework for class-based algorithms

Cluster Analysis

IOP model parameterization

IOP model/algorithm selection

• 2407 data points (NOMAD v2)• 8 clusters ‘optimal’• representations of different optical water types (OWT)• mean and covariance matrix form the basis of the fuzzy membership function.

Base OWT Definition

OWT 1 OWT 2 OWT 3 OWT 4

OWT 5 OWT 6 OWT 7 OWT 8

Mapping of the OWTs in ocean color data - example

a() = aw() + Ac()[Chl]Bc() + [acdm(440)] exp(-Sdg(-440))

Possible coefficients to parameterize on an OWT-basis ina standard semi-analytic algorithm configuration.

bb() = bbw() +[bbp(555)] [555/]Y

Red - variables Yellow - parameters that need to be set (possible OWT dependency

a() = aw() + aph(Chl) + ad(TSS) + acdom(CDOM)

bb() = bbw() + bbp(Chl,TSS)

Class–based GIOP Class–based QAA

• Sgd, Sg, Sd

• aph*()

• slope of bbp

• Sgd variable based on class• at(443) versus rrs(443)/rrs(555) class based• at(555) versus at(443) class based• aph(443) versus Chl class based aph*(443)

One could imagine applying a tuning algorithm (e.g. simulatedannealing) to each class to determine optimal

class based model coefficients.

What follows is a look at the distribution and relationships of optical properties in the context of semi (quasi)-analytic algorithms from an OWT perspective based on the NOMAD v2 and IOCCG simulated data set.

OWT NOMAD*

w/ IOPs

IOCCG

Sim.

Global Avg.

1 3 7 31

2 8 8 31

3 18 10 21

4 20 4 9

5 19 10 4

6 22 40 2

7 9 20 1

8 2 1 <1

Distribution of OWTs in Data sets vs. Ocean Obsverations (numbers are in percent)

Sg v. ag443

OWT Sg

1 0.016

2 0.016

3 0.017

4 0.015

5 0.015

6 0.016

7 0.016

8 0.017

Avg. 0.016

Points are color coded by degree of membership to the OWT (based on Rrs).

IOCCG

OWT 1

NOMAD

ag slope

ag 443

ag440 -NOMAD

Sg

(Bricaud et al, 2009)

aph*

log10 Chl

aph

OWT 1

OWT 2

OWT 4 OWT 8

OWT 7

OWT 6

OWT 5

OWT 3

OWT12345678

µ=7.87

µ=7.39

µ=3.22

µ=3.04

µ=0.148

µ=0.086

µ=0.331

µ=1.01

aph

IOCCG

NOMAD

log ag443

OWT 1

log aph443 ag443/at443 bbp slope

log ag443 bbp slopelog aph443

OWT 2

IOCCG

NOMAD

ag443/at443

OWT 3

log ag443 bbp slopelog aph443

IOCCG

ag443/at443

NOMAD

OWT 4

log ag443 bbp slopelog aph443

IOCCG

ag443/at443

NOMAD

OWT 5

log ag443 bbp slopelog aph443

IOCCG

ag443/at443

NOMAD

OWT 6

log ag443 bbp slopelog aph443

IOCCG

ag443/at443

NOMAD

OWT 7

log ag443 bbp slopelog aph443

IOCCG

ag443/at443

NOMAD

For what its worth…

Sg bbp y aph 443 ag 443

OWT N I N I N I N I

1 0.016 0.0148 0.85 2.82 0.007 0.007 0.023 0.006

2 0.016 0.0146 1.74 2.62 0.013 0.010 0.026 0.012

3 0.017 0.0146 1.38 2.33 0.021 0.016 0.025 0.026

4 0.015 0.0151 1.03 1.99 0.048 0.029 0.056 0.052

5 0.015 0.0139 0.87 0.88 0.246 0.128 0.243 0.477

6 0.016 0.0148 0.88 0.62 0.277 0.154 0.289 0.483

7 0.016 0.0159 0.99 0.48 0.116 0.210 0.197 0.491

8 0.017 0.0154 - ~0 0.132 0.314 0.187 0.252

Avg 0.016 0.0149 1.03 1.21 0.135 0.118 0.149 0.329

Averages

Chl

a ph*

Bricaud aph* function

Miscellaneous bio-optical empirical functions

OWT12345678

rrs443/rrs555

Y

(QAA)

1 50.5

2.5

2.0

1.5

1.0

0.5

0.0

OWT12345678

LAS Kd functionK

d443

rrs443/rrs555

rho

QAA

a555

0

50

75

25

“Blue Hole”

Frequency of ‘low membership’ areas

100 %

Summary

• There are some inconsistencies in the OWT-based distributions of IOPs between NOMAD and the IOCCG simulated data set.

• Both data sets are skewed towards coastal/case 2 waters.

• If a new simulated data set is being considered, the generation of IOPs and IOP pairs could be further constrained by the variance and co-variance as seen in NOMAD within different OWTs.

• In addition, the representation of data points could be guided by the global distribution of naturally occurring OWTs.

Summary (continued)

• OWT code is currently in Seadas, but has yet to receive the final green light for public usage (we see no problem here).

• Preliminary OWT-based IOP parameters now exist and can be used in the GIOP framework.

• Potential for further use in parameterizing empirical models within GIOP is being explored.

• OWTs themselves may change over time, which could effect some of the OWT-based parameters (don’t think this to be major).

• Sensitivity and performance analysis remains to be assessed for GIOP-related products.

log ag443

OWT 1

OWT 2

OWT 3

OWT 4

OWT 5

OWT 6

OWT 7

OWT 8 NOMAD ag443 OWTdistributions

ag41

1

OWT12345678

ag440 -NOMAD

Sg

(Bricaud et al, 2009)

Sd v. ad443

OWT Sd

1 0.011

2 0.011

3 0.010

4 0.010

5 0.011

6 0.011

7 0.012

8 0.011

Avg. 0.011

Sdg v. adg443

OWT Sdg

1 0.014

2 0.014

3 0.015

4 0.013

5 0.013

6 0.013

7 0.014

8 0.014

Avg. 0.014

OWT 5

aph*aph

• There are some issues with data quality that might be revealed.

Effects of aph to aph* conversion

OWT 1

OWT 4

OWT 3

OWT 2

OWT 7

OWT 6

OWT 5b bp

OWT Y

1 0.85

2 1.74

3 1.38

4 1.03

5 0.87

6 0.88

7 0.99

8 -

Avg 1.03

bbp slope estimation

Y0 0.5 1 1.5 2.0

Y

OWT Y

1 0.85

2 1.74

3 1.38

4 1.03

5 0.87

6 0.88

7 0.99

8 -

Avg.* 1.03

bbp slope estimation

* Negative values excluded

OWT 1

OWT 4

OWT 3

OWT 2

OWT 5

OWT 6

OWT 7

All