CHAPTER 11
IMAGING SPECTROMETRY OF WATER
Arnold G. DEKKER* & Vittorio E. BRANDO* & Janet M. ANSTEE* &
Nicole PINNEL* & Tiit KUTSER** Erin J. HOOGENBOOM*** & Steef
PETERS**** & Reinold PASTERKAMP**** & Robert VOS**** & Carsten
OLBERT***** & Tim J.M . MALTHUS******
*
Environmental Remote Sensing Research Group CSIRO Land and
Water, Canberra, Australia
**
CSIRO Marine Research, Canberra, Australia
***
National Institute for Coastal and Marine Management/RIKZ,
Ministry of Transport and Waterworks, The Hague, The Netherlands.
****
Institute for Environmental Studies, Vrije Universiteit, Amsterdam,
The Netherlands
***** Freie Universität Berlin, Berlin, Germany
****** Department of Geography, University of Edinburgh, Edinburgh,
Scotland
1
Introduction
Remote sensing is a suitable technique for large-scale monitoring of inland and coastal
water quality and its advantages have long been recognised. Remote sensing provides a
synoptic view of the spatial distribution of different biological, chemical and physical
variables of both the water column and if visible, the substrate. This knowledge of the
distribution is essential in environmental water studies as well as for resource management. Therefore, recent years have seen increasing interest and research in remote
sensing of water quality of inland and coastal waters ([1-6]).
The use of water colour remote sensing for the determination of an optical water
quality variable was initially developed for the oceans, as it is virtually the only method
for assessing such vast areas. The optical properties of ocean waters are in general only
affected by phytoplankton and its breakdown products. These optically relatively
simple waters are known as Case 1 waters. Imaging spectrometry is probably overkill
for these waters as a few bands in the blue to green spectral areas are sufficient to
determine chlorophyll concentrations with sufficient precision for most oceanographicbiological purposes. Apart from that argument, imaging spectrometers from space are
only available for civilian purposes since the launch of Hyperion in November 2000.
Thus up till now all imaging spectrometry remote sensing work was carried out from
IMAGING SPECTROMETRY OF WATER
3
aircraft: the scale of ocean remote sensing is not suitable for mapping from aircraft.
Therefore imaging spectrometry of ocean waters are not discussed further in this paper.
All other types of waters, i.e. those waters whose optical properties are influenced by
more than just phytoplankton, are determined to be Case 2 waters. These other optical
properties are usually a selection of dissolved organic matter from terrestrial origin,
dead particulate organic matter and particulate inorganic matter. In addition if the bottom reflectance influences the water leaving radiance signal significantly, a water is
also considered to be Case 2. In reality this distinction is in Case 1 and 2 waters is becoming less useful as more and more examples are being published, where this distinction doesn’t hold. For instance, an algal bloom of the cyanobacterium Trichodesmus in
ocean waters is technically a Case 1 water in this nomenclature. [7] realise the potential
of imaging spectrometry from space for deriving other algal pigments than chlorophyll
a in oceanic waters. The optical properties of algal blooms require different remote
sensing approaches requiring more spectral bands at longer wavelengths. The relatively
simple band ratio algorithms for clear ocean waters do not function well any more in an
algal bloom situation. A more useful approach is to describe water in terms of optically
significant properties and the substances causing these properties.
Many inland and coastal waters are highly affected by anthropogenic influences.
In combination with the complex hydrological situation, highly contrasted structures
evolve in time and space in these aquatic environments. It is obvious that a water system with different optically active substances with temporal and spatial variations is by
far more complex and requires more sophisticated models for remote sensing and separation of the water constituents than a system containing one component only like the
ocean waters. Therefore airborne imaging spectrometry mainly gets applied to coastal
and inland water environments and not to oceans. The vast dimension of oceans necessitates the use of ocean colour sensors on satellite platforms. There fore this chapter
focuses on imaging spectrometry as used for detection and monitoring of inland, estuarine, coastal and coral reef aquatic environments.
2
2.1
Light in water
INTRODUCTION TO THE THEORY
The colour of the water is a complex optical feature, influenced by scattering and
absorption processes as well as emission by the water column and of reflectance by the
substrate (Figure 1). This substrate reflectance (of seagrass, macro-algae, corals, sand,
mud, benthic micro-algae etc.) is similarly a function of absorption and scattering and
in a lesser degree emission of the substrate materials. Variations are essentially determined by the content of particulate and dissolved substances that absorb and scatter sky
and solar radiation penetrating the water surface. The water leaving multi-spectral
radiances are masked by the reflection of sun and skylight at the water surface and by
extinction and scattering processes in the atmosphere. This exposes bottlenecks in the
processing of remote sensing data to water quality maps. To address this bottleneck a
4
A.G. DEKKER ET AL.
Sensor
Direct
sunlight
Diffuse skylight
Backscattered skylight
Remote sensing signal
Absorption and scattering by atmosphere
Reflection at
water surface
Flo or o f
water
Refraction
Water surface
bo dy
CSIRO Land and Water
A. G. Dekker and H. Buettikofer
Figure.1. A schematic diagram of the various processes that contribute to the signal as measured by a remote
sensor in an optically shallow water where the substrate has a significant effect on the water leaving radiance
at the water surface.
careful and precise simulation of the radiative transfer in the water is required, at the
water to air interface and in the atmosphere as a prerequisite for the improvement and
development of new algorithms to retrieve the concentrations of selected water constituents. Therefore the relationship between the optical properties and the concentration
units of these constituents have to be known for the water column as well as the optical
properties of the substrate for substrate mapping. In regard to their optical behaviour,
optically active substances can be split into distinct classes. If the inherent optical properties of these classes are sufficiently well characterised, their contribution to water
column colour can be discriminated and their content quantified. For substrates there is
currently insufficient information on how the optical properties influence the reflectance of substrate materials; therefore it is practice to mainly determine the reflectance of
the substrate and not the concentration dependent absorption and scattering. Since the
water reflected radiation depends on the quantity and specific optical properties of one
or more water constituents, water colour carries spectral information about the concentration of some water quality parameters and possibly of the substrate. For the retrieval
of different water constituents as well as substrate cover from a remotely sensed hyperspectral signal a suite of inversion methods are available, ranging from the often used,
but less precise regression methods, through to physics based inverse modelling or inversion methods. Knowing the specific optical properties of the water constituents and
IMAGING SPECTROMETRY OF WATER
5
of the substrate and modelling the radiative transfer through water and atmosphere as a
function of these water constituents and comparing the modelled multispectral sensor
signal with the measured multi-spectral sensor signal, the water colour data can be used
to determine the concentrations of the water constituents and the substrate cover quantitatively. Analytical methods show better results than empirical, or semi-empirical
methods which use simple correlation, or reasonable band ratios only instead of sophis ticated optical models. However, the exploitation of water colour has greatly been impeded by the incapability to deal with the optical behaviour and complexity of water
constituents. The range of optical water quality properties measurable in the water column that may be estimated by remote sensing has increased from suspended matter to
include properties such as vertical attenuation coefficients of downwelling and upwelling
light, transparency, coloured dissolved organic matter, chlorophyll a contents, even red
tides and blue-green algal blooms. If the water column is sufficiently transparent and the
substrate is within the depth where a sufficient amount of light reaches the bottom and is
reflected back out of the water body it has been demonstrated that maps may be made of
seagrasses, macro-algae, sand and sandbanks, coral reefs etc.
