Remote Sensing of Environment 136 (2013) 218–224
Contents lists available at SciVerse ScienceDirect
Remote Sensing of Environment
journal homepage: www.elsevier.com/locate/rse
Comparison of chlorophyll in the Red Sea derived from MODIS-Aqua
and in vivo fluorescence
Robert J.W. Brewin a, b,⁎, Dionysios E. Raitsos a, c, Yaswant Pradhan d, Ibrahim Hoteit c
a
Plymouth Marine Laboratory (PML), Prospect Place, The Hoe, Plymouth PL1 3DH, UK
National Centre for Earth Observation, PML, Plymouth PL1 3DH, UK
King Abdullah University for Science and Technology (KAUST), Earth Sciences and Engineering Division, Thuwal, 23955-6900, Saudi Arabia
d
Met Office, FitzRoy Road, Exeter EX1 3PB, UK
b
c
a r t i c l e
i n f o
Article history:
Received 26 December 2012
Received in revised form 12 April 2013
Accepted 21 April 2013
Available online 31 May 2013
Keywords:
Phytoplankton
Ocean colour
Remote sensing
Chlorophyll
Red Sea
Comparison
a b s t r a c t
The Red Sea is a unique marine environment but relatively unexplored. The only available long-term biological
dataset at large spatial and temporal scales is remotely-sensed chlorophyll observations (an index of phytoplankton biomass) derived using satellite measurements of ocean colour. Yet such observations have rarely
been compared with in situ data in the Red Sea. In this paper, satellite chlorophyll estimates in the Red Sea
from the MODIS instrument onboard the Aqua satellite are compared with three recent cruises of in vivo fluorometric chlorophyll measurements taken in October 2008, March 2010 and September to October 2011. The performance of the standard NASA chlorophyll algorithm, and that of a new band-difference algorithm, is found to
be comparable with other oligotrophic regions in the global ocean, supporting the use of satellite ocean colour in
the Red Sea. However, given the unique environmental conditions of the study area, regional algorithms are likely to fare better and this is demonstrated through a simple adjustment to the band-difference algorithm.
© 2013 Elsevier Inc. All rights reserved.
1. Introduction
The Red Sea sustains a diverse coral reef ecosystem that resides in
one of the warmest and most saline seas in the world (Belkin, 2009;
Longhurst, 2007). It has been clustered as a fast warming large marine ecosystem (Belkin, 2009) with evidence of an abrupt temperature increase since the mid-90s, which persists till present (Raitsos
et al., 2011). Oceanic warming may have a direct or indirect impact
on marine entities and ecosystems (Edwards & Richardson, 2004);
thus, there is a need to assess past biological data to closely monitor
the relatively unexplored and fragile Red Sea ecosystem.
Long-term in situ biological datasets at large spatiotemporal scales
are not available in the Red Sea, making satellite remote-sensing of
ocean colour a vital necessity for any marine ecological studies
(Acker et al., 2008; Raitsos et al., in press), such as validation and
assimilation of biological data into ecosystem models. Over the past
two decades remote-sensing observations of ocean colour have provided unique information on the surface biomass of phytoplankton
in the Red Sea, as indexed through the chlorophyll concentration.
Yet, with the exception of a comparison between satellite and lidar
fluorescence-derived chlorophyll conducted as part of a transect between New Zealand and Italy (Barbini et al., 2004), the performance
⁎ Corresponding author at: Plymouth Marine Laboratory (PML), Prospect Place, The
Hoe, Plymouth PL1 3DH, UK. Tel.: +44 1752 633429.
E-mail address: [email protected] (R.J.W. Brewin).
0034-4257/$ – see front matter © 2013 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.rse.2013.04.018
of these satellite-chlorophyll algorithms has, to our knowledge,
never been compared with in situ data in this region. Using datasets
from three recent research cruises, covering an extensive part of the
study area, the performance of two ocean-colour chlorophyll algorithms is evaluated using in vivo fluorometric chlorophyll measurements collected over large spatial scales in the Red Sea.
2. Methodology
2.1. In situ data
Oceanographic data were collected from three research cruises
during 2008, 2010, and 2011, as part of the Research Cruises expedition programme of the Red Sea Research Center of KAUST (Fig. 1). The
R/V “Oceanus” collected data from 35 stations during October 2008
from the east coast of Saudi Arabia (21°N). In March 2010 and September to October 2011, the R/V “Aegaeo” sampled a substantial
part of the Red Sea (from 17°N to 28°N). It is the first time that the
Red Sea has been sampled in depth at such large spatial and temporal
scales.
