Spectra of a shallow sea—unmixing for class identification and monitoring of coastal waters
ABSTRACT Ocean colour-based monitoring of water masses is a promising alternative to monitoring concentrations in heterogeneous coastal
seas. Fuzzy methods, such as spectral unmixing, are especially well suited for recognition of water masses from their remote
sensing reflectances. However, such models have not yet been applied for water classification and monitoring. In this study,
a fully constrained endmember model with simulated endmembers was developed for water class identification in the shallow
Wadden Sea and adjacent German Bight. Its performance was examined on in situ measured reflectances and on MERIS satellite
data. Water classification by means of unmixing reflectance spectra proved to be successful. When the endmember model was
applied to MERIS data, it was able to visualise well-known spatial, tidal, seasonal, and wind-related variations in optical
properties in the heterogeneous Wadden Sea. Analyses show that the method is insensitive to small changes in endmembers. Therefore,
it can be applied in similar coastal areas. For use in open ocean situations or coastal or inland waters with other specific
inherent optical properties, re-simulation of the endmember spectra with local optical properties is required. However, such
an adaptation requires only a limited number of local in situ measurements.
KeywordsEndmember–Reflectance–MERIS–Water mass–Unmixing
-
Citations (0)
-
Cited In (0)
Page 1
”Dense Water” and “Fluid Sand”
”Dense water” and “Fluid Sand”
Annelies Hommersom - IVM-
Cover :
Background: MERIS image
May 4 2006
provided by the European Space Agency
Left: TriOS sensors
Center: reseach vessel Navicula (NIOZ)
Right: AC9 instrument in turbid water
Optical properties
and methods for remote sensing of
the extremely turbid Wadden Sea.
Page 2
“Dense Water” and “Fluid Sand”
Optical properties and methods for remote sensing of the
extremely turbid Wadden Sea
Annelies Hommersom
Page 3
“Dense Wat
Optical prop
Ph.D. thesis
In Dutch:
“Dik Water”
Optische eig
Proefschrift
ISBN: 97890
© 2010 Ann
This work w
and at the R
This work w
Printed on F
ter” and “Flu
perties and m
s, Vrije Unive
” en “Vloeiba
genschappen
t, Vrije Unive
086594610
nelies Homm
was carried o
Royal Nether
was financed
FSC paper by
uid Sand”
methods for
ersiteit, Amst
aar Zand”
n en method
ersiteit, Amst
mersom
ut at the Ins
rlands Institu
by NWO/SR
y PrintPartne
remote sens
terdam
den van remo
terdam
titute for En
ute for Sea R
RON Program
ers Ipskamp,
sing of the e
ote sensing v
vironmental
Research (NIO
mme Bureau
Enschede, T
xtremely tur
voor de extre
l Studies (IVM
OZ), Texel.
Space Resea
The Netherla
rbid Wadden
eem troebel
M), Vrije Uni
arch.
ands
n Sea
e Waddenze
versiteit, Am
ee
msterdam
Page 4
VRIJE UNIVERSITEIT
“Dense Water” and “Fluid Sand”
Optical properties and methods for remote sensing of the extremely turbid Wadden Sea
ACADEMISCH PROEFSCHRIFT
ter verkrijging van de graad Doctor aan
de Vrije Universiteit Amsterdam,
op gezag van de rector magnificus
prof.dr. L.M. Bouter,
in het openbaar te verdedigen
ten overstaan van de promotiecommissie
van de faculteit de Aard‐ en Levenswetenschappen
op maandag 28 juni 2010 om 15.45 uur
in de aula van de universiteit,
De Boelelaan 1105
door
Annelies Hommersom
geboren te Hengelo (ov)
Page 5
promotor:
copromotor:
prof.dr. J. de Boer
dr. S.W.M. Peters
Page 6
Table of contents
Chapter 1
Chapter 2
Chapter 3
Chapter 4
Chapter 5
Chapter 6
Chapter 7
Annex 1
Annex 2
Introduction 7
A review on substances and processes relevant for optical remote sensing
of extremely turbid marine areas, with a focus on the Wadden Sea
17
Spatial and temporal variability in bio‐optical properties of the Wadden Sea 39
Performance of the regionally and locally calibrated algorithm HYDROPT in a
heterogeneous coastal area
57
Tracing Wadden Sea water masses with an inverse bio‐optical and an
endmember model
85
Spectra of a shallow sea: unmixing for coastal water class identification and
monitoring
97
Synthesis and outlook
115
References 123
Summary and samenvatting 141
Acknowledgements and dankbetuiging 155
Glossary of terms and descriptions used in remote sensing of water quality 159
Abbreviations, acronyms, and symbols 163
