rsulines,
in Colour
ieophytin)
S) images
1istorical
shipboard
isonal ani
Seasonal
riy" fat
ation may
e a plus
ir Coastal
1981. Les
oplancton
toutefois
'rannuels)
é par une
ment, les
pigments
data and
san color
research
primary
| sensing
program,
ical CZCS
ht of the
eractions
a complex
j£ phyte
ge First,
onditions
arine and
1e basin
the St.
nic and
e remote
(Case II
plankton
ly to the
ılar, the
rences Of
The objectives of this paper are thus to use
CZCS imagery to estimate the concentrations of
the major photosynthetic pigment (chlorophyll
a plus phaeophytin) in the GSL and to charac-
terize their spatio-temporal distribution.
This study ig the first to provide ‘a synoptic
viewoof phytoplankton pigments in the whole
Qulf of St. Lawrence using remote sensing
data.
2. METHODOLOGY
Study area
The GSL system (Figure 1) is a semi-enclosed
sea of about 200 000 km?, located between 45-
57 °N and 56-70 °W. It is connected with the
Atlantic Ocean through Cabot Strait (104 km
wide and 480 m deep), and Belle-Isle Strait
(16 km wide and 60 m deep). The bathymetry of
the gulf is dominated by the Laurentian
Channel (350 m), running from Cabot Strait to
the Saguenay River with a branch toward the
northeast. The southern part of the gulf is a
large shallow area of average depth 50 m. Due
to a large annual freshwater input of about
424 kn? (Bugden et al, 1982), most of which
(75$) originating from the St. Lawrence River
system, circulation of the GSL waters is basi-
cally density-driven and the water column in
the gulf is vertically stratified into three
layers during the summer.
Image analysis
Surface phytoplankton pigments (chlorophyll a
plus phaeophytin) in the GSL were evaluated
using a set of 80 CZCS images taken between
March and September during the first three
years of the satellite life (1979-1981). only
these years were considered because the sensor
was affected by erratic loss of sensitivity
afterward (Evans and Gordon, 19945.
Radiometric corrections were done using the
Sensor sensitivity reduction correction
factors of Gordon‘ et a1.27^$1983b) and the
calibration coefficients provided by Williams
et al. (1985). Clouds and land were masked
using the Ekstein and Simpson (1991a) method,
while pixels contaminated by signal overflow
Were detected with the Mueller (1988) proce-
dure. Only gain 1 and 2 images were used in
this study because higher gains generated
problems related to molecular spread cor-
rection. The molecular scattering (Rayleigh)
Was evaluated using a single scattering model
(Gordon et al... :1983a), on the basis: that
Single-Rayleigh scattering is comparable to
the multiple-Rayleigh scattering method
(Simpson, 1993) : However, the single
Scattering model was modified using an
extraterrestrial solar radiance coefficient
(Ekstein and Simpson, 1991b) and an ozone
Optical depth value (Sturm, 1981).
ea subtraction was done using an interac-
EA technique (Arnone and LaViolette,
1). This approach was selected because many
407
problems prevented the use of the clear water
pixel method in. the. GSL. It allows the ope-
raton. to..select the optimum Angstróm
coefficient for each of the first three CZCS
channels. After the atmospheric corrections
were completed, the phytoplankton pigments
were calculated using the algorithm proposed
by Gordon et al. (1983a).
Sediment contaminated waters were detected by
transforming the reflectances at 443, 520 and
550 nm into a IHS image. Spectral signatures
of sediment contaminated waters were then cal-
culated in this new color space and used to
filter the images using a supervised classifi-
cation method.
The images were then geometrically corrected
by using the georeferenced data included in
the CZCS raw data to produce a first approxi-
mation correction which was then fine-tuned by
rectifying the images on a stereographic
projection using the World Data Bank Feature
Codes. The nearest-neighbour method was used
as the resampling scheme to avoid distorting
the pixel's pigment content.
Finally, the mean pigment value of overlaping
pixels free from clouds, overflow, and suspen-
ded matter from each geometrically corrected
image was calculated. Seasonal analysis was
based on bi-monthly groups: April-May, June-
July, and August-September, (hereafter refered
to spring, summer, and fall). Interannual
variability was evaluated using the mean of
all the available images for each of the three
years.
MAGDALEN x
SHALLOWS N Et
AGDALEN \ tei fi
"ISLAND... ^; cABOT, SN > Y
s A! . t.
NEW-BRUNSWICK
Figure 1. The St. Lawrence Gulf (from de
Lafontaine et al., 1991)
Data validation
The scarcity of available in situ measurements
simultaneous with the CZCS images made the
data validation process difficult. Using all
the known available data, the mean chloro-
phyll a plus phaeophytin values for the top
fifth of the euphotic layer was calculated to
compare with evaluated pigment values from the
satellite images (Morel and Berthon, 1989).
Results of the linear regression showed that
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996