Continuous fluorescence vertical profiles were collected at each
station using a WET Labs ECO-FLNTUs (FLNTURTD-964) fluorometer
attached to a CTD. The fluorometer was laboratory calibrated prior
to each cruise. However, the fluorometer was not field calibrated as
no independent measurements of chlorophyll (e.g. from High Performance Liquid Chromatography (HPLC) or from in vitro fluorometry)
R.J.W. Brewin et al. / Remote Sensing of Environment 136 (2013) 218–224
219
Fig. 1. The distribution of the in situ and satellite match-up data used in the study.
were taken during the cruises. The in situ chlorophyll concentrations
used in this study are therefore likely to have higher uncertainty than
those derived by HPLC or by using a field calibrated fluorometer. For
the 2010 and 2011 cruise, a consistent bias in all fluorescence profiles
was initially observed. After consultation with the sensor manufacturer, Sea-Bird Electronics, the bias was removed using out-of-water
fluorescence data, which resulted in recalibrated data within the accuracy limits of the sensor (pers. comm. Leah Trafford, Woods Hole
Oceanographic Institution). Any measurements with chlorophyll
concentrations unrealistically low (b0.01 mg m −3) were excluded
from the analysis.
For each in situ chlorophyll profile the approximate depth to
which the satellite signal is likely to penetrate was estimated. This
involved the following: (i) estimating the euphotic depth from the surface chlorophyll concentration of each profile (top five measurements
of the chlorophyll profile, representing ~5 m depth) using the model of
Morel et al. (2007); and (ii) dividing the euphotic depth by 4.6 to approximate the 1st optical depth which was assumed to be the average penetration depth of the satellite signal over the spectral range. The in situ
profiles in which the depth of the profile did not exceed the 1st optical
depth, and profiles in which the 1st optical depth was not computed
(e.g. all chlorophyll concentrations at the surface were b 0.01 mg m−3),
220
R.J.W. Brewin et al. / Remote Sensing of Environment 136 (2013) 218–224
were excluded, resulting in 401 in situ profiles, from an initial 407, that
were used in the analysis (Fig. 1). Of the 401 in situ samples the average
euphotic depth was found to be 89.5 m with a standard deviation of
11.3 m (min 51.2 m and max 119.2 m). This was consistent with
Stambler (2005), who by using a spectroradiometer computed the
mean euphotic depth of the northern Red Sea and the Gulf of Eilat
(Aqaba) to be 86.0 m ± 8.0 m. The average 1st optical depth of the
401 measurements was computed as 19.4 m with a standard deviation
of 2.5 m (min 11.1 m and max 25.9 m), and the average number of samples in each profile within the 1st optical depth was 17 with a standard
deviation of 3 (min 8 and max 25 m). The chlorophyll concentration
of each in situ sample, used to compare with the satellite chlorophyll estimates, was computed by averaging the chlorophyll concentrations in
each profile within the 1st optical depth. The average standard deviation
in this calculation was 0.012 mg m−3 chlorophyll (min 0.004 and max
0.07 mg m−3), indicating the concentrations within the 1st optical
depth were relatively homogeneous for the 401 samples. Of the 401
in situ samples, 33 were from 2008, 107 from 2010 and 261 from the
2011 cruise.
The NASA bio-Optical Marine Algorithm Dataset (NOMAD Version
2.0 ALPHA, Werdell & Bailey, 2005) was also used in the study and
processed in the following manner: all HPLC and florescencederived chlorophyll measurements, with corresponding in situ measurements of remote-sensing reflectance (Rrs) at 443, 489, 510 and
555 nm, were extracted from NOMAD (3329 samples); the NASA
OC4 chlorophyll algorithm (NASA, 2010; O'Reilly et al., 2000) was applied to the in situ Rrs measurements to estimate chlorophyll. The OC4
algorithm was used on NOMAD as opposed to the OC3, as there was a
larger percentage of in situ samples with chlorophyll and corresponding Rrs(555) when compared with the number of samples with chlorophyll and corresponding Rrs(550), particularly at low chlorophyll
concentrations. The NOMAD database was used to provide an indication of the performance of the MODIS-Aqua chlorophyll products in
the Red Sea in the context of the performance of standard NASA algorithms on a more globally-representative dataset exclusive of the Red
Sea.
2.2. Satellite data
Local Area Coverage (LAC 1 km × 1 km resolution) Level-2
MODIS-Aqua data covering the period and locations of the three
cruises were downloaded from the NASA website (http://oceancolor.
