Quantification of surface suspended sediments

int. j. remote sensing, 2002, vol. 23, no. 16, 3229–3249
QuantiŽ cation of surface suspended sediments along a river dominated
coast with NOAA AVHRR and SeaWiFS measurements: Louisiana,
USA
S. W. MYINT
Department of Geography, University of Oklahoma, Norman, OK 73019, USA;
e-mail: [email protected]
and N. D. WALKER
Coastal Studies Institute/Dept of Oceanography and Coastal Sciences,
Louisiana State University, Baton Rouge, LA 70803, USA;
e-mail: [email protected]
(Received 22 May 2000; in Ž nal form 8 August 2001)
Abstract. The ability to quantify suspended sediment concentrations accurately
over both time and space using satellite data has been a goal of many environmental researchers over the past few decades. This study utilizes data acquired
by the NOAA Advanced Very High Resolution Radiometer (AVHRR) and the
Orbview-2 Sea-viewing wide Ž eld-of-view (SeaWiFS) ocean colour sensor, coupled
with Ž eld measurements to develop statistical models for the estimation of nearsurface suspended sediments and suspended solids. ‘Ground truth’ water samples
were obtained via helicopter, small boat and automatic water sampler within a
few hours of satellite overpasses. The NOAA AVHRR atmospheric correction
was modiŽ ed for the high levels of turbidity along the Louisiana coast. Models
were developed based on the Ž eld measurements and re ectance/radiance measurements in the visible and near infrared Channels of NOAA-14 and Orbview-2
SeaWiFS. The best models for predicting surface suspended sediment concentrations were obtained with a NOAA AVHRR Channel 1 (580–680 nm) cubic model,
Channel 2 (725–1100 nm) linear model and SeaWiFS Channel 6 (660–680 nm)
power model. The suspended sediment models developed using SeaWiFS Channel
5 (545–565 nm) were inferior, a result that we attribute mainly to the atmospheric
correction technique, the shallow depth of the water samples and absorption
eVects from non-sediment water constituents.
1.
Introduction
Louisiana coastal waters receive inorganic sediments from the Mississippi River,
the largest river in North America and sixth largest worldwide in terms of discharge.
Annual water and sediment discharges average 18 400 m3 sÕ 1 and 210×106 tons yrÕ 1
(Milliman and Meade 1983). The discharge and sediment load of the Mississippi
River enters the northern Gulf of Mexico in two locations, through the Mississippi
Internationa l Journal of Remote Sensing
ISSN 0143-116 1 print/ISSN 1366-590 1 online © 2002 Taylor & Francis Ltd
http://www.tandf.co.uk/journals
DOI: 10.1080/01431160110104700
3230
S. W. Myint and N. D. Walker
Figure 1. Map of the study region showing where the Mississippi River and Atchafalaya
River enter into the northern Gulf of Mexico along the Louisiana coastline.
River bird-foot delta and the Atchafalaya River delta, 200 km to the west (Ž gure 1).
The Atchafalaya River (including the Red River  ow) carries about 30% of the
volume and 50% of the suspended sediment load of that in the Mississippi River
(Mossa and Roberts 1990). The low salinity, high turbidity plumes of these two
rivers are discharged into very diVerent coastal environmental settings. The main
branch of the Mississippi River  ows into the Gulf of Mexico through several passes,
intersecting a headland that projects about 60 km south of the mainland. The
Atchafalaya River water and sediments are discharged through Atchafalaya Bay
onto a broad and shallow shelf. The clay and Ž ne silts that remain in suspension
are transported seaward onto the inner shelf and alongshore by discharge inertia
and wind-driven coastal currents.
In this article, we investigate techniques for quantifying suspended sediment
concentrations for Louisiana coastal and bay waters using re ectance measurements
QuantiŽ cation of suspended sediments from satellite
3231
of NOAA AVHRR and radiance measurements of the Orbview-2 SeaWiFS. Satelliteacquired re ectance measurements can provide valuable information on the distribution of river water and sediments on the continental shelf as well as on the circulation
processes aVecting the fate of river-borne material. The development of better techniques for quantifying water constituents using satellite re ectance measurements
will lead to major improvements in the understanding and modelling of sediment
re-suspension and transport in river-in uenced coastal environments.
Many researchers have used optical remote sensing techniques to study the
spatial extent and temporal changes in suspended sediments of reservoirs, lakes and
rivers and the coastal ocean (e.g. Goldman et al. 1974, Gagliardini et al. 1984, Rouse
and Coleman 1976, Curran and Novo 1988, Stumpf 1988, Stumpf and Pennock
1989, Moeller et al. 1993, Wang et al. 1996, Walker 1996, Walker and Hammack
2000). Riverine and coastal waters are comprised of a diverse array of living, nonliving and once living material that vary over time and space. The main constituents
are suspended inorganic matter, suspended organic matter, phytoplankton , dissolved
organic matter and detritus (Kondratyev and Filatov 1999). In the coastal zone of
Louisiana, suspended sediments (inorganic matter) are a large component of the
in-water substances with sediment loads in the Mississippi and Atchafalaya River
typically ranging from 100 to 400 mg lÕ 1 (Mossa 1990, Walker 1996, Allison et al.
2000). Concentrations of suspended inorganic matter due to river discharge as well
as from the re-suspension of unconsolidated bottom sediments from wind-waves are
relatively high (Huh et al. 1991, Moeller et al. 1993, Huh et al. 1996, Walker 1996,
Allison et al. 2000, Walker and Hammack 2000, Huh et al. 2001). The contribution
of suspended organic matter may also be relatively high as a result of the high levels
of phytoplankto n production (Rabalais et al. 1991) as well as from detritus derived
from the extensive marshes. SAIC (1989) showed that a major optical property of
the Louisiana shelf waters is the large amount of yellow substances found in the
estuarine and river discharges.