2.2
OPTICALLY DEEP WATERS
The large variation in the concentration of suspended sediments, phytoplankton and
coloured dissolved organic matter in many inland, estuarine and coastal waters results
in a highly variable light climate. Due to the optical complexity of these (often relatively turbid) waters, optical models play a key role in understanding and quantifying
the effect of water composition on optical variables (obtained from either in situ or
remote sensing measurements). Optical modelling is preferred above (semi-) empirical
algorithms that have been the standard for many years in operational applications of
remote sensing (and still are for ocean types of water). Many (semi-) empirical algorithms make extreme simplifications about the water composition, such as the (optical)
domination of one constituent over all the others. There is a wide range of optical
models available for water, from generic radiative transfer models (eg. HYDROLIGHT ,
[8], [9]) to models based on simple analytical solutions developed for specific waters or
conditions. Analytical models have the important advantage that, due to their relative
simplicity, they can be solved very quickly. This is of great importance in a remote
sensing application where a model must be evaluated at every pixel of an image. Thus
we present an analytical optical model that describes the main light processes in both
clear and turbid waters, without and with bottom visibility, taking into account highly
variable optical conditions in the water column and the substrate as well as a complex
geometry of the incident light field and the viewing angles of a remote sensor.
First the basic optical modelling of light in water is presented: radiative transfer
theory. It explains how the radiometric properties, i.e. the radiance and irradiance,
change in the water column due to the optical properties of the medium. Next we present the so-called two-flow model for irradiance, which can be solved analytically for
the diffuse attenuation coefficient K d . With this (approximate) solution an analytical
model for the subsurface irradiance reflectance R( 0− ) is derived and compared with
other analytical models from literature. R(0-) is a measure of the colour of water. It is a
key parameter in the interpretation of remote sensing of water quality, because it links
the measured light to the optical properties of the water. Most of the mathematics and
6
A.G. DEKKER ET AL.
definitions in this chapter are based on the book “Light and water” written by Mobley
[8]. From this extensive text we have extracted those parts that are of interest within the
limited scope of this study, and combined them with the modelling by [10]. Because
the scope of this chapter is imaging spectrometry applications we cannot here build a
comp lete and consistent theoretical frame work showing all the intermediate results, in
stead we will state only important intermediate results and refer to others for the
details.
2.2.1
Optical properties of the water column for optically deep waters
This paragraph introduces the optical properties and variables that are relevant for modelling the optical processes in the water column. Thus this discussion also is relevant
for the optically deep water. The optical properties and variables are summarised in
four tables, one table for each of the four groups that can be identified:
• The inherent optical properties (IOP) are the properties of the medium itself (i.e.
water plus constituents), thus regardless the ambient light field; the IOP are measured by active (i.e. having their own light source) optical instruments (Table 1).
• The radiometric variables are the basic properties of the light that is measured by
passive optical instruments (using the sun as the light source (Table 2).
• The apparent optical properties (AOP) are combinations of radiometric variables
that can be used as indicators for the colour or transparency of the water, for
example the reflectance (Table 3).
• The diffuse inherent optical properties are a combination of IOP and AOP and play
an intermediate role in the derivation of the analytical model (Table 4).
2.2.2
The inherent optical properties
The inherent optical properties (IOP) depend only upon the medium. There are two
main optical processes, absorption and elastic scattering, quantified by the absorption
coefficient and the volume scattering function, respectively. Their definition is based
on a small volume with thickness ∆r , illuminated by a narrow collimated beam of
monochromatic light of spectral radiant power Φ i , see Figure 2 and Table 1. Some
Φ a of the incident power is absorbed within the volume of water. Some part Φ s
is scattered at an angle ψ , within a cone with solid angle ∆Ω . The remaining power
Φ t is transmitted through the volume (see[11]).
part
IMAGING SPECTROMETRY OF WATER
∆Ω
Φa
Φi
ψ
7
Φs
Φt
∆r
Figure 2. The definitions of the inherent optical properties are based on a collimated beam illuminating an
infinitesimal layer (adapted after [8].)
The absorption coefficient is defined as the limit of the fraction absorbed power when
∆r goes to zero, see Table 1. Likewise the volume scattering function is defined as the
limit of the fraction scattered light when both ∆ r and ∆Ω go to zero. From the absorption coefficient and the volume scattering function other IOP can be found, such as
the scattering coefficient and the beam attenuation coefficient. Their definitions are
summarised in Table 1. It must be kept in mind that the absorption and scattering coefficients are functions of wavelength, in other words the IOP are spectral properties. The
wavelength is omitted from the definitions for brevity.
Symbol
TABLE 1. Description and definition of the inherent optical properties.
description/definition
units/reference
a
absorption coefficient
Φa
∆r →0 Φ ∆r
i
m -1
[11]
a ≡ lim
β
volume scattering function
Φ s (ψ)
∆r → 0 ∆Ω→ 0 Φ ∆r∆Ω
i
sr -1 m -1
[11]
β(ψ) ≡ lim lim
b
scattering coefficient
m -1
π
[11]
∫
b ≡ 2 π β(ψ ) sin ψ dψ
0
bb
backscattering coefficient
m -1
π
[11]
∫
b b ≡ 2 π β(ψ) sin ψdψ
π2
c
β~
beam attenuation coefficient
m -1
c ≡ a +b
[11]
normalized volume scattering function
Sr-1
8
A.G. DEKKER ET AL.
β(ψ )
~
β (ψ ) ≡
b
P
Fα
[11]
scattering phase function
Sr-1
P(ψ ) = 4 πβ~(ψ )
[12]
forward scattering probability
-
α
Fα ≡ 2 π ~
β (ψ) sin ψdψ
[12]
∫
0
Bα
backward scattering probability
π
Bα ≡ 2π ~
β(ψ ) sin ψ dψ
[13]
∫
α
B
backscattering to scattering ratio
-
TABLE 1 (Cont.)
π
b
B ≡ 2π ~
β (ψ) sin ψdψ = b
b
π2
[14]
∫
g
asymmetry parameter
π
g ≡ 2π ~
β (ψ ) cos ψ sin ψd ψ
[8]
∫
0
ω0
single-scattering albedo
ω0 ≡
ωb
backscattering albedo
ωb ≡
2.3
b
c
bb
a + bb
[11]
[15]
RADIOMETRIC VARIABLES AND APPARENT OPTICAL PROPERTIES
The fundamental optical variable measured by most remote sensing instruments is
radiance, L . From the radiance a number of other radiometric quantities can be derived, such as the downwelling and upwelling irradiance, see Table 2 (for a more detailed description see [8]). Apparent optical properties (AOP) depend both on the medium
and on the ambient light field, but they display enough regular features and stability to
be useful descriptors of the water body. Definitions of commonly used AOP are listed
in Table 3. The fact that the AOP are relatively stable and often behave well with
depth, makes it easier to relate them to the water composition than (ir)radiance
measurements. In particular, the reflectance just below the surface, R( 0− ) , and the
diffuse attenuation coefficient for downwelling light, Kd , are very suitable, because
they are sensitive to changing water compositions. In Figure 3, the geometry of the
directional radiance vectors is defined.
IMAGING SPECTROMETRY OF WATER
9
Ξu
x
φ
Ξd
θ
y
L(s)
z
Figure 3. Definition of the geometry. s is a vector which gives the direction of the radiance. The vector is
composed of two components: the zenith angle
θ
and the azimuth angle
φ : following the notation by [12]
TABLE 2 Description and definition of the spectral radiometric variables, where
Ξ is the unit sphere (the set of all directions s with solid angle dΩ ) and µ the
cosine of the zenith angle of the horizontal plane.
Symbol
description/definition
L
(spectral) radiance
L≡
µ
Ed
∂ 4Q
∂t∂Ω∂ A∂λ
∂
is the partial derivative).
