There exists a common perception that chlorophyll a concentrations in tidal coastal waters are unsuitable to be captured by remote sensing techniques because of high water turbidity. In this study, we use band index measurements to separate active chlorophyll pigments from other constituents in the water. Published single- and multiband spectral indices are used to establish a relationship between algal chlorophyll concentration and reflectance data. We find an index which is suitable to map chlorophyll gradients in the impoundments, ditches, and associated waterways of the Hackensack Meadowlands (NJ, USA). The resulting images clearly depict the spatial distribution of plant pigments and their relationship with the biological conditions of the waters in the estuary. Since these biological conditions are often determined by land usage, the methods in this paper provide a simple tool to address water quality management issues in fragmented urban estuaries.
1. Introduction
Since
tidal coastal regions often contain suspended solids and dissolved organic
matter which confounds the existence of chlorophyll a (Chl a), there exists a
common perception that Chl a concentrations in such regions are unsuitable to
be mapped by remote sensing [1]. Light reflected off a body of water represents
a weighted sampling of contributions from water, suspended solids, and
chlorophyll [2]. It remains a challenge to develop optical measurements that
can separate photons absorbed by active chlorophyll pigments from photons
absorbed by other constituents. Currently, available narrow band airborne
spectrometers such as Hyperion, AVRIS, AISA, and CASI offer the unique
possibility to separate the effects of different constituents using remote
sensing techniques [3]. This separation would aid scientists in a variety of
ways; indeed, levels of algae, Chl a, and plant pigments have been used as
indicators of primary productivity and have been critical to the modeling and
understanding of the global carbon cycle [4].
Prior
work has focused on identifying portions of the spectrum, which are able to
accurately predict the concentrations of various constituents in water. For
example, it has been shown that the light absorption of gelbstoff and detritus
does not vary greatly and is confined to the blue region of the spectrum; therefore,
this absorption can be easily modeled and separated from light absorbed by
phytoplankton [5]. The in situ reflectance of different
water types has also been measured; for instance, it has been shown that the
reflectance ranging from 650–750 nm is a good
predictor of Chl a [3, 6, 7]. Creating effective spectral indices from
reflectance measurements allows for the large-scale discrimination of Chl a
concentrations in bodies of water. Although these spectral indices are
developed for use with reflectance measurements, in turbid waters, optical
signals correlated with Chl a are often masked by signals from detritus or total
suspended solids (TSSs) [8]. It is
well documented that in the presence of a strong absorption background,
measuring the rate of change of spectra with respect to wavelength amplifies essential
details in the spectra [9]. In particular, by using various manipulations of
first and second derivatives, it is possible to derive expressions which show
an excellent correlation with Chl a concentrations in turbid waters [2, 10].
The
purpose of this study is to test different optical measurements against actual
Chl a concentrations from shallow coastal waters by using the derivatives of
reflectance. We use our findings to classify Chl a gradients from images
captured by aircraft-mounted hyperspectral remote sensors; this allows us to delineate
those natural and human forcing functions which affect the biological conditions
of water in the estuary.
2. Methods
The
study site is located in the New Jersey Meadowlands District along the lower Hackensack River in Lyndhurst, NJ, USA
(Supplementary Figure 1 in Supplementary Materials available online at doi:10.1155/2008/146217). The impoundments have tidal influence and at high tide are no more than
four to five feet deep. The average salinity in these waters is 5–8 ppm, turbidity
varies around a baseline reading of 10 FTUs, and TSS averages at around 25 mg/L.
A field campaign was conducted to collect reflectance spectra (FieldSpec, Colo, USA Pro Full Range Spectroradiometer from
Analytical Spectral Devices) along transects starting at the edge of the
impoundment and ending twenty meters inshore with sampling points every two
meters. Immediately after each spectral measurement, a half liter water sample
was drawn from the first five inches of the surface, where the reflectance
measurement had taken place. Samples were stored on ice for 24 hours and
analyzed in the laboratory for total Chl a concentration using acetone
extraction [11] and for TSS concentration using a gravimetric method [12].
Each component of the spectral reflectance is
represented by a different Nth order polynomial. Using the Lagrange interpolation
polynomial [13], we considered the zero-, first-, and second-order derivatives
(curves) for clear water, turbid water, and algal chlorophyll in turbid water. The
spectral effects of water reflection are eliminated by the first derivative
(first-order effect) [2]. Similarly, spectral effects from turbidity are removed
by a second differentiation of the polynomial (second-order effect). Mathematica, V5.2 (Wolfram
Research, Oxfordshire, UK, 2006) was used to
calculate the first- and second-order derivatives from the raw data using a
seven-point numerical differentiation technique derived from the
Lagrange interpolation
polynomial [2]. Since differentiation tends to enhance the magnitude of noise
in the spectra, the Savitzky-Goley algorithm [14]
was applied to smooth the data prior to calculating derivatives. To determine
the index which best maps our study site, we calculate the coefficients between
various indices and Chl a/TSS values for each transect (using SPSS 11.5, III, USA, 2005). The significance
of the relationship is determined by using the analysis of variance test (ANOVA). We consider the
zero-, first-, and second-order derivatives for the following published
indices: R720 (for TSS estimation)
and R660-R695 (for Chlorophyll a estimation) from Goodin et al. [2], R680/R670
from Szekielda et al. [7], and (AVE(R650+R700)R675)/(AVE(R440+R550)) from Hladik
[15]. Since our spectral measurements
started at 450 nm, we modified Hladik’s index, replacing the reflectance values
at 440 nm with ones measured at 450 nm.