The determination of suspended sediments or total suspended solids from water
re ectance is based on the relationship between the scattering and absorption properties of water and its constituents (Maul 1985). Most of the scattering is caused by
suspended sediments whereas the absorption is controlled by chlorophyll a and
colored dissolved or particulate matter. The spectral behaviour of sediments is
dependant both on the particle size distribution and mineral composition (Maul
1985, Novo et al. 1989). For any given concentration, Ž ne-grained material contains
more particles and thus scatters more than would an equal weight of coarse-grained
material. Sediment type also eVects the relationship between re ectance and suspended sediment concentration due to the unique re ectance spectra of diVerent
sediments. For example, Novo et al. (1989) found that a sample of white clay was
four times more re ective at 600 nm than red silt. Moore (1977) demonstrated that
the peak in re ectance shifts to longer wavelengths as the suspended sediment
concentration increases (Curran and Novo 1988). The absorptive in-water components such as chlorophyll a and CDOM have been shown to lower the re ectance in
a substantial way (Curran and Novo 1988). However, the absorptive eVects of
CDOM are generally in wavelengths less than 500 nm. Based on these results, it
might be surprising to obtain a single model to hold for every synoptic situation as
the components of coastal and inland waters change over time.
In this article, we use a modiŽ ed version of an atmospheric correction technique
developed for NOAA AVHRR satellite data by Stumpf and Pennock (1989) and
S. W. Myint and N. D. Walker
3232
later applied by Froidefond et al. (1993) and Walker (1996), among others. The
modiŽ cation we present here enables development of statistical models that estimate
suspended sediment concentrations in the highly turbid waters of the Atchafalaya
River out ow region (Walker and Hammack 2000, Walker 2001). The models are
then applied to independent data collected in subsequent years to evaluate the
robustness and accuracy of these algorithms under diVerent environmental conditions. Statistical models for estimating suspended sediment concentrations are also
developed for the Orbview-2 SeaWiFS, using the standard NASA atmospheric
correction formulation (Gordon and Wang, 1994).
The primary objectives of this study were:
(1) To develop improved techniques for the quantiŽ cation of suspended sediments in turbid coastal environments (Case 2 waters) using visible and near
infrared Channels of NOAA AVHRR and Orbview-2 SeaWiFS.
(2) To apply these models to independent datasets to evaluate the robustness and
accuracy for estimating suspended sediment concentrations under varying
environmental conditions.
(3) To identify and evaluate the main sources of error in model development.
2.
Methodology
The main steps undertaken were the (a) collection of ‘ground truth’ measurements
coincident with clear-sky satellite overpasses, (b) atmospheric correction of the
satellite data to obtain water re ectances, (c) investigation of relationships between
the in-situ measurements and the satellite-derived water re ectances, (d) development
of statistical models, and (e) application, testing and assessment of the new models
on independent datasets.
2.1. Satellite data overview
The satellite data used in this study were received and analysed at the Earth
Scan Laboratory, Coastal Studies Institute, Louisiana State University. Re ectance
measurements from the NOAA-14 Advanced Very High Resolution Radiometer
(AVHRR) were used as they were obtained with the highest sun angle, in midafternoon. The AVHRR sensor has two Channels appropriate for analysing suspended material, Channel 1 in the red visible portion from 580 to 680 nm, and
Channel 2 in the near infrared from 725 to 1100 nm (Table 1). SeaWiFS radiance
measurements are obtained in six visible channels and two near infrared channels
from 402 to 885 nm (Table 1). The SeaWiFS channels oVer increased radiometric
Table 1.
NOAA-14 AVHRR and SeaWiFS ocean colour scanner spectral Bands/Channels.
Channel/band
1
2
3
4
5
6
7
8
NOAA-14 AVHRR
wavelength (nm)
SeaWiFS
Wavelength (nm)
580–680
725–1100
3550–3930
10 300–11 300
11 500–12 500
–
–
–
402–422
433–453
480–500
500–520
545–565
660–680
745–785
845–885
QuantiŽ cation of suspended sediments from satellite
3233
sensitivity over NOAA AVHRR visible and near infrared channels (Stumpf 1992) as
the sensor’s primary purpose is to estimate chlorophyll a and ocean productivity
(O’Reilly et al. 1998). Both sensors have pixel sizes of approximatel y 1.1 km at nadir.
2.2. Field data
A crucial step in the quantitative use of remote measurements is the collection
of in situ water samples coincident with clear-sky satellite overpasses. It is important
that the measurements be made as close together in time as possible. Helicopters
provide an eYcient platform for collection of water samples for several reasons.
First, one can sample a large area rapidly, enabling the sampling of a range of water
types. Second, one can locate areas of uniform colour indicating some homogeneity
in the water mass. Third, if clouds are in the Ž eld of view, one can locate cloud-free
areas more easily at the elevated altitude. The ‘ground truth’ data used in this study
were obtained by helicopter, by high speed boat and by automatic water samplers.
For the NOAA AVHRR model development, two helicopter trips were performed
over the Atchafalaya region on 26 April 1996 and 21 June 1996, with data collection
lasting about four hours, centred on the satellite overpass time. Additional ‘ground
truth’ water samples were obtained by automatic water samplers in West Cote
Blanche Bay west of Atchafalaya Bay on 26 March and 27 March 1998 within one
hour of satellite overpass. On 21 March 2001, additional water samples were collected
by a small fast coastal research vessel in Fourleague Bay and on the inner shelf, east
of Atchafalaya Bay. Also on 21 March, water samples were collected by a research
cruise on the inner shelf near the Mississippi bird-foot delta. The water samples
collected on 21 March 2001 were used to test the NOAA AVHRR models. The
SeaWiFS ground truth data were obtained on 26 April 2000, west of the Mississippi
River delta, in Barataria Bay and on the inner shelf seaward of the Bay. As of yet,
additional ‘ground truth’ data have not been obtained to test the SeaWiFS suspended
sediment model.