Units/reference
W m -1 sr -1 nm -1
[11]
cosine zenith angle
-
µ ≡ cos θ
[11]
downwelling irradiance
W m -1 nm -1
∫ µL(s)dΩ
Ed =
[11]
Ξd
Eu
upwelling irradiance
∫ µ L(s)dΩ
Eu =
W m -1 nm -1
[11]
Ξu
E0
scalar irradiance
E0 =
∫ L(s)dΩ
W m -1 nm -1
[11]
Ξ
E 0d
downward scalar irradiance
E 0d =
∫ L(s)dΩ
W m -1 nm -1
[11]
Ξd
E 0u
upward scalar irradiance
W m -1 nm -1
10
A.G. DEKKER ET AL.
E 0u =
∫ L(s)dΩ
[11]
Ξu
TABLE 3. Description and definition of the apparent optical properties.
Symbol
description/definition
R
irradiance reflectance
R≡
R( 0− )
R L ( s)
[11]
Eu
Ed
subsurface irradiance reflectance
R≡
Sr
[8]
Eu
L u (s )
[11]
Ed
E 0d
upwelling average cosine
[8]
Eu
E 0u
Kdnorm = µ d Kd
m -1
[8]
1 dE u
E u dz
normalized diffuse
downwelling light
m -1
[11]
1 dE d
E d dz
downwelling average cosine
µu ≡
Kdnorm
[8]
L u (s )
Ed
ratio of upwelling irradiance to upwelling radiance
µd ≡
µu
sr -1
diffuse attenuation coefficient of upwelling light
Q( s) ≡
µd
[12]
π Lu ( s)
Ed
diffuse attenuation coefficient of downwelling light
Ku ≡ −
Q
-
remote sensing reflectance
Kd ≡ −
Ku
[14]
radiance reflectance
Rrs (s) =
Kd
-
Eu (z = 0 )
Ed (z = 0)
R L ( s) =
Rrs (s )
units/reference
-
attenuation
coefficient
of m -1
[16]
IMAGING SPECTROMETRY OF WATER
2.3.1
11
The diffuse apparent optical properties
In addition to the IOP and AOP there is an intermediate set of optical properties, called
the diffuse apparent optical properties. They describe the absorption and scattering of
down- and upwelling irradiance (Table 4). Most of these properties are only used for
mathematical convenience in the derivation of the analytical model. Exceptions are the
shape factors for upward and downward scattering functions (Table 4 and Figure 4),
since they are not just intermediate parameters, but remain present in the final analytical model. The shape factors ‘convert’ the backscattering coefficient into the upward
and downward scattering functions. Figure 4 illustrates that the upward scattered photons partly originate from photons that are scattered forward (shaded area). Since most
particles in water scatter more light in forward directions than in backward directions,
this contribution can be significant [17].
ard
kw
bac ard
w
for
L (s' )
L (s' )
L(s' )
upward
downward
L(s)
L(s ' ) that is scattered into the directions s and
~
contributes to L(s) is given by β ( s, s' ) . The angle between the vectors s' and s is Ψ . Right: The
Figure 4. Left: The fraction of the incident radiance
shape factor for downward scattering indicates the difference between downward and forward scattering
(shaded area). Likewise, the shape factor for upward scattering indicates the difference between upward and
backward scattering (see shaded area in figure).
2.3.2
2.3.2.1
The two-flow model for irradiance
The radiative transfer equation
This chapter considers the radiative transfer equation (RTE) that describes the
behaviour of radiance in water. If we think of radiance as a beam of photons, six basic
interactions of these photons with water can be distinguished ([8] § 5.1):
• loss of photons by conversion of radiant energy to non-radiant energy (absorption)
• loss of photons by scattering to other directions without change in wavelength
(elastic scattering)
• loss of photons by scattering with change in wavelength (inelastic scattering)
• gain of photons by conversion of non-radiant energy into radiant energy (emission)
12
•
•
A.G. DEKKER ET AL.
gain of photons by scattering from other directions without change in wavelength
(elastic scattering)
gain of photons by scattering with change in wavelength (inelastic scattering)
The discussion of the radiative transfer in water will be based on absorption and elastic
scattering processes only. Inelastic scattering effects, especially fluorescence of chlorophyll will only be discussed in the applications section. Now we have defined the inherent and (diffuse) apparent optical properties we can start with the radiative transfer
equation (RTE). Many authors have elaborated on the RTE see for example [10]; [17];
[8] p251). Here we shall give a concise overview. The RTE shows how the radiance L
changes due to the optical properties of the water, hence the IOP: the beam attenuation
coefficient c , scattering coefficient b and the normalised volume scattering function
~
β . As an intermediate step to the analytical model the transfer equations for upwelling
and downwelling irradiance must be derived from the RTE for radiance. It is assumed
that the water body is source free, i.e. inelastic scattering and true emission are
neglected. In addition, it is assumed that the water body is time-independent,
horizontally homogenous with a constant index of refraction [8].
Finally, it is assumed that the absorbing and scattering particles are far apart with
respect to λ . This latter assumption is flawed when many absorption and scattering
particles are tight together, e.g. within one phytoplankton cell. In this case the scattering coefficient is not independent of absorption.
Symbol
ad
TABLE 4. Description and definition of the diffuse inherent optical properties.
description/definition
units/ref.
diffuse absorption function for downwelling irradiance
ad ≡
au
bdd
diffuse downward scattering function for downwelling irradiance
m -1
[8]
Ξ d Ξd
m -1
~
∫ ∫ L(s ' )β (s , s ' )dΩ' dΩ
[8]
diffuse downward scattering function for upwelling irradiance
m -1
bdu ≡
bud
~
∫ ∫ L(s ' )β (s, s' )dΩ' dΩ
diffuse upward scattering function for upwelling irradiance
buu ≡
bdu
b
Ed
m -1
[8]
a
µu
bdd ≡
buu
[8]
a
µd
diffuse absorption function for upwelling irradiance
au ≡
m -1
b
Eu
b
Eu
Ξ u Ξu
∫
~
∫ L(s ' )β (s, s' )dΩ' dΩ
This study
Ξ dΞ u
diffuse upward scattering function for downwelling irradiance
m -1
IMAGING SPECTROMETRY OF WATER
bud ≡
b
Ed
~
∫ ∫ L(s ' )β (s, s' )dΩ' dΩ
cd ≡
c
= ad + bud + bdd
µd
diffuse attenuation function for upwelling irradiance
cu
cu ≡
cdd
c uu
rd
c
= au + buu + bdu
µu
m -1
[8]
m -1
[8]
local transmittance functions for downwelling irradiance
m -1
cdd = ad + bud
[8]
local transmittance function for upwelling irradiance
m -1
cuu = au + bdu
[8]
shape factor for upward scattering
-
rd ≡
[8]
bud µd
bb
shape factor for downward scattering
ru
This study
Ξu Ξd
diffuse attenuation function for downwelling irradiance
cd
13
ru ≡
[8]
bdu µ u
bb
Under all these assumptions the radiative transfer equation for unpolarised
radiance is given by
(1)
dL(s )
~
µ
dz
= −cL (s ) + b ∫ L( s' ) β ( s, s ' ) dΩ'
Ξ
where Ξ is the unit sphere (here the set of all directions s' with solid angle dΩ' ) and
µ the cosine of the zenith angle. For sake of brevity the dependence on wavelength and
depth is omitted. Equation 1 describes that the change in radiance over a depth interval
dz corrected for the zenith angle by µ is equal to that part of the radiance that is not
attenuated by absorption or scattering (c = a + b) plus the contribution of the radiance
scattered at all angles projected onto the initial direction of radiance s.
2.3.2.2
Two-flow modelling
From eq. 1 expressions for the downwelling and upwelling irradiance can be derived
by integrating over all angles in the downward and upward hemisphere respectively.
With the definitions for the diffuse IOP the derivation of the radiative transfer equations for irradiance can be obtained, by integrating the RTE for radiance over all angles.