gsfc.nasa.gov/). These data products were processed by the NASA
Ocean Biology Processing Group using the most recent updates in algorithms and instrument calibration (MODIS-Aqua Reprocessing
2012.0). Standard Level 2 processing flags were applied to all satellite
data used in the match-ups. Data products that were extracted from
the Level-2 MODIS-Aqua files included: default NASA chlorophyll concentration (C), computed using the NASA OC3 algorithm (NASA, 2010;
O'Reilly et al., 2000), and Rrs at 443, 547, and 667 nm. A ±14 h window
was used between the timing of the satellite overpass and the timing of
the collection of in situ data. This ±14 h window was chosen so as to
maximise the number of match-ups acknowledging that a smaller
time window is typically used in other validation and calibration studies
(Bailey & Werdell, 2006; Hu et al., 2012; Mélin et al., 2012). However,
the time difference was noted between satellite and in situ data collection, allowing investigation into the influence of a smaller temporal
window (e.g. ±3 h) on statistical tests. When multiple match-ups
were available for a single in situ data point the closest match-up in
time was used. The nine closest pixels (3 × 3 box ~3 km2) centred
around the location of the in situ sample were extracted from the satellite data. The mean and standard deviation of the nine pixels were computed for each matchup. A group of pixels was used as it increases the
chances of a satellite match-up and assists in determining the homogeneity of the comparison point point (Bailey and Werdell, 2006). From
the 401 in situ data samples 85 corresponding match-ups were
extracted, 12 were from 2008, 25 from 2010 and 48 from the 2011
cruise. The in situ chlorophyll concentrations for these 85 match-ups
ranged from 0.046 to 0.22 mg m −3 and were predominately located
in the more oligotrophic northern half of the Red Sea (Fig. 1).
2.3. Algorithms and statistical tests
The operational chlorophyll algorithm for MODIS-Aqua is currently
the NASA OC3 algorithm (O'Reilly et al., 2000). This is an empirical
polynomial algorithm that relates the log-transformed ratio of remotesensing reflectances (X) to the chlorophyll concentration (C). The OC3
uses a three-band blue-green reflectance ratio, such that:
X ¼ log10 f½Rrs ð443Þ > Rrs ð488Þ=Rrs ð547Þg:
ð1Þ
Chlorophyll (C) is estimated according to:
a þa Xþa X 2 þa3 X 3 þa4 X 4 Þ
;
C ¼ 10ð 0 1 2
ð2Þ
where, a0 = 0.2424, a1 = − 2.7423, a2 = 1.8017, a3 = 0.0015 and
a4 = − 1.2280 (NASA, 2010).
In addition to the OC3 algorithm, the algorithm of Hu et al. (2012)
was also implemented as it has been found to perform better than the
OC3 at low chlorophyll concentrations (b 0.25 mg m −3) (Hu et al.,
2012), which are typically encountered in the Red Sea (Raitsos et al.,
in press). This algorithm is a band-difference approach that uses a colour index (CI) defined as the difference between Rrs in the green region
of the visible spectrum and a reference formed linearly between Rrs in
the blue and red region of the visible spectrum. Following Hu et al.
(2012), the colour index (CI) is defined according to
CI ¼ Rrs ð555Þ−0:5½Rrs ð443Þ þ Rrs ð670Þ:
ð3Þ
The MODIS-Aqua sensor has slightly different wavelengths to
those shown in Eq. (3). For application of Eq. (3) to MODIS-Aqua,
Hu et al. (2012) estimated Rrs(555) from Rrs(547) according to
Rrs(555) = 0.93Rrs(547), based on data in the South Pacific. They
also assumed Rrs(670) ∼ Rrs(667) considering that Rrs is very low in
oligotrophic waters at red bands, with a low signal-to-noise ratio,
due to the dominant effect of pure-water absorption. The log10transformed chlorophyll concentration is then related to CI using a
linear equation, such that
AþBCI
C ¼ 10
;
ð4Þ
where A and B are the intercept and slope of the linear regression respectively. Using NOMAD data, Hu et al. (2012) estimated A and B to
be − 0.4909 and 191.6590 respectively. Eq. (4) was designed specifically for oceanic regions with low chlorophyll (≤0.25 mg m −3).
Therefore, Hu et al. (2012) proposed a new algorithm hereafter referred to as the OCI algorithm, where at higher chlorophyll concentrations (>0.3 mg m −3) the NASA operational chlorophyll algorithm is
used (Eq. (2) in the case of MODIS-Aqua), whereas between 0.25
and 0.3 mg m −3 chlorophyll a mixture of Eqs. (4) and (2) is used, so
as to facilitate a smooth transition between equations as the chlorophyll concentration increases, and for chlorophyll concentrations
≤0.25 mg m −3 Eq. (4) is used. The OCI algorithm is expressed as
8 AþBCI
>
if ½10AþBCI ≤0:25 mg m−3
< 10
2
3
4
C ¼ α½10ða0 þa1 Xþa2 X þa3 X þa4 X Þ þ β½10AþBCI if 0:25b½10AþBCI ≤0:3 mg m−3
>
: ða0 þa1 Xþa2 X 2 þa3 X 3 þa4 X 4 Þ
10
if ½10AþBCI > 0:3 mg m−3 :
ð5Þ
The parameters α and β reflect a linear transition from Eqs. (4) to
(2) as the chlorophyll increases from 0.25 to 0.3 mg m −3 and are
R.J.W. Brewin et al. / Remote Sensing of Environment 136 (2013) 218–224
221
computed as α = (10 A + BCI − 0.25)/(0.3 − 0.25) and β = (0.3 −
10 A + BCI)/(0.3 − 0.25).
To test the performance of the satellite algorithms at deriving chlorophyll, a variety of univariate statistical tests were adopted that are
commonly used in comparisons between modelled and in situ data
(e.g. Doney et al., 2009; Friedrichs et al., 2009; Hu et al., 2012).