The 500 ml water samples, obtained from the top 0.5 m of water, were refrigerated
in the dark and processed within a few days of collection. The concentrations of
total suspended solids (TSS) were determined by Ž ltering through GF/F glass Ž ber
Ž lters following the methods described in USGS (1987). Filters were dryed at 60°C
for 12 hours and re-weighed. The inorganic sediment fraction or suspended sediment
(SS) was determined by ashing the Ž lters at 500°C for 12 h and re-weighing. For the
water samples analysed, the inorganic fraction contributed 80% or more to the total
suspended solid (TSS) weight when TSS exceeded 10 mg lÕ 1. At lower TSS concentrations, the organic fraction was as much as 50%. Visual examination of the Ž lter
papers before ashing revealed that the organic material was primarly phytoplankto n
(Walker and Hammack 2000).
2.3. Atmospheric correction of the data
The atmospheric correction of NOAA AVHRR Channel 1 and 2 data were
performed using a technique described by Stumpf and Pennock (1989) and Stumpf
(1992). This technique incorporates corrections for changes in solar irradiance,
aerosols, Rayleigh scattering and sunglint. It is considered valid over approximately
200 km in the east-west direction. Their original ‘bias correction’ method has three
main steps. First, corrections are performed to Channels 1 and 2 to compensate for
changes in downwelling solar irradiance and transmission changes based on the
Rayleigh optical depth and the gaseous absorption optical depth. Subsequently,
S. W. Myint and N. D. Walker
3234
Channel 2 re ectances are subtracted from Channel 1 re ectances. The purpose of
this step is to remove contamination from aerosols and sunglint. However, in doing
so one assumes that the water re ectance in Channel 2 is zero (Gordon and Morel
1983). Then, a clear-water pixel is identiŽ ed in the area of interest and the re ectance
value of this pixel is subtracted from the entire scene. This last step removes contamination due to Rayleigh and aerosol scattering. This technique has been used successfully in studies of Mobile Bay and Delaware Bay (Stumpf 1992) and the Mississippi
plume region (Walker 1996). For our study area, the technique required modiŽ cations
because the Channel 2 measurements in the Atchafalaya Bay region were non-zero
(Ž gure 2( b)), thus invalidating the assumption that water re ectance in Channel 2 is
negligible. This problem was solved by modifying the bias correction technique for
use with Channel 1 alone. In short, step 2 was omitted from the above procedure.
This modiŽ ed technique was successful in retaining the water re ectance patterns
within the Atchafalaya and adjacent bays and on the inner shelf. Three new variables
were created in the atmospheric correction process. The variable created by subtracting Channel 2 from Channel 1 is called ‘Ch1-Ch2’. The atmospherically corrected
Channel 1 and 2 variables are called ‘Ch1W’ and ‘Ch2W’.
Determination of water column re ectance Rd for each Channel was based on
the following steps:
R # R¾ =R ­
d
d
c
where:
C
R
bias
D
A(1)
A(2)
R=
*(1/r2)*(1/cosh)
­
c
T (1)T (1) T (2)T (2)
0
1
0
1
A(l)=albedo for Channel l, where A=G*C+I (from Kidwell 1998), G and I
are calibration coeYcients, and C=count value (0–1023 ), T (l)=exp.{­ (t (l)/2+
0
r
t (l))/cosh}, T (l)=exp.{­ t (l)/2+t (l)}, (1/r2)={1+0.0167 cos j }2 and j=2
0
1
r
0
(3.1416 ) (D­ 3)/365, D is the Julian day, cos h=cosine of solar zenith angle at
scene centre, t (l)=Rayleigh optical depth for Channel l, t (l)=ozone and water
r
0
Figure 2. NOAA-14 AVHRR ( left panel) Ch1W re ectances (%) and (right panel) Ch2W
re ectances (%) on 26 April 1996 and the locations of water samples used to develop
the suspended sediment/solids models.
QuantiŽ cation of suspended sediments from satellite
3235
vapor absorption optical depth for Channel l, R =residual re ectance deŽ ned as
bias
R for a clear atmosphere over clear water near the area of interest.
c
The SeaWiFS data were atmospherically corrected using the SeaSpace TerascanTM
version of the NASA Goddard SEADAS code based primarily on Gordon and Wang
(1994). The NASA atmospheric correction software computes water leaving radiance
for channels 1 to 5 and total radiance for 6 to 8. As in the Stumpf and Pennock
(1989) technique, the assumption is made that water leaving radiance in the near IR
channels is zero and the water leaving radiance in channels 1–5 is lowered accordingly. Throughout the paper, NOAA AVHRR data is presented as re ectance measurements in percent and SeaWiFS data is presented as radiance measurements with
units of mW cmÕ 2 mmÕ 1 srÕ 1. These are the standard outputs of our TerascanTM
image processing software. The reader is referred to Froidefond et al. (1993) for a
discussion of conversion techniques from NOAA AVHRR re ectance to radiance
values.
3. Results
3.1. Model development
Correlation, linear and nonlinear regression techniques were used to quantify the
relationships between the NOAA AVHRR satellite measurements and the in situ
measurements of total suspended solids (TSS) and suspended sediments (SS). The
SeaWiFS sensor had not been launched at the time of these helicopter over- ights.
Linear correlations between TSS, SS and the three variables derived from NOAA
AVHRR re ectance data were investigated using the near-simultaneou s measurements of satellite and Ž eld data collected on 26 April 1996 (table 2) and on 21 June
1996 (table 3). The separate correlation analyses provided information on the
strength and reliability of the relationships among the variables. Separate analyses
for 26 and 27 March 1998 were not performed since only a single measurement was
available for each date.
Table 2.
TSS
SS
Ch1–Ch2
Ch1W
Ch2W
NOAA-14 correlation matrix of variables using the 26 April 26 1996 Ž eld data.
