Abstract

1
Match-ups Chlor_a_2 and Chlor_a_3 using in
situ and MODIS Terra data By F.R. Chen1,
K. L. Carder1, J. Patch1, Jeremy Werdell2, and
Sean Bailey2 1 College of Marine Science,
University of South Florida, St. Petersburg,
Florida USA 2 SeaBASS project, GSFC, NASA.
Abstract The match-up data comparisons with
field data published on the SeaBASS web site
http//seabass.gsfc.nasa.gov show accuracies for
MODIS Terra radiances and chlor_a_2 chlorophyll-a
retrievals. The data include in situ and MODIS
Terra radiances and sea-surface temperature
values, allowing calculation of both chlor_a_2
and chlor_a_3 MODIS data products. The purpose of
this presentation is to conveniently show
side-by-side comparisons of the two chlorophyll-a
products with synoptic field data.
Chlorophyll-a retrievals using in situ radiances
compared to in situ chlorophyll-a concentrations
are accurate within 55 (linear RMS difference)
for chlor_a_2 and within 25 (linear RMS
difference) for chlor_a_3. This establishes the
inherent algorithm accuracy if all field
measurements were perfect. Chlorophyll-a
retrievals using MODIS radiances versus in situ
chlorophyll a values were only within 62 and 53
for chlor_a_2 and chlor_a_3, respectively. While
chlor_a_3 is significantly more accurate than is
chlor_a_2 when accurate radiances are used, the
inherently more accurate bands (443, 488, 551 nm)
of MODIS Terra used for chlor_a_2 retrievals
compensate for the algorithmic advantage of
chlor_a_3. Improvements in sensor calibration and
atmospheric correction are the primary
improvements needed to permit chlorophyll a
concentrations from space to reach the
chlorophyll accuracy goal of 35. It can be
reached with chor_a_3 if accuracies of
water-leaving radiance for all bands are less
than 15, but no amount of improvement of
radiance accuracies will allow chlor_a_2 values
to reach the goal of 35 accuracy for chlorophyll
a.
Results Figures 3 and 4 show chlorophyll-a
retrievals using in situ radiances with the
chlor_a_2 and chlor_a_3 algorithms, respectively,
compared to in situ chlorophyll-a values. Note
that exclusion of an apparently erroneous data
point significantly improves accuracies for both
figures. Moving the decimal point of the field
chlorophyll value by one digit places the
erroneous point appropriately amongst a cluster
of other points on each figure. Chlorophyll-a
retrievals via chlor_a_2 and chlor_a_3 algorithms
using MODIS Terra radiances are compared to field
validation chlorophyll a values(Fig. 5 and 6).
Again, one apparently erroneous data point is
removed to improve accuracies for both
comparisons.

• Discussion
•  
• An algorithm can only be as accurate as is
  permitted by the radiance data available. Given
  consistently measured in situ radiance data,
  retrievals from chor_a_2 and chlor_a_3 algorithms
  are compared (Figs. 3, 4). Chlor_a_2 has a linear
  RMS error of 54.6 due in part to inaccurate
  slope (0.84), bias (-0.11), and scatter (r2
  86.5) factors for this log-log plot. Chlor_a_3,
  however, has a much lower linear RMS error of
  25.3, due in part to more accurate slope (1.01)
  and bias (0.0017) terms, and less scatter (r2
  95.6) factors for this log-log plot. The 35
  linear accuracy goal for chlorophyll a of the
  ocean-color community (e.g. McClain et al. 1998
  Kilpatrick et al. 2002) can be met with chlor_a_3
  given accurate water-leaving radiance retrievals.
  The 5 accuracy goal for satellite ocean-color
  radiances (McClain et al. 1998 Gordon and Voss
  ATBD 1999) is probably the best that can be
  achieved for SeaWiFS even with 3x3 pixel
  smoothing due to digitization errors (Hu et al.
  2001). Digitization noise will not limit the
  accuracy of MODIS in the same way because it uses
  12-bit rather than 10-bit digitization, but
  calibration and atmospheric correction accuracies
  can be limiting. Even with accurate field
  radiances, however, chlor_a_2 does not achieve
  the 35 goal using the global database and
  algorithm parameterization presently available.
• The radiances from Terra MODIS have been
  compared to ship-borne radiance measurements as
  shown at the SeaBASS web site, with the highest
  RMS errors found for the shorter blue wavebands
  (e.g. 35.19, 25.49, 22.24, and 15.44 for 412,
  443, 488, and 551 nm, respectively). Rather than
  repeating that exercise here, we have evaluated
  the effects of the scatter inherent in these data
  on chlorophyll a retrievals. Since only chlor_a_3
  uses the 412 nm MODIS channel, and it has the
  highest RMS difference relative to field
  radiances, one would expect this inaccuracy of
  MODIS radiances to more deleteriously affect
  chlor_a_3 than chlor_a_2. Figures 5 and 6 and
  Table 1 illustrate this effect, as linear RMS
  error for chlor_a_3 has doubled from 25.3 to
  48.8, while RMS error for chlor_a_2 has only
  increased from 54.6 to 57.2. Note that Fig. 5
  has less bias than Fig. 3 with more scatter,
  accounting for the similar error values.
• Using the error model (1) and RMSs of radiance,
  the 57.2 and 48.8 errors for chlor_a_2 and
  chlor_a_3 retrievals can be simulated for MODIS
  Terra data, where 0.25492 0.15442 0.54620.5
  62.2 and 0.35192 0.25492 0.15442
  0.25320.5 52.6 (Table 1), within 5 and 4,
  respectively. If all of the MODIS radiances were
  only 15 in error, then the chlor_a_3 retrievals
  are expected to be accurate within about 36. If
  all radiances were only 10 in error, then
  chlor_a_3 retrievals would have about 30 error.
  On the other hand, if MODIS Terra radiance
  retrievals were perfect (e.g. matched field
  radiances), chlor_a_2 retrievals would not be
  more accurate than 55. Since the goal of the
  ocean color community has been to provide MODIS
  and SeaWiFS radiance accuracies of about 5
  except perhaps for the 412 nm band (e.g. Gordon
  et al., ATBD 1999), the objective of providing a
  chlor_a_3 algorithm capable of delivering
  accuracies better than 35 appears to be
  achievable using chlor_a_3 with satellite
  retrievals of water-leaving radiance of better
  than 15. However, no amount of improvement of
  radiance accuracies will allow chlor_a_2 values
  to reach the goal of 35 accuracy for chlorophyll
  a.