The log10-transformed unbiased Root Mean Square Error (Δ) and the
log10-transformed bias (δ) between satellite-estimated chlorophyll
and in situ chlorophyll, were used as an index of the precision and accuracy of the satellite-estimates respectively. Having a term available
that distinguishes systematic (δ) and random (Δ) uncertainty was
found to be useful in the context of this study. Therefore, the bias (δ)
was used as an index of accuracy in the satellite observations following
GCOS requirements (GCOS, 2006) acknowledging that other definitions
of accuracy exist in literature. The chlorophyll concentrations were
log10-transformed for these statistical tests as chlorophyll is approximately log-normally distributed over the global ocean (Campbell,
1995). In addition to these two statistics the following statistical tests
were also computed: the relative percentage Root Mean Square Error
(ψ); the unbiased relative percentage Root Mean Square Error ðÞ; the
relative percentage Absolute Mean Error (ν); the mean ratio (ϕ); the
median ratio (φ); and the squared Pearson correlation coefficient (r2) in
both linear and log10 space. These statistical tests were performed so as
to compare our results in the Red Sea with those of Hu et al. (2012)
taken in different regions of the global ocean using the OC3 and OCI
algorithms, for both SeaWiFS and MODIS-Aqua sensors. Equations used
to compute all statistical tests are provided in Supplementary Table S1.
3. Results and discussion
Scatter plots between in situ and satellite-derived chlorophyll (C)
in the Red Sea using the OC3 and OCI algorithms are shown in
Fig. 2. The results are superimposed onto scatter plots of in situ chlorophyll and in situ Rrs-derived chlorophyll (C) from the NOMAD database using the NASA OC4 algorithm (grey circles in Fig. 2).
Superimposing the Red Sea match-up data onto measurements from
NOMAD gives an indication of the performance of the OC3 and OCI algorithms in the Red Sea in the context of the performance of standard
NASA algorithms on a globally-representative dataset. It is encouraging to observe that the scatter in the OC3 and OCI algorithms around
the 1:1 line (Fig. 2) is comparable to that seen in the NOMAD data,
particularly when considering: the NOMAD data uses in situ Rrs measurements, not satellite-derived Rrs measurements; the OC4 algorithm is parameterised using NOMAD whereas the 85 Red Sea
match-ups are entirely independent of the development of the OC3
and OCI algorithms; and that the in situ chlorophyll observations
used here are based on in vivo fluorometry whereas in NOMAD they
are from a combination of HPLC, in vitro and in vivo fluorometry.
This is further verified considering Δ and δ from OC3 and OCI
(Fig. 2), between in situ and satellite-derived chlorophyll in the Red
Sea, is comparable to Δ and δ in the NOMAD data at chlorophyll concentrations b0.25 mg m −3.
Interestingly, the OCI algorithm has a lower Δ (0.128) when compared with the OC3 algorithm (0.197) indicating that the precision is
higher in the OCI algorithm. Alternatively, the OCI algorithm has a
positive bias (δ = 0.184) when compared with the OC3 algorithm
(δ = −0.009), indicating that the accuracy is higher in the OC3 algorithm. The relationship between chlorophyll and in vivo fluorescence
in surface waters can be affected by daytime-fluorescence quenching
(Cullen & Lewis, 1995). In such cases, the fluorescence signal can be
depressed in the upper layer of the water column when the phytoplankton is exposed to high irradiance. For the 2010 and 2011 cruises
(86% of satellite match-ups), corresponding measurements of Photosynthetically Available Radiation (PAR) were available on the CTD.
The positive bias in OCI algorithm was observed in both high and
Fig. 2. Scatter plots of the match-ups between in situ and satellite-derived chlorophyll
(C) in the Red Sea using the OC3, OCI and the bias-corrected OCI algorithm denoted
OCI⁎. Results are superimposed onto 3329 globally-representative measurements of
in situ chlorophyll (both High Performance Liquid Chromatography and fluorescencederived chlorophyll) from the NOMAD database (Werdell & Bailey, 2005) which are plotted against chlorophyll-estimated from corresponding in situ reflectance measurements
using the NASA OC4 algorithm.
low light samples (see Supplementary Fig. S1), suggesting that its
origin is not related to daytime-fluorescence quenching.
