Evaluation of ocean color atmospheric correction for

Report
Comparison of ocean color
atmospheric correction approaches for
operational remote sensing of turbid,
coastal waters
Jeremy Werdell
Bryan Franz
NASA Goddard Space Flight Center
13 Jun 2012
PJW, NASA, 13 Jun 2012, AtmCorr in
outline
remote sensing of turbid, coastal waters is difficult
no one uses the “black pixel assumption” anymore
most of the approaches to account for Rrs(NIR) > 0 sr-1 overlap
a bio-optical model for Rrs(NIR) provides one viable approach
comparing various approaches requires consistency
PJW, NASA, 13 Jun 2012, AtmCorr in
remote sensing of turbid, coastal waters is difficult
temporal & spatial variability
satellite sensor resolution
satellite repeat frequency
validity of ancillary data (SST, wind)
resolution requirements & binning options
straylight contamination (adjacency effects)
non-maritime aerosols (dust, pollution)
region-specific models required?
absorbing aerosols
suspended sediments & CDOM
complicates estimation of Rrs(NIR)
complicates BRDF (f/Q) corrections
saturation of observed radiances
AERONET
COVE
Chesapeake Bay
Program
anthropogenic emissions (NO2 absorption)
PJW, NASA, 13 Jun 2012, AtmCorr in
remote sensing of turbid, coastal waters is difficult
temporal & spatial variability
satellite sensor resolution
satellite repeat frequency
validity of ancillary data (SST, wind)
resolution requirements & binning options
straylight contamination (adjacency effects)
non-maritime aerosols (dust, pollution)
region-specific models required?
absorbing aerosols
suspended sediments & CDOM
complicates estimation of Rrs(NIR)
complicates BRDF (f/Q) corrections
saturation of observed radiances
AERONET
COVE
Chesapeake Bay
Program
anthropogenic emissions (NO2 absorption)
PJW, NASA, 13 Jun 2012, AtmCorr in
the experiment
Chesapeake Bay provides our case study site
run multiple long-term time-series of MODIS-Aqua
Lower Chesapeake Bay, June 2002 - December 2008
processing configuration follows Reprocessing 2010
QC metrics: exclude cloudy days & high sensor zenith angles
final analyses use ~ 13 days per month
generate frequency distributions and monthly time-series
use in situ measurements as reference
consider potential for application in an operational environment
PJW, NASA, 13 Jun 2012, AtmCorr in
atmospheric correction & the “black pixel” assumption
t() = w() + g() + f() + r() + a()
TOA
water
glint
foam
need a() to get w() and vice-versa
PJW, NASA, 13 Jun 2012, AtmCorr in
air
aerosols
atmospheric correction & the “black pixel” assumption
t() = w() + g() + f() + r() + a()
TOA
water
glint
foam
air
aerosols
need a() to get w() and vice-versa
the “black pixel” assumption (pre-2000):
a(NIR) = t(NIR) - g(NIR) - f(NIR) - r(NIR) - w(NIR)
PJW, NASA, 13 Jun 2012, AtmCorr in
0
atmospheric correction & the “black pixel” assumption
t() = w() + g() + f() + r() + a()
TOA
water
glint
foam
air
aerosols
need a() to get w() and vice-versa
the “black pixel” assumption (pre-2000):
0
a(NIR) = t(NIR) - g(NIR) - f(NIR) - r(NIR) - w(NIR)
calculate aerosol ratios,  :
(748,869) ≈
≈
(,869)
a(748)
a(869)
a()
(748,869)
a(869)
PJW, NASA, 13 Jun 2012, AtmCorr in
no one uses the “black pixel assumption” anymore
PJW, NASA, 13 Jun 2012, AtmCorr in
no one uses the “black pixel assumption” anymore
PJW, NASA, 13 Jun 2012, AtmCorr in
what happens if we don’t account for Rrs(NIR) > 0?
use the “black pixel” assumption (e.g., SeaWiFS 1997-2000)
PJW, NASA, 13 Jun 2012, AtmCorr in
approaches to account for Rrs(NIR) > 0 sr-1 overlap
many approaches exist, here are a few examples:
assign aerosols () and/or water contributions (Rrs(NIR))
e.g., Hu et al. 2000, Ruddick et al. 2000
use shortwave infrared bands
e.g., Wang & Shi 2007
correct/model the non-negligible Rrs(NIR)
Siegel et al. 2000
used in SeaWiFS Reprocessing 3 (2000)
Stumpf et al. 2003
used in SeaWiFS Reprocessing 4 (2002)
Lavender et al. 2005
MERIS
Bailey et al. 2010
used in SeaWiFS Reprocessing 2010
Wang et al. 2012
GOCI
use a coupled ocean-atmosphere optimization
e.g., Chomko & Gordon 2001, Stamnes et al. 2003, Kuchinke et al. 2009
PJW, NASA, 13 Jun 2012, AtmCorr in
fixed aerosol & water contributions (ex: MUMM)
assign  & w(NIR) (via fixed values, a climatology, nearby pixels)
PJW, NASA, 13 Jun 2012, AtmCorr in
advantages & disadvantages
advantages:
accurate configuration leads to accurate aerosol & Rrs(NIR) retrievals
several configuration options: fixed values, climatologies, nearby pixels
method available for all past, present, & future ocean color satellites
disadvantages:
no configuration is valid at all times for all water masses
requires local knowledge of changing aerosol & water properties
implementation can be complicated for operational processing
PJW, NASA, 13 Jun 2012, AtmCorr in
use of NIR + SWIR bands
use SWIR bands in “turbid” water, otherwise use NIR bands
PJW, NASA, 13 Jun 2012, AtmCorr in
use of SWIR bands only
compare NIR & SWIR retrievals when considering only “turbid pixels”
PJW, NASA, 13 Jun 2012, AtmCorr in
advantages & disadvantages
advantages:
“black pixel” assumption largely satisfied in SWIR region of spectrum
straightforward implementation for operational processing
disadvantages:
only available for instruments with SWIR bands
SWIR bands on MODIS have inadequate signal-to-noise (SNR) ratios
difficult to vicariously calibrate the SWIR bands on MODIS
must define conditions for switching from NIR to SWIR
PJW, NASA, 13 Jun 2012, AtmCorr in
bio-optical model to estimate Rrs(NIR)
estimate Rrs(NIR) using a bio-optical model
operational SeaWiFS & MODIS processing ~ 2000-present
PJW, NASA, 13 Jun 2012, AtmCorr in
advantages & disadvantages
advantages:
method available for all past, present, & future ocean color missions
straightforward implementation for operational processing
disadvantages:
bio-optical model not valid at all times for all water masses
PJW, NASA, 13 Jun 2012, AtmCorr in
summary of the three approaches
defaults as implemented in SeaDAS
PJW, NASA, 13 Jun 2012, AtmCorr in
approaches to account for Rrs(NIR) > 0 sr-1 overlap
t() = w() + g() + f() + r() + a()
TOA
water
glint
foam
air
coupled ocean-atm
Chomko & Gordon 2001
Stamnes et al. 2003
Kuchinke et al. 2009
aerosols
SWIR
Wang et al. 2007
a(NIR) = t(NIR) - g(NIR) - f(NIR) - r(NIR) - w(NIR)
(748,869) ≈
≈
(,869)
a(748)
a(869)
assign  and/or Rrs(NIR)
Hu et al. 2000
Ruddick et al. 2000
a()
a(869)
PJW, NASA, 13 Jun 2012, AtmCorr in
model Rrs(NIR)
Siegel et al. 2000
Stumpf et al. 2003
Lavendar et al.
2005
Bailey et al. 2010
Wang et al. 2012
bio-optical model to estimate Rrs(NIR)
initial Rrs(670) measured by satellite (using Rrs(765) = 0)
PJW, NASA, 13 Jun 2012, AtmCorr in
bio-optical model to estimate Rrs(NIR)
initial Rrs(670) measured by satellite (using Rrs(765) = 0)
model a(670) = aw(670) + apg(670)
= 0.1 m-1
aw(670) = 0.44 m-1
PJW, NASA, 13 Jun 2012, AtmCorr in
bio-optical model to estimate Rrs(NIR)
initial Rrs(670) measured by satellite (using Rrs(765) = 0)
model a(670) = aw(670) + apg(670)
estimate bb(670) using Rrs(670), a(670), & G(670) [Morel et al. 2002]
Rrs (670)  G(670)
bb (670)
a(670)  bb (670)