Publications 167
Page 7
Page 8
Chapter 1
Introduction
Page 9
Chapter 1
8
1 Introduction
1.1 Optical remote sensing of water quality
Remote sensing means “detecting from a distance”. A sensor used for detection can be hand‐held,
employed from an airplane (air‐borne remote sensing) or be part of a satellite (space‐born remote
sensing) and the technique can be passive or active. Optical remote sensing, as used in this thesis, is
passive: it uses reflected sun light in the visible part of the spectrum (~ 300‐800 nm, Figure 1.1). Optical
remote sensing techniques can for example be used to examine land use changes (Valbuena et al.,
2009), to monitor seasonal dynamics in vegetation (Zuritta‐Milla et al., 2009), to calculate carbon
accumulation in peat lands (Scheapman‐Strub et al., 2008), to detect erosion (Vrieling et al., 2007), or to
monitor water quality (Robinson et al., 2008). Active remote sensing includes radar (wavelengths <1 mm
to 1 m, for example Synthetic Aperture Radar: SAR) and laser (infrared, visible, ultraviolet, for example
LIDAR), where signals are sent to earth and the subsequently reflected (radar) or induced (laser) signal is
detected. These techniques can be used to detect 3‐dimensional features of objects, waves, and land‐
water boundaries. This thesis addresses optical remote sensing of water quality, which is commonly
referred to as ocean colour remote sensing (although lakes and rivers require similar approaches).
Figure 1.1. Spectrum of visible light.
Ocean colour remote sensing dates back to the 19th century, when explorers noticed large differences in
water colour (e.g. Figure 1.2) between coastal waters and oceans (Wernand, 2010). For example Joseph
Luksch determined 367 observations of the water colour, measured with the Forel Ule colour scale
(Luksch, 1901; Wernand, 2010) during the expedition of the S.M.S. Pola in the Mediterranean and
Aegean Sea (1890‐1894) and during the Austro‐Hungarian Expedition to the Red Sea (1895‐1898). Ocean
colour remote sensing became popular in the 1960’s and 1970’s with increasing awareness of water
quality. Secchi discs and air borne remote sensing were introduced to study turbidity and water colour.
The first ocean colour sensor (the Coastal Zone Scanner, CZSC) was launched in 1978 and soon after,
other satellite sensors for Ocean Colour research became available.
Figure 1.2. Various water colours seen in the Wadden Sea
To use airborne or space borne optical remote sensing data of earth surfaces, absorption and scattering
by the atmosphere must first be removed from the remote sensing signal. The atmosphere and clouds
Page 10
Introduction
9
reduce the sunlight penetration to the earth surface to ~55% (Pidwirny, 2006). A large portion of
reflected light is again absorbed and scattered on the way back to the satellite. Correcting signals for
this atmospheric influence is a research field of its own. In remote sensing of land, the light reflected by
the (land) surface is the variable that tells something about this surface. In ocean colour remote sensing,
interest lies with substances in the water column (ray 5 in Figure 1.3); reflectance from the (water)
surface (ray 6 in Figure 1.3) obscures this signal. Also the water itself can influence the reflected signal
and, at shallow or clear enough locations, so can the bottom (Figure 1.3). Consequently, only a small
portion of the light received by the sensor contains information on the content of the water and can be
used for water quality monitoring. Therefore, remote sensing of water requires different techniques
than remote sensing of land.
Figure 1.3. The most common routes of sunlight on its way to a satellite above the sea. Ray 1 is absorbed by the
atmosphere and never reaches the water surface and the sensor. Rays 2 and 3 are respectively absorbed and
scattered by the water or its contents and so never reach the sensor. Rays 4, 5, 6 and 7 reach the sensor. However,
only ray 5 is interesting for water quality monitoring. This ray is partly absorbed by water or its contents, but enough
is scattered over an angle > 90˚ (backscattered) to reach the sensor. Rays 4, 6, and 7 are, respectively, reflected by
the sea floor, by the water surface and scattered by the atmosphere. Combinations of these routes can also occur.