It is worth noting that both the OC3 and OCI algorithms are
designed for global application and that, given the unique environmental conditions in the Red Sea, it is likely that region-specific algorithms may perform better. This has already been demonstrated
in the nearby oligotrophic Gulf of Eilat (Aqaba) using a band-ratio
algorithm (Iluz et al., 2003). Here, the benefits of developing a regionspecific (Red Sea) algorithm are demonstrated using a band-difference
approach. Acknowledging that such a retuning could have be conducted
using either a band-difference (OCI) or a band-ratio algorithm (OC3), a
band-difference approach was chosen considering: (i) the high precision
of the OCI algorithm compared with OC3 (Fig. 2); (ii) the simplicity of
re-tuning a linear equation (Eq. (4)) when compared with a 4th order
polynomial (Eq. (2)); and (iii) its potential benefits in oligotrophic
222
R.J.W. Brewin et al. / Remote Sensing of Environment 136 (2013) 218–224
waters (Hu et al., 2012). Eq. (4) was re-tuned so as to remove the bias between the OCI-derived chlorophyll and in situ chlorophyll. This involved
minimising Eq. (4) using in situ chlorophyll and CI obtained through
Eq. (3) and MODIS-Aqua Rrs, keeping the parameter B in Eq. (4) (slope
of the linear regression between CI and log10-chlorophyll) constant and
allowing the parameter A in Eq. (4) (intercept of the linear regression
between CI and log10-chlorophyll) to vary. A value of −0.6721 was computed for A, which is lower than −0.4909 used in the OCI algorithm (Hu
et al., 2012). The OCI algorithm with the adjusted A parameter, tuned to
data in the Red Sea, is hereafter referred to as OCI⁎, with the superscript *
denoting that it has a different parameterisation to the OCI algorithm.
The results plotted in Fig. 2 highlight no change in Δ (0.128) when compared with OCI. However, the bias has been removed (δ = 0.000).
The reasons for this change in the parameter A could be as follows:
(i) a modification in the relationship between chlorophyll and other
non-phytoplankton components (e.g. non-algal suspended material
and/or coloured-dissolved organic matter) relative to the average relationships observed in the global oceans (Iluz et al., 2003); (ii) differences in the phytoplankton community structure in the Red Sea
compared with the average community structure observed in the
global ocean for a given chlorophyll concentration, perhaps leading
to a different chlorophyll-specific absorption coefficient; or (iii) the
unique environmental characteristics of the Red Sea, being among
the warmest and most saline seas (Belkin, 2009; Longhurst, 2007),
such that a similar phytoplankton community to that typically observed in the global ocean is present, but adapted to the unique environmental conditions, and thus altering the chlorophyll-specific
absorption coefficient. This adjustment to the parameter A in Eq. (4)
may also incorporate a slightly different relationship between Rrs(555)
and Rrs(547) in the Red Sea, to that used by Hu et al. (2012) derived
from the South Pacific. Nonetheless, it remains to be revealed whether
this adjustment to the OCI algorithm for the Red Sea is an artifact of
the limited number of match-ups available. Another possible source of
the differences is the use of in vivo fluorescence to estimate chlorophyll.
The fluorescence yield can vary between species (Kiefer, 1973b;
Strickland, 1968) and within a single species subjected to different environmental conditions (Kiefer, 1973a; Slovacek & Bannister, 1973).
Additional research campaigns in the Red Sea, incorporating both chlorophyll (HPLC and/or in vitro fluorescence) and optics, are required to
test these hypotheses related to the potential sources of the differences
in the regional algorithm compared with the global.
The statistical performance of the OC3, OCI and OCI⁎ algorithms
using the match-ups in the Red Sea was compared with that of Hu
et al. (2012) taken in different regions of the global ocean, for both
SeaWiFS and MODIS-Aqua sensors at chlorophyll concentrations
b0.25 mg m −3 (Table 1). Consistent with the results shown in
Fig. 2, the performance of the OC3 and OCI algorithms in the Red
Sea is comparable to the performance of these algorithms on a matchup dataset acquired in other regions of the global ocean, as indexed
by similar performance in statistical tests (Table 1). In the study of
Hu et al. (2012), they used a ±3 h window between satellite and
in situ observation whereas a ±14 h window was used in this study.
Table 2 shows the effect of changing the temporal window of our
match-ups to ±12 h, ±8 h and ±3 h on Δ and δ. No improvement in
Table 2
Statistical results of the match-ups in the Red Sea for different time windows between
satellite and in situ data.
Algorithm
Δ
δ
N
Time window [h]
OC3
OC3
OC3
OC3
OCI
OCI
OCI
OCI
OCI⁎
OCI⁎
OCI⁎
OCI⁎
0.195
0.201
0.204
0.236
0.128
0.132
0.139
0.137
0.128
0.132
0.139
0.137
-0.009
-0.006
0.006
0.050
0.185
0.186
0.212
0.298
0.000
0.001
0.028
0.114
85
73
48
18
85
73
48
18
85
73
48
18
b14
b12
b8
b3
b14
b12
b8
b3
b14
b12
b8
b3
N refers to the number of samples used to compute statistics.
statistical metrics was observed when decreasing the temporal
window. For the 85 match-ups, the average standard deviation divided
by the mean chlorophyll concentration of the 3 × 3 box of pixels surrounding the in situ sample was 7.5% ± 5.4%, indicating that the comparison points were relatively homogeneous and perhaps explaining
why there was no significant improvement when decreasing the temporal window. It is also worth noting that decreasing the temporal
window will reduce the number of samples (Table 2) which will also
influence statistical tests. The choice of the ±14 h window appears to
be an adequate compromise between the number of match-ups and
the quality of statistics for this particular dataset.