TSS
SS
Ch1–Ch2
Ch1W
Ch2W
1
0.999
0.677
0.833
0.939
1
0.680
0.836
0.941
1
0.967
0.862
1
0.963
1
Notes: Correlations are all signiŽ cant at the 0.01 level (Pearson Correlation), N=19.
Table 3.
TSS
SS
Ch1–Ch2
Ch1W
Ch2W
NOAA-14 correlation matrix of variables using the 21 June 1996 Ž eld data.
TSS
SS
Ch1–Ch2
Ch1W
Ch2W
1
0.996
0.862
0.920
0.952
1
0.863
0.921
0.955
1
0.986
0.862
1
0.935
1
Notes: Correlations are all signiŽ cant at the 0.01 level (Pearson Correlation), N=21.
3236
S. W. Myint and N. D. Walker
SigniŽ cant linear correlations (at the 0.01 level ) were determined between the
satellite re ectances, TSS and SS. The linear correlations were highest between Ch2W
and the TSS/SS on both dates. The correlations were relatively low between the
Ch1–Ch2 variable and the TSS/SS measurements. The linear correlations were
highest using the Ž eld measurements of 21 June 1996 when the Atchafalaya River
discharge was relatively high compared with the April 1996 dataset. River discharge
on 21 June was 11 558 m3 sÕ 1 and on 26 April it was 5920 m3 sÕ 1.
The ‘ground truth’ measurements of TSS and SS are shown in Ž gure 3 (see
Ž gure 2 for locations). TSS ranged from 5–140 mg lÕ 1 and SS ranged from 0 to
120 mg lÕ 1, indicating that inorganic sediments dominated the water constituents. A
similar graph was constructed for the satellite re ectance variables (Ž gure 4). The
Ch1W re ectance measurements were higher and exhibited a larger range in values
than those of Ch2W, a result that would be expected for this shorter wavelength
Channel. By comparing Ž gures 3 and 4, it is evident that the Ch1–Ch2 variable does
not follow the curves of TSS and SS above concentrations of about 60 mg lÕ 1. We
attribute this result to the failure of the atmospheric correction scheme at the higher
levels of TSS and SS.
Scatter-plots of satellite re ectances (independent variables) and the Ž eld measurements of suspended sediments (dependant variables) demonstrated that the relationships were non-linear using Ch1W re ectances and linear using Ch2W re ectances
(Ž gure 5). Previous algorithms developed using NOAA AVHRR Ch1 measurements
were also non-linear in form (Stumpf 1992, Walker 1996, Walker and Hammack
2000 ).
The SPSS software package was used to determine the best predictive models
for the estimation of suspended sediments and suspended solids from the re ectance
measurements. Both linear and nonlinear regression techniques were investigated.
Figure 3. Atchafalaya region Ž eld measurements of total suspended solids (TSS) and suspended sediments (SS): April/June 1996. Sample points 1–21 were obtained in June
and 22–44 were obtained in April.
QuantiŽ cation of suspended sediments from satellite
3237
Figure 4. Satellite re ectances (%) corresponding to the Ž eld measurements of Ž gure 3.
Sample points 1–21 were obtained in June and 22–44 were obtained in April.
The nonlinear models that were investigated included logarithmic, inverse, quadratic,
cubic, power, compound, S-curve, logistic, growth and exponential. Initially, models
were chosen based mainly on the coeYcient of determination (R2) values and Fratios. Table 4 depicts the best model results using Ch1W and Ch2W as independent
variables.
From the statistical results shown in table 4 and graphical data displays, the
most robust models were chosen for the estimation of both suspended sediments
(SS) and total suspended solids (TSS) using Ch1W and Ch2W. In Ž gure 6, the
predicted concentrations of SS computed from the two diVerent re ectance models
are shown. Table 5 corresponds to Ž gure 6, listing summary statistics for the chosen
models including R2, the standard error of the estimate (SEE), root mean square
error (RMS), F-ratio and the tabled F value.
The cubic models (SS1C, TSS1C) yielded the best estimates of suspended sediment
and suspended solids using Ch1W (table 5, Ž gure 6). The linear models (TSS2L,
SS2L) gave the best results using Ch2W (table 5, Ž gure 6).
The equations for the selected predictive models for TSS and SS are given below:
TSS=­
10.26+(14.8288 Ch1W)­
SS=­
10.746+(12.7179 Ch1W)­
TSS=­
8.1358+(20.0235 Ch2W)
SS=­
9.3438+(17.3593 Ch2W)
(3.1684 (Ch1W)2)+(0.2691 (Ch1W)3)
(2.7548 (Ch1W)2)+(0.2353 (Ch1W)3)
All of the chosen regression models were highly signiŽ cant as the F-ratio exceeded
the tabled F value by a large margin in each case. The model results using Ch1W
as input had the lowest SEE values (9.09 and 6.94 for TSS and SS, respectively).
Although the summary statistics were impressive for all the models, comparisons of
the two panels of Ž gure 6 clearly revealed that the Ch1W model estimations were
3238
S. W. Myint and N. D. Walker
Figure 5. Scatter-plot of NOAA AVHRR (top panel ) Ch1W re ectances (%) and (bottom
panel ) Ch2W re ectances (%) against suspended sediment (SS) concentrations
(mg lÕ 1): April/June 1996.
more similar to the Ž eld measurements, particularly at the higher levels of suspended
sediment.
In Ž gure 7, the spatial distribution of suspended sediments (SS) on 26 April 1996
produced using the Ch1W cubic model and the Ch2W linear model are shown.
Model-derived values on the shelf were observed to be similar. DiVerences in the
estimated values were detected within and close to the river channels where the
QuantiŽ cation of suspended sediments from satellite
Table 4.
3239
Linear and nonlinear estimations for TSS and SS using Ch1W and Ch2W.