Hyperspectral
imagery of the entire lower Hackensack River
(8.288 hectares) was
collected on October
5, 2004 using the Airborne Imaging Spectroradiometer for
Applications (AISA) [16]. We utilize a mask to select only pixels that are
associated with open water; further, we ensure that pixels used to estimate Chl
a concentration were free of floating vascular vegetation and did not include
areas of exposed mud flats. However, brightness differences between flight
lines and shadows remain a significant image-related error. It is our
assumption that these errors are associated with flight line direction. Hladik’s
index, which showed a strong correlation to Chl a concentration for all
transects, was selected to create gradient images for the estuary. The index
was entered in ENVI’s BandMath function which results in a single-band image
where each pixel acquires the index value. Finally, trophic state classes were
assigned using Chl a concentration as follows: oligotrophic < 2.6 μg/L, mesotrophic
2.6–7.3 μg/L,
eutrophic 7.3–20 μg/L, and
hypereutrophic > 20 μg/L [17].
3. Results and Discussion
Chl a concentration along our
transects varied from 0.2 μg/L to
35 μg/L, similar to what was observed for the
fall season in other studies [18, 19]. According to published trophic scales in
the fall season, these waters are oligotrophic or mesotrophic and have low to
moderate productivity with intermediate to low clarity [17]. The results of the
field study show that TSS remains
almost constant (r2< 0.4) along the entire length of the transects,
while Chl a concentration increased significantly (r2> 0.85; P< .05) from near shore
to inshore (see Supplementary Figure 2). Spectra collected from 15 cm above the water
surface display the typical peaks and troughs associated with living plant
pigments (see Supplementary Figure 3).
The
relationship between the indices and water constituents was stronger in the
overall model (N=27, after removal of three outliers) as compared to each
individual transect (Supplementary Table 1). Our field data not only agreed with several band indices from the
literature but also conflicted with other published claims. For example, Goodin
et al. [2] suggested that the first derivative of R720 may estimate TSS, but our data showed no relationship between
the first derivative of R720 and TSS.
It is important to note that this is consistent with claims within [2] since
the wavelengths chosen by Goodin et al. are for comparison purposes to evaluate
the performance of the derivative method. In particular, these wavelengths were
not intended to be predictive indicators. On the other hand, our data clearly
verified that the index of Szekielda et al. (R680)/(670) is a good estimator of
Chl a showing a strong relationship with plant pigment in all transects as well
as in the overall model. Additionally, Hladik’s index showed a very strong
relationship between the raw/first derivative and Chl a concentration; this
relationship held true for both individual transects and for the overall model
(see Figure 1 Supplementary Table 1).
Chlorophyll
a concentrations graphed against Hladik’s index [15]. The figure depicts the
coefficient of correlation for each individual transect (T1, T2, and T3) and
for the overall model. The statistical significance of the relationship is
calculated using the ANOVA test and details can be found
in supplementary Table 1.
Based
on our field measurement results, Hladik’s index was selected for mapping the Chl
a concentration in the District’s hydrological network, the results of which
are shown in Figure 2. First, we used Hladik’s overall model, with r2=0.853 to
classify all open water pixels. Since
this model is not as accurate for concentrations lower than
2.6 μg/L, we
selected all pixels that were classified by the overall model as having 2.6 μg/L
or less of Chl a and reclassified them using Hladik’s T3 index model, with r2=0.741. Figures
2(a)–2(d) show the
results of the Chl a image classification. The entire open water surfaces in
the District are shown in Figure
2(a). Tide restricted impoundments to the
south west show the greatest productivity; this is in agreement with field
observations in these areas which are characterized by chronic algal scum and
macrophyte problems. On the other hand, the main channel of the river is mainly
oligotrophic with mesotrophic areas to the north, which are connected to
eutrophic tributaries that have their origins in urbanized areas. Figure 2(b) presents
a tributary showing pockets of hypereutrophic waters along its course. The
largest of these pockets to the north is clearly linked to several industrial
facilities. As the tributary meanders through wetlands and away from developed
sites, it becomes less productive emphasizing the role of wetlands in improving
water clarity. It also shows the trophic status increasing upstream as it
connects with urban development. The hydraulic connectivity beneath an
abandoned railroad connecting a tide-restricted impoundment with hypereutrophic
waters to an oligotrophic water body off the main river channel is captured by
Figure 2(c). Finally,
Figure 2(d) shows a network of ditches and channels
holding stagnant waters, which over time become a breeding ground for
mosquitoes.