Substances in the water column that can be detected from optical remote sensing are: the water itself,
pigments, suspended particulate matter (SPM), and coloured dissolved organic matter (CDOM). The
pigment most abundant in marine phytoplankton, chlorophyll‐a (Chl‐a), is usually taken to represent the
pigments. SPM, Chl‐a and CDOM are (indirect) indicators for other water quality parameters such as
nutrient concentration, river runoff, resuspension or decay. As SPM, Chl‐a and CDOM have a significant
influence on the water colour, these three substances are called optically active substances in this
thesis. In the absence of optically active substances, water mainly absorbs red light, while the size of the
water molecules leads to scattering of blue light. The colour of pure water is therefore blue. Water
absorbs more than it scatters, and so very deep water bodies look dark from above. Chl‐a absorbs blue
Page 11
Chapter 1
10
and red light, turning water green in high concentration (e.g. the central picture in Figure 1.3). Inorganic
SPM scatters light efficiently and can therefore lead to a high reflectance. This high reflection in
combination with absorption of organic SPM, mostly in the blue wavelengths, results in bright water
with often high red reflectance. CDOM absorption spectra are similar to those of organic particles, but
CDOM does not scatter. Therefore water with high CDOM concentrations, for example the Baltic Sea,
looks yellow‐brownish or even blackish (Berthon et al., 2008). The left picture in Figure 1.2 contains
some Chl‐a, SPM and CDOM, leading to greyish water; the picture on the right contains much SPM and
CDOM, which makes the water look brown‐red.
Figure 1.4. Reflectance spectra (Rrs) of different water colours that can be found in the Wadden Sea.
The colour of water is quantified by means of reflectance spectra, usually given per wavelength (λ) over
the range of visible light (Figure 1.4). Reflectance can be measured with hand‐held spectrometers, from
an air plane, or by satellite. When the distinct absorption and scattering properties of Chl‐a, SPM and
CDOM (Figure 1.5) are known, their abundances can theoretically be calculated from the reflectance.
Various types of algorithms (for example ratio algorithms, neural networks and inverse bio‐optical
models) are available to derive high quality results from remote sensing data of the open ocean.
Especially in coastal zones, Chl‐a, SPM and CDOM occur in mixtures, which complicates the derivation of
their concentrations from reflectances. The local specific absorption and scattering properties of these
substances (the specific inherent optical properties, SIOPs) (Figure 1.5) may also vary. For example, SPM
primarily scatters when it consists of sand, but might have a strong absorption when it mainly consists of
organic particles or clay. Pigment absorption varies for different phytoplankton species, while the water
source (e.g. North Sea water, river discharge, land‐runoff) determines the absorption properties of
CDOM. The levels of scattering and absorption, as well as the spectral shape of the SIOPs can vary.
Deriving concentrations from remote sensing data is therefore more complex in coastal areas than in
the open ocean and the results are often not yet precise enough for water quality monitoring.
High SPM concentrations, caused by resuspension, can mask the effects of Chl‐a and CDOM in shallow
coastal areas. Another problem is that reflectance of the bottom can influence the derived signal at
clear and shallow locations, while pixels at the coastline often contain a mixture of land and water. In
airborne and space‐born remote sensing high reflection of vegetation at the coast can (due to the height
of the sensor) lead to noise in the near‐by water reflectances, which is called the adjacency effect
(Santer and Schmechtig, 2000).
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
350450550650750
Rrs
λ (nm)
Page 12
Introduction
11
Despite the difficulties mentioned above, remote sensing is an interesting possibility for water quality
monitoring of coastal zones. The Water Framework Directive regulations from the European Union force
member states to monitor all their coastal areas (Environment Directorate‐General of the European
Commission, 2000). Monitoring is an important means to support maintenance of ecological and
economic values of coastal zones, which are often highly populated areas (Robinson et al., 2008).
Monitoring of all coastal waters by ship poses severe logistic problems, apart from being costly and time
consuming. Remote sensing offers an attractive alternative because of its high spatial resolution (e.g.
the image at the cover). While remote sensing cannot detect all properties for water quality assessment
recommended by water authorities, such as concentrations of PCB’s ‐the substances held responsible
for the in decline in seal populations in the Wadden Sea in the late 1980’s (Reijnders, 1986)‐ it provides
information on transparency (Kd), Chl‐a, SPM (important for the visual acuity of the seal (Weiffen et al.,
2006)) and CDOM. These properties are highly relevant for the Water Framework Directive (Peeters et
al., 2009), while manual monitoring can always augment remote sensing in the case of unexpected
changes. Remote sensing only derives information from the upper layers of a water body, but, when
space borne remote sensing is used, a high temporal resolution (Fanton d’Andon et al., 2005) is
obtained.