TSS
Ch1W
SS
Ch1W
TSS
Ch2W
SS
Ch2W
R2
F-ratio
R2
F-ratio
R2
F-ratio
R2
F-ratio
LIN
QUA
CUB
EXP
POW
0.71
96.97
0.70
94.86
0.91
379.52
0.90
349.14
0.92
213.63
0.92
217.05
0.91
207.56
0.91
189.80
0.95
218.03
0.95
224.38
0.95
224.77
0.95
233.48
0.91
393.54
0.83
194.13
0.83
193.92
0.71
96.92
0.83
192.99
0.78
140.13
0.87
268.53
0.76
129.12
Notes: LIN=Linear model; QUA=Quadratic model; CUB=Cubic model; EXP=
Exponential model; and POW=Power model.
sediment levels were highest. The Ch2W model yielded higher levels of SS in these
regions. This was also observed in the actual Ch2 re ectance measurements (Ž gure 2).
It is most likely that the diVerence may be attributed to diVerential absorption by
non-sediment water constituents including gelbstoVe, marsh detrital material and
phytoplankto n (Wang et al. 1996). Absorption eVects would have impacted Ch1
more than Ch2, lowering re ectance values in Ch1. Additional information on water
constituents will be needed to clarify the observed diVerences between re ectance
patterns of the two channels.
3.2. Application and testing of the models
Water samples were obtained on 21 March 2001 in the Atchafalaya Bay region
and in the Mississippi River plume, coincident with a clear sky NOAA AVHRR
image. The Atchafalaya and Mississippi Rivers were in  ood with discharges
of 10 214 m3 sÕ 1 and 24 209 m3 sÕ 1 ( http://www.mvn.usace.army.mil), respectively.
Twelve surface samples were collected in the Atchafalaya region and six in the
Missisissippi Plume region (Locations, Ž gures 8 and 9). The satellite re ectance
values were extracted from individual pixels closest to the location of the water
samples. The models developed in previous secions were then applied to this
independent set of data.
The estimates of SS and TSS from the Ch1W and Ch2W models were better
than expected. RMS errors were lowest using the Ch2W linear algorithm and ranged
from 5.8–7.2 mg lÕ 1 (table 6). The Ch1W cubic algorithm produced concentration
estimates with RMS values of 10.6–10.9 mg lÕ 1. Bias values were negative using
Ch1W (­ 22 to ­ 24 mg lÕ 1) and positive using Ch2W (13–21 mg lÕ 1). The relatively
high bias values are attributable to the larger range of SS and TSS encountered on
21 March 2001.
The spatial distribution of surface suspended sediments hindcast using the NOAA
AVHRR re ectance models are shown for the Atchafalaya region (Ž gure 8) and for
the Mississippi delta region (Ž gure 9). The two diVerent algorithms yielded similar
values in Atchafalaya Bay and on the inner shelf. However, the estimated values
diverged in the river channels (Atchafalaya River and Wax Lake, west of the River),
as was also observed in Ž gure 7. The estimates were also diVerent in the far western
bay. A water sample obtained mid-stream in the Atchafalaya River in March 2001
3240
S. W. Myint and N. D. Walker
Figure 6. Model estimates of SS concentrations (mg lÕ 1) using the NOAA AVHRR (top
panel ) Ch1W cubic model and ( bottom panel ) Ch2W linear model, compared with
Ž eld measurements of SS. ‘SS1C’ and ‘SS2L’ refer to the suspended sediment predictions
using the Ch1W cubic model and the Ch2W linear model, respectively.
of 317 mg lÕ 1 indicates that the higher suspended sediment concentrations hindcast with Ch2W re ectances were more representative of Ž eld measurements. The
data suggest that Ch1W re ectances were too low, perhaps due to absorption
from organic sources such as marsh detritus, phytoplankton or yellow substances.
QuantiŽ cation of suspended sediments from satellite
Table 5.
Variable
Ch1W
Ch2W
3241
Summary statistics for the non linear cubic regression model using NOAA AVHRR
Ch1W and the linear regression model using NOAA AVHRR Ch2W data.
Models
R2
SEE
RMS
F-ratio
Tabled F
0.05
TSS
SS
TSS
SS
0.95
0.95
0.91
0.90
9.09
6.94
10.38
9.38
7.69
6.60
10.14
9.16
218.03
224.38
379.52
349.14
4.08
4.08
4.08
4.08
Figure 7. Model-estimated regional SS concentrations (mg lÕ 1) on 26 April 1996 using ( left
panel ) the Ch1W cubic model and (right panel ) the Ch2W linear model. Contour
intervals of 10, 25, 50 and 100 mg lÕ 1 are shown.
Figure 8. Model-estimated regional SS concentrations (mg lÕ 1) for the Atchafalaya region
on 21 March 2001 using ( left panel ) the Ch1W cubic model and (right panel ) the
Ch2W linear model. Contour intervals of 10, 25, 50 and 100 mg lÕ 1 are shown. The
region in which ‘ground truth’ data was obtained is denoted by a box.
S. W. Myint and N. D. Walker
3242
Figure 9. Model-estimated regional SS concentrations (mg lÕ 1) for the Mississippi delta
region on 21 March 2001 using (left panel ) the Ch1W cubic model and (right panel )
the Ch2W linear model. Contour intervals of 10, 25, 50 and 100 mg lÕ 1 are shown.
The location of ‘ground truth’ data collection is depicted using dots.
Table 6.
RMS and bias values for the estimation of TSS and SS from NOAA-14 AVHRR
Ch1W and Ch2W re ectance data on 21 March 2001.
RMS
TSS
SS
TSS
SS
Ch1W
Ch1W
Ch2W
Ch2W
10.94
10.57
7.18
5.81
­
­
Bias
22.19
24.02
21.07
13.44
Extensive chlorophyll blooms have been previously observed in Vermilion Bay during
Ž eld sampling, however, water samples were only obtained from Fourleague Bay in
March 2001. In the Mississippi delta region (Ž gure 9), the surface sediment distribution patterns were similar, but somewhat higher using the Ch2W linear algorithm.
