An Algorithm for Coastal Water and the Status of its Implementation into the MODIS Processing

Stream

by

Howard.R. Gordon and Roman. M Chomko, Department of Physics University of Miami

Coral Gables, FL 33124

R.E. Evans, J.W. Brown, S. Walsh and W. Ba ringer RSMAS

University of Miami Miami, FL 33146

(Gordon and Morel, 1983)

Atmospheric Correction

)()()()()()( λρλλλρλρλρ wsvArt tt++=

Case 1 waters : • ρw(765) ≈ ρw(865) ≈ 0, ⇒ NIR can be used to assess the aerosol

influence . Case 2 waters : • ρw(NIR) ≠ 0, ⇒ no bands "tailor made" for assessing the aerosol. • Case 2 waters contain large quan tities of dissolved organic material that

influence ρt in a manner similar to strongly -absorbing aerosols. • Strongly absorbing aerosols are often found near the coast. Approach for Case 2 waters: model ρA(λ) and ρw(λ), and then use spectral optimization to find the best values of the model parameters.

The Aerosol Model

Uses a Junge Power -Law Size Distribution :

0=dD

dN , D < D0,

11

+= νD

K

dD

dN

, D0 ≤ D ≤ D1,

1+= νD

K

dD

dN

, D1 ≤ D ≤ D2,

0=dD

dN

, D > D2,

D0 = 0.06 µm, D1 = 0.20 µm, and D2 = 20 µm.

Mie theory is used to compute aerosol properties

• m = m r − imi, where mr is either 1.50 or 1.333, and mi = 0, 0.001, 0.003, 0.010,

0.030, and 0.040. • ν ranges from 2.0 to 4.5 in steps of 0.5. • 72 separate aerosol models (2 values of mr × 6 values of m i × 6 values of ν).

)(),,,,()(),,,,(

)(),,,,()(),,,,(),,,,(

43

2

λτνλλτνλ

λτνλλτνλνλρ

irir

iririrA

mmGdmmGc

mmGbmmGammG

++

+=

• Interpolate to essentially give a continuum of models.

The Water Model

(Garver and Siegel, 1997)

ρw = ρw(bb/(a+bb))

a = aw + aph + acdm bb = bbw + bbp

aph(λ) = aph0(λ) C acdm(λ) = acdm(443) exp[-S(λ−443)}] bbp(λ) = bb(443) [443/λ]n

ρw = ρw(λ,C,acdm(443),bbp(443))

Note, the parameters aph0(λ), S, and n are provided by fitting the model to experimental data. For Case 1 waters, S = 0.0206 nm-1 and n = 1.03 (Maritorena, et al., 2002).

The Optimization

)()()()()()( λρλλλρλρλρ wstvtArt ++=

,

)(),(),(),(),,( λρλλλρλρ wsvAAw GtGtGmeasuredG +≡

.

The modeled counterpart of is

)).443(),443(,,(ˆ),,,,,(ˆ),,,,,(ˆ

),,,,,(ˆ))443(),443(,,,,,,,(ˆ

bpcdmwairsairv

airAbpcdmairAw

baCmmGtmmGt

mmGbaCmmG

λρτνλτνλ

τνλρτνλρ

+

≡

Assuming ρA(765) and ρA(865) = 0 gives estimation of the parameters ν and τa as functions of mr and m i, i.e., ν( mr,m i) and τa( mr,mi). Given the constraints ν( mr,mi) and τa( mr,m i) we minimize the quantity

{ }∑ −i

measuredGbaCmmG iAwbpcdmairiAwλ

λρτνλρ 2),,())443(),443(,,,,,,,(ˆ

In effect, we have optimized for 7 parameters:

C, acdm(443), bbp (443), ν, τa, mr, and mi;

This is generally all that is needed in Case 1 waters.

To validate this algorithm, we use the SeaWiFS image from Day 279 (left on previous slide) and compare the retrievals of acdm from the algorithm with estimates of aCDOM from the AOL. The AOL measurements are made along the triangular path drawn on the nex t two images.

SOA acdm(443) (m-1)

Note Tracks

SOA acdm(443) (m-1)

North-SouthTrack

0.000

0.010

0.020

0.030

0.040

0.050

36.0 36.5 37.0 37.5 38.0 38.5

Latitude

acd

m(4

43)

(m

-1)

AOL (S = 0.0206/nm)

SOA (S = 0.0206/nm)

Comparison of SOA and AOL acdm(443)along the North-South Track

The value of S required to bring the SOA retrieved acdm(443) into confluence with the AOL-retrieved aCDOM(443) at each point along the track the track was determined and shown in the next slide. The resulting S values show a clear trend of decreasing into the mesotrophic waters as would be expected (Green and Blough, 1994). Similar results are found for the other two tracks.

Required "S" for Exact AOL-SOA Agreement

Along North-South Track

0.010

0.015

0.020

0.025

36.0 36.5 37.0 37.5 38.0 38.5

Latitude (Degrees)

S (

nm

-1)

SOA Chl a (mg/m3)

1.50

1.00

0.50

0.10

SeaWiFS 8-day mean Chl a

Note Tracks

Comparison with SeaWiFS

Comparison with SeaWiFS

bbp(443) (m-1)

0.001

0.010

0.030

0.003

ν4.5

4.0

3.5

3.0

2.5

2.0

ω0

1.0

0.9

0.8

0.7

0.6

Extension to Case 2 Waters

• In Case 2 waters, we operate the algorithm as in Case 1 waters, i.e., assuming that ρw(NIR) = 0.

• Then we use the retrieved values of C, acdm(443), and

bbp(443) to provide an estimate of ρw in the NIR, and the retrieved value s of ν, τa, mr, and mi to estimate tv and ts and the NIR.

• These estimates are subtracted from the total, i.e.,

)()()()()()( NIRANIRrNIRwNIRsNIRvNIRt tt λρλρλρλλλρ +=−

.

• The ν −τa, portion of the algorithm is then operated with

)()()()( NIRwNIRsNIRvNIRt tt λρλλλρ −

,

instead of ρ t(λNIR), to estimate new constraints ν( mr,mi) and τa( mr,m i), and to initiate a new optimization , etc.

Incorporation into the MODIS Code : A Status Report

Processing philosophy

• Spectral Optimization Algorithm is slow, so at present we must restrict application to sub -granuals.

• Unlike the Case 1 ρw(λ) model, the Case 2 ρw(λ) model

will most likely be site specific, i.e., the parameters in the GS97 model {aph0(λ), S, and n} will depend on the target location.

• Our goal is to provide processing code that can be used

for any location, given model parameters for that location. Individual investigators must su pply aph0(λ), S, and n.

SeaWiFS

bbp (m-1)

MODISSeaWiFS

bbp (m-1)

MODISSeaWiFS

bbp (m-1)

acdm (m-1)

SeaWiFS 0.003

MODISSeaWiFS

acdm (m-1)

MODISSeaWiFS

acdm (m-1)

Chl (mg m-3)

SeaWiFS

MODISSeaWiFS

Chl (mg m-3)

MODISSeaWiFS

Chl (mg m-3)

La commedia è finita