Open waters of the New Jersey Meadowlands District
showing (a) the entire hydrological network of the District, (b) the higher
trophic state of tributaries as compared to the main channel, (c) hydraulic connectivity
beneath an abandoned railroad line, and (d) stagnant water in a poorly draining
ditch network.
4. Conclusion
Our results show that there are band indices which
effectively capture plant pigment concentrations in highly turbid waters. Additionally,
our study depicts the ability, in some cases [7], for the reflectance rate of
change expressed through a mathematical derivative to further separate the
effects of turbidity from those of Chl a; this is an essential aspect of
mapping turbid waters since it strengthens the index-plant pigment
relationship. We find that Hladik’s index shows the strongest relationship with
Chl a when all data points (N=27) are taken into account. This index captures the
area of variability for light absorption and reflection in the red and NIR;
further, the index is normalized with respect to the dissolved organic fraction
in the blue segment of the spectra. Our field transects cover a representative
chlorophyll range for the entire estuary and regressions result in a highly
significant overall model. The resulting images clearly showed Chl a gradients
as represented by the trophic state. Thus, our method allows for an inspection
of the Chl a concentration in relation to human land use and provides a clear
link between the different manmade
forcing functions that are driving the biological conditions of the waters in
the estuary.
Acknowledgments
The
technical assistance of Dr. Jian-sheng Yang and Dr. Ruji
Yao is gratefully acknowledged. The authors would like to thank Robert Saverino
for his editorial work. This project was funded by the Meadowlands Environmental Research Institute and the Herchel Smith Research Fellowship.
Del CastilloC. E.CobleP. G.ConmyR. N.Müller-KargerF. E.VanderbloemenL.VargoG. A.Multispectral in situ measurements of organic matter and chlorophyll fluorescence in seawater: documenting the intrusion of the Mississippi River plume in the West Florida Shelf200146718361843GoodinD. G.HanL.FraserR. N.RundquistD. C.StebbinsW. A.SchallesJ. F.Analysis of suspended solids in water using remotely sensed high resolution derivative spectra1993594505510QuibellG.The effect of suspended sediment on reflectance from freshwater algae199112117718210.1080/01431169108929642BehrenfeldM. J.FalkowskiP. G.A consumer's guide to phytoplankton primary productivity models199742714791491RoeslerC. S.PerryM. J.CarderK. L.Modeling in situ phytoplankton absorption from total absorption spectra in productive inland marine waters198934815101523GitelsonA.SzilagyiF.MittenzweyK.-H.Improving quantitative remote sensing for monitoring of inland water quality19932771185119410.1016/0043-1354(93)90010-FSzekieldaK.-H.GoblerC.GrossB.MosharyF.AhmedS.Spectral reflectance measurements of estuarine waters20035329810210.1007/s10236-003-0027-xDemetriades-ShahT. H.StevenM. D.ClarkJ. A.High resolution derivative spectra in remote sensing1990331556410.1016/0034-4257(90)90055-QMartinA. E.Difference and derivative spectra1957180457923123310.1038/180231a0RundquistD. C.HanL.SchallesJ. F.PeakeJ. S.Remote measurement of algal chlorophyll in surface waters: the case for the first derivative of reflectance near 690 nm1996622195200WebbD. J.BurnisonB. K.TrimbeeA. M.PrepasE. E.Comparison of chlorophyll a extractions with ethanol and dimethyl sulfoxide/acetone, and a concern about spectrophotometric phaeopigment correction199249112331233610.1139/f92-256EPA Method 160.2EPA method 160.2 for the analysis of samples for Total Suspended Solids2002, http://www.epa.gov/region09/qa/pdfs/160_2.pdfBarnettS.CroninT. M.1986London, UKLongmanTsaiF.PhilpotW.Derivative analysis of hyperspectral data for detecting spectralfeatures1997312431245HladikC. M.2004Omaha, Neb, USACreighton UniversityArtigasF. J.Francisco.Artigas@njmeadowlands.govYangJ. S.Hyperspectral remote sensing of marsh species and plant vigour gradient in the New Jersey Meadowlands200526235209522010.1080/01431160500218952CarlsonR. E.A trophic state index for lakes1977222361369USGSNatural Water Information System1994, http://nwis.waterdata.usgs.gov/nwisMERIMeadowlands Environmental Research Institute Scientific Data2006, http://merigis.njmeadowlands.gov/vdv/Index.php