Through several steps of integration and rewriting (see [8] ) the transfer equation for
downwelling irradiance can be obtained
(2)
dE d
= − ( ad + bud ) E d + bdu E u
dz
14
A.G. DEKKER ET AL.
Equation 2 describes the change in downwelling irradiance with depth is equal to the
downwelling irradiance that is not diffusely absorbed or diffusely scattered upwards
plus the diffuse downward scattered fraction of the upwelling irradiance at that depth
interval. Following the same line of reasoning, integration of the RTE over all angles in
the upward hemis phere gives the irradiance transfer equation for upwelling irradiance
dE
(3)
− u = − (a u + bdu ) E u + bud E d
dz
Equations 2 and 3 form the two-flow model for the source-free case, as illustrated in
Figure 5. We see that the downwelling irradiance:
• decreases with depth because of absorption of E d ;
•
decreases because of scattering of E d into E u ;
•
increases because of scattering of E u into E d .
2.3.2.3
An analytical solution of the irradiance RTE
Under certain conditions the two-flow model developed in the previous chapter can be
solved for the vertical attenuation coefficient. It is assumed the medium is homo geneous, i.e. it is assumed that a set of effective IOP can be used that are constant over
depth. Another assumption is that the water is optically deep so bottom effects can be
neglected. Furthermore, it is assumed that the downwelling irradiance decays exponentially with depth (known as Beer’s law)
(4)
E d ( z ) = Ed ( 0) exp ( − Kd z ) .
[10] derives an analytical expression for Kd that goes one step beyond the single
scattering approximation, since it includes a second order scattering effect in the
second term. In clear waters this second term is often neglected Kd ≈ cdd
K d = cdd −
bdubud
cuu + cdd
(5)
This analytical model for Kd can be rewritten in terms of the absorption and backscattering coefficients. Substituting the relevant definitions in Table 4 gives
(6)
a
bb
ru µ d
bb
r µ +r µ
Kd =
k= d u u d .
,
1 + rd 1 −
µ d
a µ u + µ d a + kbb
µu + µd
Equation 6 can be considered as a generic model that is exp ected to be valid in both
clear and turbid waters. In order to compare the concept of equation 6 with other models found in the literature, it is convenient to neglect the second term in equation 6
(7)
b
a
Kd =
1 + rd b
µ d
a
Several analytical models similar to eq. 7 can be found in literature. For instance, setting the shape factor rd to 1 gives the model of [12] and if, in addition, the µ d is
approximated by µ 0 (the cosine of sun zenith) we arrive at the model of Gordon et al.
(1975).
IMAGING SPECTROMETRY OF WATER
15
Ed
bud E d
bud E d
a u Eu
ad Ed
bdu E u
bdu E u
Eu
a
b
Figure 5. (a) The change with depth of the downwelling irradiance can be interpreted in terms of absorption
and scattering functions, (b) idem for the upwelling irradiance.
TABLE 5. Several analytical models for the diffuse attenuation coefficient can be found in
literature.
Model
ref.
a bb
1+
µ 0
a
1 b
K d = a 1 +
6 a
a
b
Kd =
1 + G (µ 0 , g )
µ0
a
[14]
Kd =
[18]
[19]
2.236
0.849
G (µ 0 , g ) = µ 0
− 2.447 −
− 0.739
g
g
Kd =
a
µd
bb
1 + a
bb
ru µ d
bb
,
1 + rd 1 −
a µ u + µ d a + kbb
r µ +r µ
k = d u u d
µu +µd
b
a
Kd =
1 + rd b
µ d
a
Kd =
a
µd
[10];
[12]
this study
[10]
An analytical model is presented for the diffuse attenuation coefficient that can be
expected to be valid for turbid waters. It relates the total absorption and backscattering
coefficient to the Kd , and to specify the model in eq.6 four parameters (AOP) are re quired: µ d , µ u , rd and ru . Unfortunately relatively little is known about the values for
the shape factors in turbid waters. In most analytical models the shape factors are set to
1. However, Stavn & Weidemann (1989) find that rd can vary between 1.3 and 10 and
that ru can vary between 1.8 and 20, during the development of a phytoplankton bloom
16
A.G. DEKKER ET AL.
in Case I (ocean type) waters. These results indicate that the (variation in the) shape
factors must be taken into account. As far as we know their values are not yet determined for turbid water types. Hence, research on the shape factors in these waters is
highly recommended.
As will be evident from the discussion of optical models in optically shallow
waters presented in paragraph 2.4 a clear understanding of the nature of attenuation
with depth is essential for remote sensing of bathymetry or a substrate or a substrate
cover.
2.3.3
An analytical model for the irradiance reflectance
We refer to [10] for a complete derivation of the analytical model for the irradiance
reflectance. The reason for choosing this model is that it can act as a reference for
understanding all other models of this kind found in literature. In terms of the backscattering and absorption coefficients the [10] analytical model for irradiance re flectance
can be written as
(8)
rµ
bb
r µ +r µ
R( 0− ) = d u
,
k= d u u d
µ u + µ d a + k bb
µ u + µd
This equation for R( 0− ) states that R( 0− ) is proportional to the backscattering divided by the sum of absorption and the second order backscattering. In more detail the
equation states that the irradiance reflectance is equal to a factor times the backscattering divided by the sum of absorption and the second order backscattering (whereby the
second order backscattering is multiplied by a factor that accounts for up and downwelling shape factors and the average cosines for up and downwelling irradiance). The
multiplication factor takes into account the downwelling shape factor and average cosines of the up and downwelling irradiances.
Although even this model contains approximations as explained by [10] it may be
expected to yield quite accurate results for turbid waters. Various authors have developed analytical models for the subsurface reflectance, which can be related to the remo te
sensing reflectance measured from (far) above the water surface. Therefore, the
subsurface irradiance reflectance plays an important intermediate role in many remote
sensing applications on water quality. Most of the models are developed and validated
for relatively clear waters. From the comparison of the models summarised in Table 5
we see that the model in eq. 8 is generic in the sense that most of the other models can
be obtained by substituting approximate values for the AOP. For instance, if we set the
shape factors to unity, we get the model of [12]. In case of a diffuse upwelling light
field ( µ u =0.5) and, µ d being approximated by µ 0 (the cosine of sun zenith), we get the
second model by [12]. If in addition the sun zenith is 0 and the backscattering term in
the denominator is neglected, we get the well-known model by [14]. If we assume a
totally diffuse light field ( µ u = µ d =0.5) the model of [15] is obtained. Finally, if we
assume unit shape factors and µ u = µ d the exact solution given in [10]) and underlying
the derivation of equation 8 simplifies to the model of [20]. Kirk’s models [21, 22]
were derived in a different fashion from the more analytically derived models as they
are based on Monte Carlo simulations of the underwater light field. If the
parameterisation of Kirk’s models applies to the waters under study they are perhaps
IMAGING SPECTROMETRY OF WATER
17
the easiest to apply as apart from the absorption and backscattering coefficients only
the average cosine of all photons just under the water surface are required. In a more
general applicability and because there is so little information available yet on the
shape factors, we recommend using the [12] first model as it has the least assumptions
and is easiest to use in simulation models.
TABLE 6. Analytical models for the subsurface reflectance found in literature.
model
ref.
a + bb +
3
R (0− ) =
[20]
bb
R (0− ) =
∑
n =0
(a + bb )
b
fn b
a + bb
R (0− ) = 033
.
2
− bb2
[14]
n
[23]
bb
a
R (0− ) = ( 0.975 − 0.629 µ 0 )
[24]
bb
a
R ( 0− )
b
= 0.095 b
Q
a + bb
[25]
bb
a
[26]
R(0− ) = f
2
f = 063
. + − 022
.