The Ch2W re ectance data and predictions were observed to contain more noise
and more atmospheric contamination in the 21 March 2001 image, compared with
those of 1996.
The March 2001 data set provided the opportunity to improve the SS algorithms
due to the larger range of sediment concentrations encountered in the Ž eld (maximum
‘ground truth’ data was 209 mg lÕ 1 in March 2001). Scatter-plots of Ch1W and
Ch2W re ectances with SS are shown in Ž gure 10. The summary statistics for the
new SS regression models, developed by combining Ž eld and satellite measurements
from 1996 and 2001, are summarized in table 7. The cubic model using Ch1W yielded
a coeYcient of determination (R2) of 0.9. The linear model using Ch2W yielded a
coeYcient of determination of 0.93. The equations for these new models for SS is
given below.
SS=­
22.49+(22.76 Ch1W)­ (5.36 (Ch1W)2)+(0.42 (Ch1W)3)
SS=­
11.32+(16.78 Ch2W)
QuantiŽ cation of suspended sediments from satellite
3243
Figure 10. Scatter-plot of NOAA AVHRR (top panel ) Ch1W re ectances (%) and ( bottom
panel ) Ch2W re ectances (%) against SS concentrations (mg lÕ 1) using all Ž eld
measurements: 1996 –2001.
3.2. Development of suspended sediment models using SeaW iFS
A dedicated ‘ground truth’ collection survey was conducted on 26 April 2000 for
the Orbview-2 SeaWiFS sensor. This Ž eld experiment was undertaken in Barataria
Bay, west of the Mississippi River bird-foot delta and the adjacent coastal ocean.
Twenty-two water samples were obtained along a line that ran from 10 km oVshore
S. W. Myint and N. D. Walker
3244
Table 7.
Ch1W
Ch2W
Summary statistics of suspended sediment regression models using all Ž eld data
with NOAA AVHRR Ch1W and Ch2W variables.
R2
F-ratio
R2
F-ratio
LIN
QUA
CUB
EXP
POW
0.64
104.07
0.93
816.15
0.88
206.39
0.94
436.11
0.90
174.15
0.94
285.79
0.81
242.18
0.66
110.22
0.73
154.56
0.73
156.55
Notes: LIN=Linear model; QUA=Quadratic model; CUB=Cubic model; EXP=
Exponential model; and POW=Power model.
to the interior of the bay (Ž gure 11). Two additional samples were obtained to the
east of the line within the central bay region. The samples in the vicinity of the tidal
pass were excluded from the analysis as they were too close to land to be valid with
the nominal 1.1 km pixels of SeaWiFS. The correlation matrix obtained for the
radiance values, SS and TSS is shown in table 8. The concentration of suspended
sediments (SS) and total suspended solids (TSS) were found to be highly correlated
(0.998). Channel 6 (670 nm) was more highly correlated with SS and TSS than was
Channel 5 (555 nm). Correlation coeYcients between Ch 6 radiances, TSS and SS
were 0.85 (table 8). Correlation coeYcients between the Ch 5 and TSS/SS were only
0.46–0.47. The relatively high correlations with Channel 6 were not unexpected as
Figure 11. Orbview-2 SeaWiFS image of 26 April 2000 showing the normalized water
leaving radiance at ( left panel) 555 nm and (right panel ) normalized radiance at
670 nm (mW cmÕ 2*steradianÕ 1*micrometerÕ 1) with ‘ground truth’ station locations
superimposed.
QuantiŽ cation of suspended sediments from satellite
Table 8.
3245
Correlation matrix showing the relationship between TSS, SS, SeaWiFS 555 nm
and SeaWiFS 670 nm radiances on 26 April 2000.
TSS
SS
SeaWiFS-555
SeaWiFS-670
TSS
SS
SeaWiFS-555
SeaWiFS-670
1
0.998**
0.456*
0.845**
1
0.468*
0.853**
1
0.552*
1
Notes: ** Correlation is signiŽ cant at the 0.01 level (Pearson Correlation), N=20.
*Correlation is signiŽ cant at the 0.05 level (Pearson Correlation), N=20.
it is most similar to NOAA AVHRR Ch1 measurements. The relatively poor performance of Channel 5 may be partially attributed to over-correction for the atmosphere
by the standard NASA algorithm, although SS concentrations were not as high as
in the Atchafalaya region (maximum of 48 mg lÕ 1). Another source of error may
have been the use of water samples from the upper 0.5 m of the water column, if the
Ch5 radiances represented an integration over a deeper water column.
Statistical models were then developed using the radiance values of the 555 nm
and 670 nm Channels (table 9). For estimation of SS, the power model out-performed
other models. Only the model using the 670 nm Channel is considered a reasonable
approximation of the surface suspended sediment levels (Ž gure 12). The modelestimates of suspended sediment concentrations are compared with actual values in
Ž gure 13. The RMS error for this model was 8.24 mg lÕ 1.
4.
Discussion and conclusions
The technique of ‘ground truthing’ satellite data via helicopter and high speed
boat proved very eVective for obtaining water samples close in time to the satellite
re ectance measurements. This is of great importance in coastal regions where
re ectance patterns change rapidly as a result of water movements due to tidal and
wind-driven currents, mixing and re-suspension processes.
The near simultaneous acquisition of Ž eld measurements and regional synoptic
satellite data enabled the development and testing of statistical models for estimating
near-surface total suspended solids (TSS) and suspended sediments (SS) in surface
Table 9.
Summary statistics of the selected regression models using SeaWiFS 555 and 670 nm
Channel data.