Figs. 3, 4 and 5 show in situ chlorophyll concentrations (white and
red circles, where red indicates a satellite match-up) superimposed
using the same colour scale onto a clear MODIS-Aqua image acquired
around the time of each cruise, for the 2008, 2010 and 2011 cruises
respectively. This allows a visual comparison between the in situ
and satellite-derived chlorophyll concentrations. MODIS Level-2
quality control flags were turned off in these images to show the circulation features and maximise coverage for this visual comparison.
The reader is referred to Supplementary Fig. S2, S3 and S4, for the
same images but with the Level-2 quality control flags turned on.
Scatter plots of satellite and in situ match-ups are also plotted for
the OC3, OCI and OCI⁎ algorithms in each figure (note Level-2 quality
control flags were turned on for all match-ups, see Section 2.2). The
performance of algorithms are seen to vary between cruises. The
OC3 algorithm has a lower Δ and a δ closer to zero in the 2008 and
2010 cruises (12 and 24 match-ups respectively) when compared to
the OCI algorithm, whereas the OCI algorithm has a lower Δ in the
2011 cruise (48 match-ups). Correcting for the bias using all
match-ups through adjusting the parameter A in Eq. (4) results in
the OCI⁎ algorithm having a δ closer to zero in the 2010 and 2011
cruises (86% of the match-ups),when compared with the OCI algorithm, but not for the 2008 cruise. When overlaying all the in situ samples on a clear MODIS-Aqua image taken around the period of each
cruise (Figs. 3, 4 and 5) it is encouraging to observe that the spatial
pattern in the satellite estimated chlorophyll is reflected in the spatial
pattern of the in situ observations, particularly when considering the
Table 1
A statistical comparison of the match-ups performed in the Red Sea with those of Hu et al. (2012) for the global ocean at chlorophyll concentration b0.25 mg m−3.
Study
Sensor
Region
Algorithm
ψ [%]
[%]
ν [%]
ϕ
φ
2
rlinear
2
rlog
N
Hu et al. (2012)
Hu et al. (2012)
Hu et al. (2012)
Hu et al. (2012)
This study
This study
This study
SeaWiFS
SeaWiFS
MODIS-Aqua
MODIS-Aqua
MODIS-Aqua
MODIS-Aqua
MODIS-Aqua
Global
Global
Global
Global
Red Sea
Red Sea
Red Sea
OC4
OCI
OC3
OCI
OC3
OCI
OCI⁎
535.8
91.8
77.7
43.9
45.9
78.1
33.2
54.2
47.2
44.2
32.7
42.9
49.3
28.8
41.5
36.8
32.0
25.4
35.3
60.8
24.7
1.79
1.40
1.24
1.15
1.08
1.60
1.05
1.19
1.16
1.05
1.04
1.00
1.51
0.99
0.01
0.31
0.42
0.62
0.57
0.35
0.35
0.33
0.39
0.66
0.71
0.48
0.31
0.30
357
357
63
63
85
85
85
N refers to the number of samples used to compute statistics.
R.J.W. Brewin et al. / Remote Sensing of Environment 136 (2013) 218–224
223
trophic range, the performance of satellite-chlorophyll algorithms at
higher concentrations in the Red Sea are still unknown. Furthermore,
considering all match-ups were in the oligotrophic northern half of the
Red Sea, the performance of satellite–chlorophyll algorithms in the
more shallower and optically-complex southern Red Sea, and in the
coastal zone, is still yet to be tested. Analytical approaches that partition
the influence of the different water constituents on the reflectance
spectrum (IOCCG, 2006) may be better suited to these environments.
Nonetheless, it is encouraging to observe that at higher chlorophyll
concentrations, and in the southern Red Sea, the approximate spatial
pattern in the in situ observations is reflected in single MODIS-Aqua
scene taken around the time period of each cruise (e.g. Fig. 5).
4. Summary
Fig. 3. In situ chlorophyll concentrations (white and red circles, where red indicates a
satellite match-up) from the October 2008 cruise superimposed onto a clear MODIS
image on 24 October 2008 for the OC3, OCI and the OCI⁎ algorithms. MODIS Level-2
quality control flags were turned off to show the circulation features and maximise
coverage. Scatter plots of the corresponding satellite (SAT) match-ups are also
shown for the 2008 cruise. Nin denotes the number of in situ samples used in the
2008 cruise and Nm denotes the number of satellite match-ups.
temporal mis-match between a single MODIS-Aqua scene and in situ
observations collected over multiple days.