R (0− ) = f
bb (w)
b (w )
bb (w)
− 0.05 b
− 0.31 − 0.25
µ0
bb
bb
bb
[27]
bb
a + bb
R (0− ) = (1018
.
− 0657
. µ0 )
bb
a + bb
1
bb
R (0− ) =
µ
1 + d a + bb
µu
1
bb
R (0− ) =
1 + 2µ0 a + bb
rµ
bb
R= d u
,
µ u + µ d a + kbb
bb
a + 0.361bb
[[22]
[15]
R (0− ) = 05
.
[12]
[12]
k=
rd µ u + ru µ d
µu + µ d
[10],
this study
Most studies neglect the variation in the shape factor and use empirical corrections based on the sun zenith angle in stead of average cosines. [28] investigated the
validity of the model by Gordon in three turbid water samples with maximum bb a of
18
A.G. DEKKER ET AL.
0.5. They were not able to fit one model due to the limitations of the Gordon model for
variations in the illumination conditions. Main errors that [28] identify are the
combined effect of various solar zenith angles and skylight illu mination, and the nondiffuse distribution of the upwelling light. These findings indicate that for turbid waters
the values for the average cosines for downwelling and upwelling light play a
significant role.
A potential problem is that it is probably impossible to use one set of typical values for µ d , µ u , rd and ru . Near many coasts a large spatial gradient of turbidity occurs
in the first 20 km. Near the coast and in intertidal area’s large temporal variations in
turbidity are measured, caused by the large variation in the concentration of suspended
particles: from a few to more than 1000 g m-3 within one tidal period. Since the optical
conditions can be so different it may be necessary to use a two-step approach. First, the
IOP (and subsequently the constituent concentrations) are calculated from the reflectance using a set of average values for µ d , µ u , rd and ru . Second, using the calculated IOP
as input these four parameters are calculated with a RTE-model such as HYDROLIGHT and then the IOP are calculated again with these adapted values. It is recommended to investigate the range of values for the average cosine and the shape factors
that may occur in inland, estuarine and coastal waters. Most of the models in Table 5
are developed for the reflectance just below the air-water interface. The subsurface
reflectance is most relevant for remote sensing applications. However, in situ measurements can be carried out at any depth. In many cases it is preferred to measure at some
distance from the surface to minimise wave effects. From our analysis it appears that
the model is valid for any depth, provided we assume that the downwelling irradiance
decays exponentially with depth and that the water is optically deep.
2.4
OPTICAL SHALLOW WATERS
[29] present a clear discussion of the physics of an optical shallow water body where
part of the reflectance at the surface is composed of a bottom signal. They describe
their analytical model for optically shallow water in the same terms used for describing
the physics of the underwater light field for an optically deep system. Therefore the
following text is mainly derived from their text. They use an approach derived from
the two-flow equations to obtain approximate formulae based on a set of simplifying
assumptions.
In optically shallow waters Eu (0), is the sum of upwelling irradiance originating
within the water column (where none of the photons have interacted with the substrate), Eu (0)C, and the upwelling irradiance reflected from the substrate (where each of the
photons have interacted with the substrate), Eu (0)B
(9)
E ( 0) = E ( 0) + E ( 0)
u
u
C
u
B
To estimate the first term consider an infinitely thin layer of thickness dZ at depth Z,
where the downwelling irradiance is Ed (Z). At this depth the fraction of upwelling irradiance created by this layer is
(10)
dE ( Z ) = b E ( Z ) dZ
u
bd
d
Ed (Z) can be expressed as in equation 4. Before it reaches the surface , d Eu (Z) is attenuated along the path of Z to the surface, expressed by
exp( −κZ )
(11)
IMAGING SPECTROMETRY OF WATER
19
where κ is the vertical diffuse attenuation coefficient for Eu (Z) as defined by [30]. It is
important to realise at this stage that Ku is the vertical attenuation coefficient for diffuse
upwelling light Eu measuring it from the surface downwards where as κ is the vertical
attenuation coefficient for diffuse upwelling light originating in each layer of the water
column and measuring it depth upwards. The contribution of the considered layer in eq.
10 to the upwelling irradiance just below the water surface is expressed as
(12)
dE ( Z → 0) = b E ( 0) exp[ −( K + κ ) Z ]dZ
u
bd
d
d
If it is assumed that bbd , Kd and κ are not depth-dependent, the contribution of all layers
between Z and 0 is
z
(13)
Ed ( 0, Z ) = bbd Ed (0) ∫ exp[ −( K d + κ ) Z ]dZ
0
Equivalent to:
(14)
Eu ( 0, Z ) = ( K d + κ ) −1 bbd Ed (0)(1 − exp[ −( K d + κ ) Z ]dZ )
For an infinite water depth eq. 14 reduces to :
(15)
Eu ( 0, ∞ ) = ( K d + κ ) −1bbd Ed ( 0) = R(0, ∞) E d (0)
R(0, ∞) is in the case of an optical deep water equal to R(0-) as given in Table 6. If we
assume a totally absorbing substrate at depth H, eq. 14 becomes:
(16)
E ( 0, H ) = R E ( 0)(1 − exp[ −( K + κ ) H ]) = E (0)
∞
u
d
d
u
C
equivalent to the first term in eq. 9. For optically shallow water with an albedo A, the
upwelling irradiance originating from reflection at the substrate at a level H (imme diately above the bottom) is
(17)
E ( 0) = AE ( 0) exp[ −( K + κ ) H ])
u
B
d
d
Filling in eqs 16 and 17 into eq 9 the following equation is obtained
Eu ( 0) = Ed (0)( R∞ (1 − exp[ −( K d + κ ) H ])) + A exp( −( K d + κ ) H ))
(18)
When eq. 18 is divided with Ed (0) the expression is derived for the reflectance just
below the surface of a homogeneous water body with a reflecting substrate (identical to
that of [31])
(19)
R( 0, H ) = R + ( A − R∞ ) exp[ −( K + κ ) H ]
∞
d
Because there are actually two upwelling light streams: one from the bottom and one
from the water column κ can be described as κB and κC respectively. Equation 19 then
becomes
(20)
R( 0, H ) = R + exp( − K H )[ A exp( −κ H ) − R exp( −κ H )]
∞
d
B
∞
C
20
2.5
A.G. DEKKER ET AL.
LIGHT ABOVE WATER
2.5.1
Water surface effects
Previously we described what happens to downwelling irradiance once it has penetrated the water surface. Downwelling irradiance above the water surface will undergo
one of two effects. It will either be reflected from the surface itself back into the atmosphere or it will pass across the air-water interface into the water, being refracted in the
process (Figure 6). The surface reflected component is an unwanted signal in remotely
sensed imagery used for water quality assessment. We are interested in that fraction of
light which passes into the water column, interacts with it and perhaps with the substrate and then may be reflected back across the interface to be detected by a sensor.
i
Air
Reflection
Refraction
Water
j
48.8°
Figure 6. Reflectance and refraction at the water surface.
Refraction can be calculated according to Snell’s law: n a . sin i = n w . sin j , where n w and
n a are the refractive indices for both water and air, respectively. Ordinarily, n a is usually defined as equal to 1 and for most purposes the refractive index of sea water can be
regarded as being 1.338 (although it is affected by both water temperature and salinity,
[32]):
sin φ a
sin φ w
=
nw
na
=
1.338
For freshwater n w = 1.333. The implications of refraction at the air water interface are
that, for a flat sea surface, the whole of the hemispherical irradiance from the atmosphere which passes across the interface is compressed into a cone of underwater light
with a half angle of 48.8° (Figure 6). This phenomenon also has implications for reflected radiance. Any backscattered light travelling upwards and striking the surface at
angles greater than 48.8° will be totally internally reflected - they will not penetrate the
surface. Similarly, the flux contained within the solid angle below the surface will be
IMAGING SPECTROMETRY OF WATER
21
spread out because of the refraction above the surface when it passes across the interface.