TSS
555 nm
SS
555 nm
TSS
670 nm
SS
670 nm
R2
F-ratio
R2
F-ratio
R2
F-ratio
R2
F-ratio
LIN
QUA
CUB
EXP
POW
0.21
4.73
0.22
5.04
0.72
45.05
0.73
48.28
0.56
10.73
0.56
10.98
0.72
21.77
0.73
22.97
0.56
6.91
0.57
7.14
0.72
21.77
0.73
22.97
0.37
10.49
0.42
13.01
0.79
66.86
0.80
70.08
0.52
19.42
0.58
24.90
0.83
89.34
0.84
97.24
Notes: LIN=linear model; QUA=Quadratic model; CUB=Cubic model; EXP=
Exponential model; and POW=Power model.
3246
S. W. Myint and N. D. Walker
Figure 12. Scatter-plot of satellite-derived normalized radiances (670 nm) and SS
concentrations (mg lÕ 1) along transect line shown in Ž gure 11.
Figure 13. Model estimates of SS concentrations (mg lÕ 1) using the SeaWiFS 670 nm power
model compared with Ž eld measurements of SS.
waters along the river-in uenced Louisiana coastline. The high levels of suspended
sediments in near-shore regions required the development of a new atmospheric
correction technique for use with the NOAA AVHRR re ectance measurements.
The improved atmospheric correction enabled development of a robust algorithm
QuantiŽ cation of suspended sediments from satellite
3247
for estimating suspended sediments that was applicable to imagery acquired several
years later.
The NOAA AVHRR Ch1W non-linear model and the Ch2W linear model yielded
R2 values from 0.9–0.95 for SS and TSS. Subsequent testing with an independent
dataset yielded acceptable RMS values (<10 mg lÕ 1) in a region where sediments
ranged from near 0 to over 200 mg lÕ 1. The combination of all Ž eld measurements
enabled development of a more comprehensive SS model with R2 values for Ch1W
and Ch2W of 0.9 and greater. These results indicate that the main contributor to
re ectance was backscatter from inorganic suspended sediment in the water column.
The close Ž t between the Ž eld measurements and the model-estimates of TSS and
SS suggest that light attenuation from gelbstoVe, detritus, phytoplankton pigments,
etc. were minimal in the regions sampled.
The regional model estimates of SS using the Ch1W and Ch2W models were
found to be similar over large portions of the Atchafalaya and Mississippi plume
region. However, they did diVer in the Atchafalaya River and in an interior bay where
the near-IR Channel re ectances were higher and appeared to be more representative
of suspended sediment levels, based on the Ž eld data. The Ch1 re ectances may have
been lowered by absorption due to the presence of non-sediment constituents in the
water in these regions.
The suspended sediment models developed using SeaWiFS Channel 5 were not
as good as those developed using SeaWiFS Channel 6 or NOAA AVHRR Channels
1 and 2. SeaWiFS Channel 6 (670 nm) radiances were well correlated with SS and
TSS (R2 of 0.84 ). The superiority of Channel 6 may be attributed to its longer
wavelength and reduced eVects from pigments associated with organic substances
(CDOM or pigments from phytoplankto n species). The relatively low correlations
between Channel 5 radiances and the Ž eld measurements may have resulted from
over-correction for aerosols in the atmosphere, the shallow depth of the water
samples (<0.5 m), and absorption eVects from other pigments.
Important questions have arisen in the course of this study that merit further
investigation. ‘What is the ideal depth for the ‘‘ground truth’’ water sample?’ The
depth of the water sample may be critical to the relationships developed. It would
be informative to obtain water samples from several depths to improve attempts to
quantify suspended sediment concentrations. The second question is more general
in nature. ‘When comparing the sediment concentration of a 500 ml water sample
with a 1 km spatial average from satellite, how good of a relationship can one really
expect to obtain’
In this paper, using a modiŽ ed atmospheric correction technique for NOAA
AVHRR data, statistical models were developed that estimated SS and TSS in an
independent dataset with R2 values exceeding 0.9. Future model developments using
SeaWiFS and other higher resolution sensors such as Terra and Aqua MODIS will
also require modiŽ ed atmospheric correction techniques for turbid coastal regions.
References
Allison, M. A., Kineke, G. C., Gordon, E. S., and Goni, M. A., 2000, Development and
reworking of a seasonal  ood deposit on the inner continental shelf oV the Atchafalaya
River. Continental Shelf Research, 20, 2267–2294.
Curran, P. J., and Novo, E. M. M., 1988, The relationship between suspended sediment
concentration and remotely sensed spectral radiance: A review. Journal of Coastal
Research, 4, 351–368.
3248
S. W. Myint and N. D. Walker
Froidefond, J. M., Casting, P., Jouanneau, J. M., and Prud’homme, R., 1993, Method for
the quantiŽ cation of suspended sediments from AVHRR NOAA-11 satellite data.
International Journal of Remote Sensing, 14, 885–894.
Gagliardini, D. A., Karszenbaum, H., Legeckis, R., and Klemas, V., 1984, Application of
Landsat MSS, NOAA/TIROS AVHRR, and Nimbus CZCS to study the La Plata
River and its interaction with the ocean. Remote Sensing of Environment, 15, 21–36.
Goldman, C. R., Richards, R. C., Paerl, H. W., Wrigley, R. C., Oberbeck, V. R., and
Quaide, W. L., 1974, Limnological studies and remote sensing of the Upper Truckee
river sediment plume in Lake Tahoe, California. Remote Sensing of Environment,
3, 49–67.
Gordon, H. R., and Morel, A. Y., 1983, Remote assessment of ocean color for interpretation
of satellite visible imager: a review. In L ecture Notes on Coastal and Estuarine Studies,
edited by R. T. Barber, C. N. K. Mooers, M. J. Bowman, B. Zeitzschel (New York:
Springer-Verlag), 114 pp.
Gordon, H. R., and Wang, M., 1994, Retrieval of water-leaving radiance and aerosol optical
thickness over the oceans with SeaWiFS: a preliminary algorithm. Applied Optics,
33, 433–452.