Methods Estimates of chor_a_2 and chlor_a_3
were made using both in situ and MODIS Terra
radiances (Collection 041 ocean processing) along
with MODIS sea-surface temperatures for the in
situ, match-up sites. The latitudes and
longitudes for the match-up stations used at the
SeaBASS web site (Fig. 1) were used for
extraction of MODIS SST values required by the
chor_a_3 algorithm. Chlor_a_2 retrievals of
chlorophyll a were determined using the OC-3
empirical algorithm of OReilly et al. (1998)
along with remote-sensing reflectance or
normalized water-leaving radiance data. Chlor_a_3
values were determined using the semi-analytic
method of Carder et al. (1999) along with an
updated nitrate-depletion-temperature (NDT)
method of adjusting phytoplankton absorption for
the pigment-package effect (Carder et al. 2004
Advances in Space Research). NDT values used in
the chlor_a_3 algorithm for the globe are shown
in Figure 2 in degrees Celsius (after Kamykowski
et al. 2002). SST values are compared to NDT
values to determine the appropriate
chlorophyll-specific absorption coefficient.
Modeling the error fields as the sum of the
squares of the algorithm accuracies and each
MODIS radiance accuracy for the channels used
provides the following This simple
model basically assumes that the various sources
of error are random and sum together based upon
the inherent algorithm error and the errors for
each of the bands used in the algorithm. We can
use this model to simulate the retrieval accuracy
expected if each of the MODIS bands were more
accurate. Errors are shown in Table 1.
Applications are shown in Discussion.
Fig. 1 SeaBASS validation data points (23 Jan.,
2004).
Fig. 2 Nitrate-depletion temperature map based
upon Kamykowski et al. (2002).
Fig. 3 Chlor_a_2 retrieval accuracy for in situ
radiances.
Fig. 4 Chlor_a_3 retrieval accuracy for in situ
radiances.
References Carder, K.L., F.R. Chen, Z.P. Lee,
S.K. Hawes, and D. Kamykowski, 1999. Semianalytic
Moderate-Resolution Imaging Spectrometer
algorithms for chlorophyll-a and absorption with
bio-optical domains based on nitrate-depletion
temperatures, JGR 104, 5403-5421. Carder, K.L.,
F.R. Chen, J.P. Cannizzaro, J.W. Campbell, and
B.G. Mtichell, 2004. Performance of the MODIS
semi-analytical ocean color algorithm for
chlorophyll-a, Advances in Space Research, in
press, http//www.sciencedirect.com/. Hu, C.,
K.L. Carder and F.E. Muller Karger, 2001. How
precise are SeaWiFS ocean color estimates?
Implications of digitization-noise errors, Remote
Seansing of Environment 76 (2), 239-240. Gordon,
H., and K. Voss, 1999. MODIS Normalized
Water-leaving Radiances ATBD vol. 4,
http//modis-ocean.gsfc.nasa.gov/qa/. Kamykowski,
D., S. Zentara, J. M. Morrison, and A. C.
Switzer, 2002. Dynamic global patterns of
nitrate, phosphate, silicate, and iron
availability and phytoplankton community
composition from remote sensing data, Global
Biogeochemical Cycles, vol. 16, No. 4,
1077. Kilpatrick K., E. Kearns, K. Voss, R.
Evans, W. Esaias, and V. Salomonson, 2001. Status
of the MODIS Ocean Color Calibration,
http//modis-ocean.gsfc.nasa.gov/qa/. McClain,
C.R., M.L. Cleave, G.C. Feldman, W.W. Gregg, S.B.
Hooker, and N. Kuring, 1998. Science quality
SeaWiFS data for global biosphere research, Sea
Technology 39, 10-16. OReilly, J.E., S.
Maritorena, B. G. Mitchell, D.A. Siegal, K.L.
Carder, S.A. Garver, M. Kahru, and C.R. McClain,
1998. Ocean color algorithms for SeaWiFS, J.G.R
103, 24,937-24,953.
Table 1. MODIS Terra errors for normalized
water-leaving radiance and the inherent (field)
errors for chlorophyll-a algorithms chlor_a_2 and
Chlor_a_3. Algorithm and total error values are
linearized from figures as RMSlinear
0.5(10RMS1 - 1) (1 10 -RMS1). Band nLw
Algorithm Model Total
Model Model ? Error Error
Error Error Error 15 Error 10 412
0.3519 10.253 30.526 50.488
70.36 70.307 443 0.2549
20.546 40.622 60.572 488
0.2218 551 0.1544 1. In situ chlor_a_3
4. chlor_a_2 RMS error model 2. In situ
chlor_a_2 5. chlor_a_3 measured error
(MODIS) 3. chlor_a_3 RMS error model 6.
chlor_a_2 measured error (MODIS)
7. Chlor_a_3 performance with more accurate
nLws
Fig. 5 Chlor_a_2 retrieval accuracy for MODIS
Terra radiances.
Fig. 6 Chlor_a_3 retrieval accuracy for MODIS
Terra radiances.