The effects of surface roughness - The surface of a natural water body is almost
never flat; wind driven waves will have a major effect on the ability of light to pass
across the air water interface. The effect of wind roughening is generally to widen the
solid angle through which light will penetrate, i.e. some light will penetrate the water at
angles greater than 49 degrees. The presence of slicks and/or whitecaps will further
modify the light field in different ways from that of wave action [33] [34]). Oil slicks,
apart from having a dampening effect on wave action will cause higher reflectance in
certain regions of the spectrum.
2.5.2
Atmospheric effects and atmospheric correction
Although the physics of atmospheric correction of remote sensing data over waters is
essentially the same as for terrestrial targets, there are a few practical differences that
need to be addressed. For any water body it is the signal coming from within the water
body that is the desired signal. On land it is the surface reflected signal that is of interest. For water bodies the surface reflected signal is a signal that is considered as noise,
and is composed of the reflected component of diffuse skylight and of the direct sunlight impinging on the water surface. Water bodies in general reflect (as subsurface
irradiance reflectance) in the range of 1 to 15% of downwelling irradiance. The majority of waters reflect between 2 and 6% of downwelling irradiance. Thus to obtain e.g.
40 levels of irradiance reflectance in the range of 2 to 6% reflectance we need a minimal accuracy of atmospheric correction to 0.1% reflectance.
Water body surfaces show swell, waves and capillary waves with facets of tens of
meters to a few centimetres. Although their distribution can be predicted in a stochastic
way, the inherent chaotic nature complicates adequate removal of surface reflectance
effects. Therefore flight planning of airborne imaging spectrometry campaigns needs to
consider the solar zenith angles and azimuths that minimise (given the FOV of the
scanner) imaging of sunglint effects of the water surface. A rule of thumb is that solar
zenith angles of 30º to 60º are optimal over water targets and that flight paths should be
flown at 0º or 180º headings with respect to the solar azimuth. At low latitudes this will
mean flying in a short period around noon in summer to achieve a maximal amount of
irradiance. At mid-latitudes the flight time will be dependent on the season: as the maximum solar zenith angles increases going from summer to winter -the flight time envelope decreases from approximately 6 to 8 hours surrounding noon to two hours surrounding noon. At low latitudes solar noon must be avoided to avoid sunspot effects (direct
reflectance from horizontal water surfaces into the FOV); thus a situation arises with
two periods: one in the morning and one in the afternoon.
22
3
3.1
A.G. DEKKER ET AL.
Optically deep and shallow waters: applications and case studies
INTRODUCTION
The previous paragraphs described one of the theoretical approaches to describing the
processes in the underwater light field. Now we will present literature reviews and case
studies of spectral measurement, modelling, simulation and imaging spectrometry applications. First, inland waters and estuaries as the two most studied optically deep
water systems will be discussed. Next, seagrasses and coral reefs as the two most studied optically shallow systems will be discussed. After the literature review for inland
waters a case study is presented for lakes in Germany where an inverse modelling
method was applied to derive images of chlorophyll and suspended matter. These two
variables often confuse simpler algorithms as chlorophyll is the pigment in the algae
and the algae constitute part of the biomass that is part of the suspended matter. The
estuary example will be discussed as it represents an application for optically deep
water where it is currently possible to parameterise most of the inherent and apparent
optical properties. The estuary example is for modelling and measurement based on in
situ spectra only, as there are very few publications that describe actual airborne imaging spectrometers flown over estuaries, where there is no bottom visibility.
After these optically deep waters the optically shallow waters are discussed. First a
discussion on the combined effects of a water column and a substrate takes place. In
this discussion bathymetry play an important role. Next a literature review and a case
study on seagrass remote sensing is presented. The final subject discussed are the coral
reefs. In terms of the analytical model, the seagrass and coral reef examples represent a
more hybrid situation where not everything can yet be described in terms of IOPs and
AOPs. Therefore, it is necessary to rely more on in situ measured reflectance spectra in
combination with analytical modelling or numerical modelling of the effects of the
water column.
3.2
3.2.1
OPTICALLY DEEP INLAND AND ESTUARINE WATERS
Imaging spectrometry of optically deep inland waters
A review of satellite and airborne remote sensing of aquatic ecosystems is given in [35],
summarily updated in [24]. [36] gave a review of airborne remote sensing. [2] wrote a
comprehensive review of satellite and airborne remote sensing of inland waters, including
imaging spectrometry. [37] present a sound treatise on remote sensing of inland and
coastal waters, where the emphasis of the applications is on the Laurentian Great lakes in
the USA. [4] reviewed the literature on satellite remote sensing of lakes and [5] presented
a review of satellite remote sensing of inland and coastal waters.
After 1984 remote sensing of inland waters has taken place mainly using data from satellite based sensors such as Landsat Thematic Mapper and SPOT-HRV (IRS –LISS
series, CZCS, NOAA-AVHRR) and airborne remote sensing using instruments varying
from multispectral scanners to line spectrometers and imaging spectrometers such as the
CASI, AISA, AVIRIS, HYMAP and DAIS-7915. The CASI and AISA (and in a lesser
degree HYMAP and AVIRIS) systems are not each one sensor with fixed capabilities.
IMAGING SPECTROMETRY OF WATER
23
They are a family of sensors, whereby there is a progression in sophistication of the sensor with each new model developed. E.G for the CASI there are now approximately 20
systems operational. Each one of them has slightly different capabilities, whereby each
upgrade to an existing sensor (specifically the case for AVIRIS) or each new sensor outperforms the previous version. Notice must be taken that the results of a CASI or AVIRIS
flown in 1990 are not the same as the results for a CASI or AVIRIS flown in 2000 because the performance of the instrument has greatly increased. Moreover for the CASI an
extra complication in comparing results is that it is a programmable imaging spectrometer, meaning that each application may have a unique spectral band s et applied. For the
development of high spectral resolution remote sensing applications, both imaging and
non-imaging (either line or point measurements) data are of interest. Ground-based surface and subsurface spectral measurements may serve as surface calibration and as the link
between the remotely sensed signal and the inherent optical properties.
Specific Inherent Optical Properties Inland Waters
2.00
0.05
-1
0.04
1.50
0.03
1.00
0.02
0.50
0.01
0.00
400
-1
2.50
0.06
a(w)&a* (CDOM)&a*(tripton)(m )
3.00
0.07
(m )
b(w)&b*(phyto)&a*/b*(cpc)&b*(tripton)
0.08
b(w)
a*(phyto)
b*(phyto)
a*(tripton)
a*(CPC)
b*(CPC)
a(w)
a*(CDOM)
b*(tripton)
0.00
450
500
550
600
650
700
750
Wavelength (nm)
Figure 7. Inherent optical properties of Dutch inland waters (Dekker 1993). Left axis: b(w), a*(ph), b*(ph),
a*(CPC), b*(CPC) and b*(tr); right axis: a(w), a*(CDOM)norm 440 and a*(tr). Units: a and b in ( m -1) ; a* and b*
of phytoplankton, CPC in (mg m -2 ) and a*(tr) and b*(tr) in (g m -2 ).
To summarise relevant information from the literature, a literature review is carried out
discussing only those studies that actually made use of an airborne spectroradiometer
(hand-held) or imaging spectrometer. This selection criteria is strict and excludes a significant amount of excellent work that discusses underwater and just above water measurements of inherent and apparent optical properties. Interested readers are referred to
the reviews by [2], [37],[4]and [5]. In order to put the airborne spectrometry review into
a correct perspective, simulations of reflectance using a bio-optical model are presented
for algae dominated water and for total suspended matter dominated water.