Figure 1.5. Example spectral shapes of the specific absorption (a) and scattering (b) by water (m‐1), chlorophyll‐a (m2
mg‐1), SPM (m2 g‐1) and CDOM (‐) at MERIS wavelengths (Table 1.1). Black lines are related to the vertical axis on the
left, gray lines to the vertical axis on the right.
0
0.02
0.04
0.06
0.08
0.1
0
0.1
0.2
0.3
0.4
400450500550600650700
absorption or scattering
absorption or scattering
λ (nm)
aWatera*CDOMb*SPMa*Phytoplankton a*SPMbWater
Page 13
Chapter 1
12
1.2 Research questions
This thesis studies the possibilities for water quality and water mass monitoring in an optically complex
coastal area, using two promising modelling techniques. The case study area is the Wadden Sea, an area
with clear ecological and economic values, which has recently been added to the UNSECO World
Heritage List (UNESCO, 2009). The Wadden Sea is also extremely turbid and optically heterogeneous,
and therefore very pertinent to the study of remote sensing in optically complex coastal areas. The
following research questions are addressed in this thesis:
What are the ranges in concentrations and the spatial and temporal variations in optically active
substances and (specific) optical properties in the Wadden Sea? Which processes are
responsible for variations in these concentrations and (specific) optical properties? How does
this affect (the accuracy of) ocean colour data of the area?
To what extent is water quality and/or water mass monitoring based on (MERIS) satellite data
and using a regionally calibrated inverse bio‐optical model or an end‐member approach
possible, in an extremely turbid and heterogeneous area such as the Wadden Sea?
1.3 Approach and outline
To identify gaps in knowledge in remote sensing research in extremely turbid areas, a review with a
focus on the Wadden Sea was made. This review, in Chapter 2, examines the concentration ranges of
optically active substances that occur in the Wadden Sea and the most important processes influencing
them. Gaps in the knowledge are identified, namely: little research was done on algorithms for
extremely turbid areas to simultaneously derive various substances, and there is a lack of information
on SIOPs and on the apparent optical properties (AOP’s) of the area. Information on SIOPs is essential to
calibrate algorithms and information on AOPs is essential for validation of these algorithms.
Sampling campaigns were conducted to fill the gaps in the knowledge needed for remote sensing.
Results of measurements on SIOPs and AOPs, measured simultaneously with the concentrations of
optically active substances are described in Chapter 3. This chapter also examines the processes
influencing variations in these parameters in time and space.
Two promising approaches are applied to predict concentrations of optically active substances and
model water masses from in situ reflectance measurements and MERIS data. The first approach,
described in Chapter 4, is an inverse bio‐optical model (called HYDROPT) which is calibrated with local
SIOPs derived from samples of the Wadden Sea. Inverse bio‐optical models, based on inherent optical
properties, will theoretically lead to the best retrievals, because the connection between IOPs and water
leaving radiance is direct and physical. Only atmospheric correction will influence the derived properties
(IOCCG, 2006). For greater precision and to avoid ambiguity various researchers argue for regionally
calibrated models (Defoin‐Platel and Chami, 2007, IOCCG, 2000, Lutz et al., 1996). Chapter 4 also shows
the retrieval of concentrations with HYDROPT with in situ spectra and with MERIS data as input, and
addresses the ambiguity of the input spectra and possible solutions for the lack of quality control
encountered. Chapter 4 concludes by comparing “water types” (water masses having similar SIOPs)
Page 14
Introduction
13
derived from satellite images using the inverse‐bio optical model, with general knowledge on water
sources and currents in the Wadden Sea.
A second approach to identify water masses is presented in Chapter 5. This approach is an endmember
model, based on unmixing reflectance spectra in percentages of pre‐defined endmembers. Endmember
techniques are often applied in land remote sensing but have not been used before for water mass
modelling. Chapter 5 compares the water masses derived from the inverse bio‐optical model (“water
types”) with the “water classes” derived from the endmember technique.
The endmember model was applied to several MERIS images. Results of the model derived from images
acquired in different four seasons, with either high and low wind situations and on various moments
during the tidal cycle are compared in Chapter 6. This Chapter examines the extent to which the model
is able to visualise variations in optical properties as described in Chapter 2 and 3. Although no exact
concentrations can be obtained with the endmember technique, the results are promising.