The in situ chlorophyll concentrations for the 85 match-ups ranged
from 0.046 to 0.22 mg m−3. The lower range was due to the location of
the match-ups being predominately in the northern oligotrophic Red
Sea, and that March, September and October, when all in situ samples
were taken, lie either at the beginning or end of the phytoplankton seasonal succession (Raitsos et al., in press), when lower chlorophyll concentrations are observed in comparison to winter months (Acker et al.,
2008). Considering the match-ups were not representative of the entire
We made use of 401 in vivo fluorometric chlorophyll profiles
collected from three research cruises covering a large area of the
Red Sea. Eighty-five of these chlorophyll observations were matched
to MODIS-Aqua satellite remote sensing reflectance (Rrs) data, at a
±14 h window and at a spatial resolution of ~ 3 km 2, ranging from
0.046 to 0.22 mg m −3. Two satellite algorithms were applied to the
Rrs data to estimate chlorophyll, this included the standard NASA default algorithm (OC3) and a new band-difference algorithm (OCI)
designed specifically for low chlorophyll waters. Using a suite of statistical tests the performance of the two algorithms was compared
with in situ chlorophyll concentrations, and with similar statistics
published using globally-representative match-up data from other regions to that of the Red Sea.
The precision and accuracy of the standard NASA algorithm in
oligotrophic waters of the Red Sea were found to be comparable with
other areas in the global ocean. The band-difference algorithm had a
better precision than the OC3 but was less accurate, as indexed by a
larger bias. Correcting for this bias improved its overall performance
and implies that, whereas the performance of standard global algorithms are comparable in the Red Sea with that in other regions of the
oceans, region-specific algorithms are likely to perform better. Our
results support the use of ocean-colour data in the Red Sea, which is
encouraging considering it is the only available long-term biological
dataset at large spatial and temporal scales. Ongoing research campaigns
will verify our conclusions, and ascertain the extent to which they hold
for higher chlorophyll waters and in the more optically-complex southern portion of the Red Sea.
Acknowledgements
The authors would like to thank the captains and the crews of the
R/V “Aegaeo” of the Hellenic Centre for Marine Research (HCMR), and
of the R/V “Oceanus” from the Woods Hole Oceanographic Institution
Fig. 4. In situ chlorophyll concentrations (white and red circles, where red indicates a satellite match-up) from the March 2010 cruise superimposed onto a clear MODIS image on 29
March 2010 for the OC3, OCI and the OCI⁎ algorithms. MODIS Level-2 quality control flags were turned off to show the circulation features and maximise coverage. Scatter plots of
the corresponding satellite (SAT) match-ups are also shown for the 2010 cruise. Nin denotes the number of in situ samples used in the 2010 cruise and Nm denotes the number of
satellite match-ups.
224
R.J.W. Brewin et al. / Remote Sensing of Environment 136 (2013) 218–224
Fig. 5. In situ chlorophyll concentrations (white and red circles, where red indicates a satellite match-up) from the September-October 2011 cruise superimposed onto a clear
MODIS image on 10 October 2011 for the OC3, OCI and the OCI⁎ algorithms. MODIS Level-2 quality control flags were turned off to show the circulation features and maximise
coverage. Scatter plots of the corresponding satellite (SAT) match-ups are also shown for the 2011 cruise. Nin denotes the number of in situ samples used in the 2011 cruise and
Nm denotes the number of satellite match-ups.
(WHOI), who made the data collection possible. We thank in particular Leah Trafford and Amy Bower for their assistance on the cruise
data used in this study. We thank Shubha Sathyendranath and
Chuanmin Hu for suggestions on an earlier version of the manuscript
and two anonymous reviewers for their useful comments. This research was supported by the King Abdullah University for Science
and Technology (KAUST), Kingdom of Saudi Arabia and the UK National Centre for Earth Observation, and is a contribution to the
Ocean Colour Climate Change Initiative of the European Space Agency
(ESA).
Appendix A. Supplementary data
Supplementary data to this article can be found online at http://
dx.doi.org/10.1016/j.rse.2013.04.018.
References
Acker, J., Leptoukh, G., Shen, S., Zhu, T., & Kempler, S. (2008). Remotely-sensed chlorophyll a observations of the northern Red Sea indicate seasonal variability and influence of coastal reefs. Journal of Marine Systems, 69(3–4), 191–204.
Bailey, S. W., & Werdell, P. J. (2006). A multi-sensor approach for the on-orbit validation of ocean color satellite data products. Remote Sensing of Environment, 102,
12–23.
Barbini, R., Colao, F., De Dominicis, L., Fantoni, R., Fiorani, L., Palucci, A., et al. (2004).
Analysis of simultaneous chlorophyll measurements by lidar fluorosensor, MODIS
and SeaWiFS. International Journal of Remote Sensing, 25(11), 2095–2110.
Belkin, I. M. (2009). Rapid warming of large marine ecosystems. Progress in Oceanography,
81, 207–213.