3.2.1.1
Simulations using a bio-optical model
Figures 7 to 9 demonstrate the variability of inland water spectra. Figure 7 shows the
inherent optical properties of inland waters as determined for Dutch lakes by [27]. In the
case of inherent optical properties such as the absorption by the sum of chlorophyll a and
phaeophytin (CHL) the specific inherent optical property is given (meaning the amount
of absorption or scattering or backscattering per unit weight). Figure 8 shows a
24
A.G. DEKKER ET AL.
simulation run where CHL varied from 0-90 in 10 µ g l-1 steps, CPC varied in 0-135
µ g l-1 steps and a(cdom) 440 varied from 1.0 to 1.9 m-1 . The non-chlorophyllous
suspended matter (tripton) was kept fixed at 1 mg l-1 . This is thus a simulation of a deep
lake where a cyanobacterial dominated phytoplankton bloom is occurring.
Subsurface reflectance
run1
2
3
4
0.035
0.030
R(0-)
0.025
5
6
7
8
0.020
0.015
0.010
9
10
0.005
0.000
400
450
500
550
600
650
700
750
Wavelength
run
1
2
3
4
5
6
7
8
9
10
Seston
1.0
1.7
2.4
3.1
3.8
4.5
5.2
5.9
6.6
7.3
Tripton
CHL
CPC
1.0
0
0
1.0
10
15
1.0
20
30
1.0
30
45
1.0
40
60
1.0
50
75
1.0
60
90
1.0
70
105
1.0
80
120
1.0
90
135
a(cdom)440
1.0
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
Figure 8. Simulation run where CHL varied from 0-90 in 10
µ g l-1 steps, CPC varied in 0-135 µ g l-1 steps
and a(cdom)440 varied from 1.0 to 1.9 m -1 .
Figure 9 shows a simulation run where CHL is fixed at 1 µ g l-1 , CPC is zero and
a(cdom) 440 is fixed at 1.0 m-1 . The non-chlorophyllous suspended matter (tripton) was
varied from 10 mg l-1 to 100 mg l-1 . This is thus a simulation of a lake or river where a
substantial amount of suspended matter is entering the water column, either through river
input or through wave-induced resuspension of bottom sediments. Figure 9 shows that if
the main feature varying is TSM, reflectance increases over the entire spectrum, and this
increase tends to saturate at higher concentrations of TSM. From the two figures it is
clear that CDOM and pigments such as CHL and CPC as well as the tripton all contribute
to lowering the reflectance at the blue wavelengths. Centred at 624 and 676 are the CPC
and CHL induced reflectance troughs. These troughs are flanked by local reflectance
peaks at 570-600 nm, 650 and 704-710 nm respectively. Note that no fluorescence term
was required to simulate the often occurring reflectance peak at 706 nm. Many authors
erroneously contribute this peak entirely to fluorescence whereas it is mostly due to a
combined minimum in absorbing features at this wavelength. Indeed, the work [38], who
originally conceived the idea of measuring fluorescence by remote sensing in the eighties,
IMAGING SPECTROMETRY OF WATER
25
demonstrates that above 10 to 20 µ g l-1 CHL the fluorescence signal centred at 683-685
nm is absorbed by the broadening absorption of CHL.
Subsurface reflectance
run
1
0.250
2
3
4
R(0-)
0.200
0.150
5
6
7
0.100
8
9
10
0.050
0.000
400
450
500
550
600
650
700
750
Wavelength
run
1
Seston
Tripton
CHL
CPC
a(cdom)440
2
3
4
5
6
7
8
9
10
10.1
10
20.1
20
30.1
30
40.1
40
50.1
50
60.1
60
70.1
70
80.1
80
90.1
90
100.1
100
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
1
1
1
1
1
1
1
1
1
Figure 9. Simulation run where CHL is fixed at 1
µ g l-1 , CPC is zero and a(cdom)440 is fixed at 1.0 m -1. Only
the tripton part of seston is increased.
The simulations in Figures 8 and 9 may be used for optimal spectral band location of
programmable imaging spectrometers such as the CASI or AISA.
3.2.1.2
Literature review of imaging spectrometry of optically deep inland waters
The earliest airborne imaging spectrometry results are published for the Programmable
Multispectral Imager on Canadian waters in 1985; PMI and CASI campaigns over eutrophic Dutch lakes in 1988 and 1990; AVIRIS used for oligotrophic lake Mono and
the saline lake Tahoe in the USA from 1990-1993; a CASI for Tennessee Valley Reservoirs mapping in the period of 1991-1992. The period 1993-1996 saw imaging spectrometers - mostly CASI’s- deployed in Australia, Netherlands and German lakes. These
more recent researches established a more analytical approach to hyperspectral remote
sensing, and are thus to be considered as a breakthrough towards more quantitative,
intercomparable, methods.
The largest data set and most consistent application was in the Netherlands where
imaging spectrometry missions were flown in 1988, 1990, 1992, 1993,1995 and 1997
using PMI, CAESAR and a sequence of CASI sensors. A development in time is clear
from initial more empirical and semi-empirical approaches towards more analytical
26
A.G. DEKKER ET AL.
approaches involving the use of parameterized bio-optical models that are subsequently
inverted. Especially the work by [39] and [40] and [41] illustrate the use of inversion
methods, whereby optical water quality variables are inverted from the remotely sensed
signal taking into account partially covering other optical water quality variables: e.g.
estimating chlorophyll contents and TSM, whereby account is taken of the effect of the
chlorophyll associated biomass of algae to the TSM contents.
From 1997 onwards, increased activity in Scandinavian countries and Germany is
evident. Especially the work by [42] and Schaale et al. (1998) is original and a portent
of future methodologies involving the creation of massive amounts of simu lated data
based on (i) bio-optical modelling, (ii) air-water interface and atmospheric modelling,
as well as (iii) incorporating the sensor look geometry across track, after which (iiii)
either look-up tables or neural networks are created for inversion methods. See the case
study of the Berlin lakes CASI images discussed later. From the literature (26 studies)
it is evident that CHL is the prime variable measured in all cases (26); the blue-green
(or cyanobacterial) pigment CPC in 8 cases; total suspended matter (or an equivalent
thereof-the definitions and methods of measuring TSM are highly variable) in 16 cases;
Secchi depth transparency in 18 cases; vertical attenuation of PAR in 8 cases and a few
cases of turbidity (6) and CDOM . The CDOM determinations from the remote sensing
data were not really successful as the sensors used all showed low sensitivity in the
blue wavelengths, where CDOM absorption is most noticeable. The range in optical
water quality variables detected during these campaigns was large:
TABLE 7. Review of measured water quality variables.
variable
CHL (µg l-1 )
CPC (µg l-1 )
TSM (mg l-1 )
SD
(m)
Kd
(m-1 )
NTU
Min
0
0
1
0.05
0.4
1
Max
1010
1519
700
8.5
10
640
An excellent example of the use of imaging spectrometry is the determination of
cyanophycocyanin, as only an instrument that can parameterise the absorption feature
of CPC at 624 nm, by simultaneously measuring the local reflectance peaks at 600 and
648 nm (See Figure 8) is capable of determining the presence of cyanobacteria [27]
[39].
3.2.1.3
Case study: inland water quality with inverse modelling
A case study is presented of an inverse modelling approach using a radiative transfer
model which is a coupled water-atmosphere model, in this case the MOMO model
based on the matrix operator method [43] Fischer, 1984 #382 [44]. MOMO is a highly
sophisticated model suited especially for the simu lation of the radiative transfer in clear
and turbid atmospheres including the water body with a rough water surface.