In the last chapter (Chapter 7), a synthesis of the results and conclusions of this thesis is presented. This
chapter provides recommendations for future research and on application of optical remote sensing of
extremely turbid coastal waters.
To accommodate non‐expert readers a glossary of terms and descriptions used in ocean colour remote
sensing is provided (Annex 1). A list of abbreviations, acronyms, and symbols can be found in Annex 2.
1.4 Satellite data: MERIS
The satellite data used in this thesis originates from the MEdium Resolution Imaging Spectrometer
(MERIS), aboard the Envisat satellite (ESA, 2009a), deployed by the European Space Institute (ESA).
MERIS is specially developed for monitoring coastal waters, with a relatively small ground resolution of
~300 x 300 m and nine spectral bands in the visible part of the light spectrum (Table 1.1). These nine
bands enable the distinction of the spectral shapes of the absorptions by SPM, Chl‐a and CDOM.
Table 1.1. MERIS bands
Band number Wavelength (λ) DescriptionBandwidth (λ)
1 412.5 Visible10
2 442.5 Visible 10
3 490 Visible10
4 510 Visible10
5 560 Visible10
6 620 Visible 10
7 665 Visible10
8 681.25 Visible7.5
9 708.75 NIR 10
10 753.75 NIR7.5
11 760.625 NIR3.75
12 778.75 NIR15
13 865 NIR20
14 885 NIR10
15 900 NIR10
Page 15
Chapter 1
14
1.5 Study area: the Wadden Sea
The Wadden Sea is located to the north of The Netherlands, northwest of Germany and west of
Denmark (Figure 1.6). It is a shallow sea and comprises large areas of tidal flats, and is sheltered from
the North Sea by a series of islands. North Sea water enters via tidal inlets (“channels”) around the
islands and fresh water enters via river discharges. With a length of almost 500 km and an area of about
8 000 km2, it is the largest mudflat area in the world and provides breeding, feeding and roosting
grounds for migratory birds. It is also an important breeding and nursery ground for (commercial) fish
and shellfish.
Large parts of the mudflats are tidal, emerging when the water retreats. Water in small basins almost
completely disappears with low tide, in large basins the disappearing and remaining water volumes at
low tide are about equal (Reise and De Jong, 1999). Mudflats can cover a very high percentage of the
surface area in some areas: for example intertidal flats in the Dollard account for 85 % of the surface (De
Jonge, 1995).High tide brings about 15 km3 saline water into the Wadden Sea twice a day. The islands in
the Dutch and East Frisian (German) Wadden Sea are oriented west to east, while the islands in the
North Frisian Wadden Sea are oriented north to south (Figure 1.6). The rivers and Lake IJssel together
discharge about 60 km3 freshwater yearly (Van Beusekom et al., 2001). Residence times of 11‐12 tidal
cycles, or about one week, are typical for most of the Wadden Sea (Postma, 1982). The water column is
well mixed (Postma, 1982; Tillmann et al., 2000) although salinity differences can cause weak
stratification and density differences near river mouths (Postma, 1982).
Large dikes were built in the past. In Denmark, the islands Mandø, Rømø and Sylt were connected to the
mainland and in The Netherlands a dike was built closing off the former “Zuiderzee”, now called Lake
IJssel. The dikes changed local currents and increased the tidal range in the Wadden Sea (Misdorp et al.
1989). A comprehensive overview on the (historical) human transformations of the Wadden Sea is given
by Lotze et al. (2005).
Contents of the saltwater inflow can change drastically in only a few years due to changes in the oceanic
climate, inducing large changes in the Wadden Sea phytoplankton community and ecosystem
(Lindeboom et al., 1995; Edwards et al., 2002). A complete overview of the geomorphology and the
hydrography of the Wadden Sea is given in the report series of the Wadden Sea Working Group edited
by respectively Dijkema et al. (1980) and Postma (1982). For ecological articles, interested readers are
referred to the overview of historical ecological changes in the Wadden Sea given by Lotze (2005).
View other sources
Hide other sources
-
Available from Steef Peters · 3 Oct 2012
-
Available from Steef Peters · 28 Nov 2012
-
Available from vu.nl
-
Available from vu.nl
-
Available from vu.nl
-
Available from vu.nl