Huh, O. K., Roberts, H. H., and Rouse, L. J. Jr., 1991, Fine grain sediment transport and
deposition in the Atchafalaya and Chenier Plain sedimentary system. American Society
of Civil Engineers Proceedings f rom Coastal Sediments ’91. Seattle, Washington,
pp. 817–830.
Huh, O. K., Moeller, C. C., Merzel, W. P, Rouse, L. J. Jr., and Robert, H. H., 1996,
Remote sensing of coastal and estuarine waters: A method of multispectral water-type
analysis. Journal of Coastal Research, 12, 984–995.
Huh, O. K., Moeller, C. C., and Walker, N. D., 2001, Sedimentation along the eastern
Chenier Plain coast: down drift impact of a delta complex shift. Journal of Coastal
Research, 17, 72–81.
Kidwell, K. B., 1998, NOAA Polar Orbiter Data User’s Guide, US Dept. of Commerce,
NOAA, NESDIS, National Climatic Data Center, Suitland, MD.
Kondratyev, K. Y., and Filatov, N. N., 1999, L imnology and Remote Sensing. A
Contemporary Approach (Chichester: UK Springer-Praxis), 406 pp.
Maul, G. A., 1985, Introduction to Satellite Oceanograph y (Boston: Martinus).
Milliman, J. D., and Meade, R. H., 1983, World-wide delivery of river sediment to the ocean.
Journal of Geology, 91, 1–21.
Moeller, C. C., Huh, O. K., Roberts, H. H., Gumley, L. E., and Menzel, W. P., 1993,
Response of Louisiana Coastal Environments to a Cold Front Passage. Journal of
Coastal Research, 9, 435–447.
Moore, G. K., 1977, Satellite surveillance of physical water quality characteristics. Proceedings
of the 12th International Symposium on Remote Sensing of Environment, University of
Michigan, Ann Arbor, Environmental Research Institute of Michigan, pp. 445–461.
Mossa, J., 1990, Discharge-suspended sediment relationships in the Mississippi-Atchafalaya
River System, Louisiana, PhD thesis, Louisiana State University, 179 pp.
Mossa, J., and Roberts, H. H., 1990, Synergism of riverine and winter storm related sediment
transport processing in Louisiana coastal wetlands. Gulf Coast Association of
Geological Societies, 40, 635–642.
Muller-Karger, F. E., McClain, C. R., and Richardson, P. L., 1988, The dispersal of the
Amazon’s water. Nature, 333, 56–58.
Novo, E. M. M., Hansom, J. D., and Currain, P. J., 1989, The eVect of sediment type on
the relationship between re ectance and suspended sediment concentration.
International Journal of Remote Sensing, 10, 1283–1289.
O’Reilly, J. E., Maritorena, S., Mitchell, B. G., Siegel, D. A.,, Carder, K. L., Garver,
S. A., Kahru, M., and McClain, C., 1998, Ocean color chlorophyll algorithms for
SeaWiFS. Journal of Geophysical Research, 103, 24 937–24 953.
Rabalais, N. N, Turner, R. E., and Wiseman, W. J., Jr, and Boesch, D. F., 1991, A brief
summary of hypoxia on the northern Gulf of Mexico continental shelf: 1985–1988. In
Modern and Ancient Continental Shelf Anoxia, edited by P. V. Tyson and T. H. Pearson
(London: Geological Society Special Publication), pp. 35–47.
QuantiŽ cation of suspended sediments from satellite
3249
Rouse, L. J., and Coleman, J. M., 1976, Circulation observations in the Louisiana Bight
using LANDSAT imagery. Remote Sensing of Environment, 40, 635–642.
Science Applications International Corporation, 1989, Gulf of Mexico Physical Oceanography Program, Final Report: Year 5. Volume II: Technical Report. OCS Report/
MMS-89–0068, U.S. Dept. of the Interior, Minerals Management Service, Gulf of
Mexico OCS Regional OYce, New Orleans, LA, 333 pp.
Stumpf, R. P., 1988, Remote Sensing of suspended sediments in estuaries using atmospheric
and compositional corrections to AVHRR data. Proceeding of the 21st International
Symposium on Remote Sensing of Environment, Ann Arbor, Michigan. pp. 205–222.
Stumpf, R. P., 1992, Remote Sensing of Water clarity and suspended sediment in Coastal
Waters. Proceedings of the 1st T hematic Conference on Remote Sensing for Marine and
Coastal Environments, New Orleans, L ouisiana, 15–17 June, 1992. 1930, pp. 293–305.
Stumpf, R. P., and Pennock, J. R., 1989, Calibration of a general optical equation for remote
sensing of suspended sediments in a moderately turbid estuary. Journal of Geophysical
Research, 94, 14 363–14 371.
US Geological Survey., 1987, Methods for collection and analysis of aquatic biological and
microbiological samples, Chapter A4. In Techniques of water resources investigations
of the U.S. Geological Survey, edited by L. J. Britton and P. E. Greesonm (Washington,
DC: US Geological Survey), 127–130.
Walker, N. D., 1996, Satellite assessment of Mississippi River plume variability: causes and
predictability. Remote Sensing of Environment, 58, 21–35.
Walker, N. D., and Hammack, A., 2000, Impacts of Winter Storms on Circulation and
Sediment Transport: Atchafalaya-Vermilion Bay Region, Louisiana. Journal of Coastal
Research, 16, 996–1010.
Walker, N. D., 2001, Tropical storm and hurricane wind eVects on water level, salinity
and sediment transport in the river-in uenced Atchafalaya-Vermilion Bay system,
Louisiana, USA, 24, 498–508.
Wang, M., Lyzengo, D. R., and Klemas, V. V., 1996, Measurement of Optical Properties
in the Delaware Estuary. Journal of Coastal Research, 12, 211–228.