Campbell, J. W. (1995). The lognormal distribution as a model for bio-optical variability
in the sea. Journal of Geophysical Research, 100(C7), 13237–13254.
Cullen, J. J., & Lewis, M. R. (1995). Biological processes and optical measurements near
the sea-surface: Some issues relevant to remote sensing. Journal of Geophysical
Research, 100(C7), 13,255–13,266 (13).
Doney, S. C., Lima, I., Moore, J. K., Lindsay, K., Behrenfeld, M. J., Westberry, T. K., et al.
(2009). Skill metrics for confronting global upper ocean ecosystem-biogeochemistry
models against field and remote sensing data. Journal of Marine Systems, 76, 95–112.
Edwards, M., & Richardson, A. J. (2004). Impact of climate change on marine pelagic
phenology and trophic mismatch. Nature, 430, 881–884.
Friedrichs, M. A. M., Carr, M. -E., Barber, R. T., Scardi, M., Antoine, D., Armstrong, R. A.,
et al. (2009). Assessing the uncertainties of model estimates of primary productivity in the tropical Pacific Ocean. Journal of Marine Systems, 76(1–2), 113–133.
GCOS (2006). Systematic observation requirements from satellite-based data products
for climate. Tech. rep. 7 bis, avenue de la Paix, CH-1211 Geneva 2, Switzerland:
World Meteorological Organisation (WMO).
Hu, C., Lee, Z., & Franz, B. (2012). Chlorophyll a algorithms for oligotrophic oceans:
A novel approach based on three-band reflectance difference. Journal of Geophysical
Research, 117, C01011.
Iluz, D., Yacobi, Y. Z., & Gitelson, A. (2003). Adaptation of an algorithm for chlorophyll-a
estimation by optical data in the oligotrophic Gulf of Eilat. International Journal of
Remote Sensing, 24(5), 1157–1163.
IOCCG (2006). Remote sensing of inherent optical properties: Fundamentals, tests of
algorithms, and applications. In Z. P. Lee (Ed.), Tech. rep. Reports of the international
ocean-colour coordinating group, No. 5, Dartmouth, Canada: IOCCG.
Kiefer, D. A. (1973a). Chlorophyll a fluorescence in marine centric diatoms: Responses
of chloroplasts to light and nutrient stress. Marine Biology, 23, 39–46.
Kiefer, D. A. (1973b). Fluorescence properties of natural phytoplankton populations.
Marine Biology, 22, 263–269.
Longhurst, A. R. (2007). Ecological geography of the sea (2nd ed.): Elsevier.
Mélin, F., Zibordi, G., & Berthon, J. -F. (2012). Uncertainties in remote sensing reflectance from MODIS-Terra. IEEE Geoscience and Remote Sensing Letters, 9, 432–436.
Morel, A., Huot, Y., Gentili, B., Werdell, P. J., Hooker, S. B., & Franz, B. A. (2007). Examining the consistency of products derived from various ocean color sensors in open
ocean (case 1) waters in the perspective of a multi-sensor approach. Remote
Sensing of Environment, 111, 69–88.
NASA (2010). Ocean color chlorophyll (OC) v6. URL. http://oceancolor.gsfc.nasa.gov/
REPROCESSING/R2009/ocv6/
O'Reilly, J. E., Maritorena, S., Siegel, D., O'Brien, M. C., Toole, D., Mitchell, B. G., et al.
(2000). Ocean color chlorophyll a algorithms for SeaWiFs, OC2, and OC4:. Tech.
rep. In S. B. Hooker, & E. R. Firestone (Eds.), SeaWiFS postlaunch technical report series.
Vol. 11. SeaWiFS postlaunch calibration and validation analyses, part 3 (pp. 9–23).
Greenbelt, Maryland: NASA, Goddard Space Flight Center.
Raitsos, D. E., Hoteit, I., Prihartato, P. K., Chronis, T., Triantafyllou, G., & Abualnaja, Y.
(2011). Abrupt warming of the Red Sea. Geophysical Research Letters, 38 L14601.
Raitsos, D. E., Pradhan, Y., Brewin, R. J. W., Stenchikov, G., & Hoteit, I. (2013). Remote
sensing the phytoplankton seasonal succession of the Red Sea. PloS One (in press).
Slovacek, R., & Bannister, T. (1973). NH4Cl activation of the fluorescence yield in CO2
starved Chlorella pyrenoidosa. Biochimica et Biophysica Acta, 325, 114–119.
Stambler, N. (2005). Bio-optical properties of the northern Red Sea and the Gulf of Eilat
(Aqaba) during winter 1999. Journal of Sea Research, 54, 186–203.
Strickland, J. (1968). Continuous measurement of in vivo chlorophyll: A precautionary
note. Deep Sea Research, 15, 225–227.
Werdell, P. J., & Bailey, S. W. (2005). An improved in-situ bio-optical data set for ocean
colour algorithm development and satellite data production validation. Remote
Sensing of Environment, 98, 122–140.