PJW, NASA, 13 Jun 2012, AtmCorr in
bio-optical model to estimate Rrs(NIR)
initial Rrs(670) measured by satellite (using Rrs(765) = 0)
model a(670) = aw(670) + apg(670)
estimate bb(670) using Rrs(670), a(670), & G(670) [Morel et al. 2002]
model h using Rrs(443) & Rrs(555) [Lee et al. 2002]


Rrs (443) 
h  2.0 11.2 exp0.9

Rrs (555) 



from Carder et al. 1999
PJW, NASA, 13 Jun 2012, AtmCorr in
bio-optical model to estimate Rrs(NIR)
initial Rrs(670) measured by satellite (using Rrs(765) = 0)
model a(670) = aw(670) + apg(670)
estimate bb(670) using Rrs(670), a(670), & G(670) [Morel et al. 2002]
model h using Rrs(443) & Rrs(555) [Lee et al. 2002]
estimate bb(765) using bb(670) & h
670 h
bb (765)  bbw (765)  bbp (670)  
765 

PJW, NASA, 13 Jun 2012, AtmCorr in
bio-optical model to estimate Rrs(NIR)
initial Rrs(670) measured by satellite (using Rrs(765) = 0)
model a(670) = aw(670) + apg(670)
estimate bb(670) using Rrs(670), a(670), & G(670) [Morel et al. 2002]
model h using Rrs(443) & Rrs(555) [Lee et al. 2002]
estimate bb(765) using bb(670) & h
reconstruct Rrs(765) using bb(765), aw(765), & G(765)
Rrs (765)  G(765)
bb (765)
a w (765)  bb (765)
aw(765) = 2.85 m-1

PJW, NASA, 13 Jun 2012, AtmCorr in
bio-optical model to estimate Rrs(NIR)
initial Rrs(670) measured by satellite (using Rrs(765) = 0)
model a(670) = aw(670) + apg(670)
estimate bb(670) using Rrs(670), a(670), & G(670) [Morel et al. 2002]
model h using Rrs(443) & Rrs(555) [Lee et al. 2002]
estimate bb(765) using bb(670) & h
reconstruct Rrs(765) using bb(765), aw(765), & G(765)
iterate until Rrs(765) changes by <2% (typically 3-4 iterations)
PJW, NASA, 13 Jun 2012, AtmCorr in
bio-optical model to estimate Rrs(NIR)
black = land; grey = Chl < 0.3 mg m-3; white Chl > 0.3 mg m-3
not applied when Chl < 0.3 mg m-3
weighted application when 0.3 < Chl < 0.7 mg m-3
fully applied when Chl > 0.7 mg m-3
PJW, NASA, 13 Jun 2012, AtmCorr in
bio-optical model to estimate Rrs(NIR)
approaches used previously by the NASA OBPG:
Bailey et al. 2010, Optics Express 18, 7521-7527
Stumpf et al. 2003, SeaWiFS Postlaunch Tech Memo Vol. 22, Chapter
9
Siegel et al. 2000, Applied Optics 39, 3582-3591
others
PJW, NASA, 13 Jun 2012, AtmCorr in
comparing approaches requires consistency
comparison of approaches benefits from consolidation of software
permits isolation of mechanisms & algorithms to evaluate
limits interference by & biases of other factors (e.g., look up tables)
for example
Lavendar et al. 2005, Bailey et al. 2010, & Wang et al. 2012 all present
bio-optical models for estimating Rrs(NIR)
inclusion of all 3 into L2GEN permits isolated comparison of bio-optical
model while controlling Rayleigh tables, aerosol tables, etc.
uncertainties
PJW, NASA, 13 Jun 2012, AtmCorr in
comparisons with MERIS CoastColour
SeaWiFS
MODIS-Aqua
MERIS
in situ
Middle Bay
2005-2007
Rrs() 412-670
PJW, NASA, 13 Jun 2012, AtmCorr in
comparisons with MERIS CoastColour
SeaWiFS
MODIS-Aqua
MERIS
in situ
Middle Bay
2005-2007
derived products
Chl, IOPs, Kd,
TSM
PJW, NASA, 13 Jun 2012, AtmCorr in
PJW, NASA, 13 Jun 2012, AtmCorr in
Turbid Water Atmospheric
Correction: w(NIR) ≠ 0
guess
w(670) = 0
model
1) convert w(670) to bb/(a+bb)
via Morel f/Q and retrieved Chla
w(NIR) = func w(670)
2) estimate a(670) = aw(670) + apg(670)
via NOMAD empirical relationship
Correct
'a(NIR) = a(NIR) – t w(NIR)
a670  e
lnCa 0.93893.7589
retrieve
3) estimate bb(NIR) = bb(670) (/670)h
via Lee 2010
iw(670)

test
|wi+1 (670) - i(670)|
< 2%
done
  a 670
w

h  2.0 * 1. - 1.2 * e-0.9*R
rs
 443 R rs 555
4) assume a(NIR) = aw(NIR)
no

5) estimate w(NIR) from bb/(a+bb)
via Morel f/Q and retrieved Chla

SNR transect for MODIS-Aqua NIR & SWIR bands
PJW, NASA, 13 Jun 2012, AtmCorr in
Aqua Chl “match-ups” for NIR & SWIR processing
PJW, NASA, 13 Jun 2012, AtmCorr in
MODIS-Aqua a(443)
PJW, NASA, 13 Jun 2012, AtmCorr in
distribution of the turbidity index using in NIR-SWIR
PJW, NASA, 13 Jun 2012, AtmCorr in
MODIS-Aqua vs. SeaWiFS
default processing ~ OC3 for MODIS-Aqua & OC4 for SeaWiFS
PJW, NASA, 13 Jun 2012, AtmCorr in
atmospheric correction & the “black pixel” assumption
ocean color satellites view the top of the atmosphere
this signal includes contributions from:
model
Rayleigh (air molecules)
surface reflection
model
aerosols
water
0
to remove the aerosol signal, we make some assumptions
about the “blackness” of the water signal in near-infrared (NIR)
bands
PJW, NASA, 13 Jun 2012, AtmCorr in

similar documents