Comparisons with a water-atmosphere Monte Carlo model as well as comparisons with
IMAGING SPECTROMETRY OF WATER
27
airborne and underwater measurements show excellent agreement [45]. This model
allows the computation of spectral, sensor and solar zenith and azimuthal resolved
radiances and reflectance. The type and concentration of atmo spheric aerosols and
water constituents are introduced by the extinction coefficient, single scattering albedo
and the phase (volume scattering) function. The accuracy of the retrie val of water
constituents depends on the choice of optical properties.
The CASI was flown on 23 April 1995 over the lake ‘Tegeler See’ an important
fresh water reservoir for the city of Berlin, Germany. The instrument was operated in
spatial mode with six bands in the visible spectral range and a spatial resolution of 2.5
x 3.5 m during a period of highest biological activity. The overall approach was to (i)
convert the CASI sensor signals into radiances by a calibration and a radiometric
correction [46], (ii) to obtain the surface reflectance needed in the further analysis [47]
through atmospheric correction of the remotely sensed data, (iii) perform geometric
correction, (iiii) apply the inversion to the water areas of the images.
The radiative transfer model MOMO was applied for the water constituents CHL,
TSM and CDOM, using their optical properties [42]. The substance concentrations and
the observation/illumination geometry were varied systematically in a given range
(Table 8) to set up a look-up table for the further analysis whilst atmospheric parameters were fixed to realistic values.
TABLE 8. Substance concentration and observation geometry range for the MOMO run.
Variable
Range
Step size
# of values
Chlorophyll a
0 – 220 µg l-1
2 µg l-1
111
Suspended matter
0 – 100 mg l-1
2.5 mg l-1
41
CDOM absorption
0 – 35 m -1 @ 254 nm 2.5 m -1 @ 254 nm
15
Sun zenith angle
0° – 82.15°
15.70° – 19.11°
6
Sun-observer azimuth distance 0° –180°
11.25°
17
Observer zenith angle
0° – 82.15°
15.70° – 19.11°
6
At the end of the model simulation runs the look-up table contained 42 million multispectral reflectance vectors for six bands. The inversion problem considered here may
be solved by a simple look-up table approach, which is computational very time consuming, or by a faster neural network approach. The look-up table approach is straightforward and immediately applicable. The neural network approach consists of a sophisticated mathematical model and needs an intensive training of the network prior to its
application.
Using the look-up table approach both the substance concentrations as well as the
observation/illumination geometry has to be interpolated linearly. The neural network
approach can be used as a multi-variable non-linear regression method. It was found
more useful to represent look-up tables by a neural network including their non-linear
interpolating and fast application properties. Each inverse modelling method needs to
set up a look-up table where for a number of concentrations of water constituents based
on a bio-optical model the reflectance spectra are known. Then, the concentrations, and
their spatial distribution can be retrieved from remotely sensed reflectance spectra out
of the database. For the inversion of the remote sensing measurements using the lookup table approach for each multi-spectral reflectance vector, or pixel, respectively, this
corresponding simulated multi-spectral reflectance vector is searched for within the
look-up table, where the Euclidean distance finds its minimum. The observation/illumination geometry was interpolated linearly. The concentrations for chlorophyll a, sus-
28
A.G. DEKKER ET AL.
pended matter, and CDOM are the result of this inversion. A significant advantage of
this method is that no further pre-processing of the data is necessary. For the inversion
of the remote sensing measurements using the neural network approach several manipulations were applied to the model data. Several forms of noise were added to the
data.
Figure 10. Spatial distribution of chlorophyll a (left) and suspended matter (right) for lake ‘Tegeler See’ on
23 April 1995 retrieved by a look-up table approach.
Then, up to 1 ‰ random chosen input/output data sets were selected from the entire
look-up table, and used for the training of an adapted feed-forward radial basis function
(RBF) network [48]. 500 training steps in total were used for the training of a total of
25 neurons. The number of learning cycles should be high enough for a slow decrease
of the cost function to obtain a robust solution. Three different neural networks were
trained for each of the three substances: chlorophyll a, suspended matter, and CDOM.
The application of both inverse modelling methods to the CASI data reveals clear
and well structured concentration maps for chlorophyll a as well as for suspended matter (Figure 10 and 11; only chlorophyll is shown here for illustration purposes). The
main (north east) part of the lake shows low chlorophyll a and suspended matter concentrations while the attached waterways (south west) show higher concentrations. This
is confirmed by in situ measurements that are in the same order of concentration ranges. Beside this main feature a variety of local features can be extracted: high gradients
of concentration changes at local points (especially on the shore), traces of ships and
possibly bottom effects along the shore. For CDOM no useful results were achieved.
The spectral characteristics of CDOM overlap those of chlorophyll in the blue spectral
region, therefore it is not simple to extract that part of absorption which is caused by
CDOM.
The obtained results demonstrate the successful use of inverse modelling methods
using MOMO for the retrieval of the spatial distribution of at least two optically active
substances (chlorophyll a and suspended matter) by multi-spectral remote sensing measurements above inland waters. Compared to the look-up table approach the neural network approach produces similar results more efficiently, by requiring less time. The
higher effort for the training of the neural network in preparation of the application is
the only disadvantage. Nevertheless, both methods allow the use of time consuming
radiative transfer models for the simulation of more than one water constituent independently, and reasonable maps of the quantitative and spatial distribution of several
inland water quality parameters can be generated.
IMAGING SPECTROMETRY OF WATER
29
Figure 11. Spatial distribution of chlorophyll a (left) and suspended matter (right) for lake ‘Tegeler See’ on
23 April 1995 retrieved by a neural network approach.
3.2.2
Imaging spectrometry of optically deep estuaries
Estuaries are usually dynamic environments where freshwater meets ocean water. They
are among the most productive aquatic ecosystems. Pressures from often conflicting
uses, causes them to be studied extensively. [49] summarises the developments in the
period of 1990-2000 in actual applied remote sensing of estuaries. However, few
publications exis t on imaging spectrometry applications over estuaries, based on
inversion of analytical models. Some papers have been published on simple empirical
methods, but these have no multitemporal or multi-site applicability. The study by [50]
is a good illustration of the frequency required for begin ning to understand some of the
dynamics in an estuary using remote sensing. Although they did not use imaging
spectrometers, they collected in situ spectroradiometric reflectance measurements, a
suite of aerial photography taken on different days as well as Airborne Thematic
Mapper data (from which they mainly use the thermal band). With all this data they
studied frontal systems in the Tay Estuary in Scotland. Another example is the work by
[51] who studied the extent of a massive river Oder flood event using several types of
satellite imagery, but basing their algorithms on previously carried out measurement
and modelling of the underwater light field. They were able to produce maps of CHL,
TSM and CDOM from the experimental satellite MOS sensor.
3.2.2.1
Optical properties of estuarine waters
The majority of studies that address the issues of spectral reflectance of estuarine
waters, focus on the spectral optical properties of estuarine waters, with the aim of
developing algorithms for processing satellite images of these estuaries. The following
literature study discusses the studies that focussed on understanding the optical variability of estuarine waters. Next, the papers are discussed that proceed from these optical
properties to inversion methods. After that, the literature is discussed of studies that
continue this thread of analysis and actually apply it to i) non-hyperspectral satellite
data, ii) airborne imaging spectrometry data and iii) studies in preparation of MERIS.
30
A.G. DEKKER ET AL.
MERIS will be the first dedicated imaging spectrometry system for coastal waters in
space in the near future. It is arguable that MODIS and MOS sensors warrant dis cussion here as well. However these sensors are in essence multispectral systems designed
for ocean colour analysis and less for inland and coastal waters.
[52] studied the inherent optical properties of the Po Delta in northern Italy,
focussing on the absorption coefficients of algal cultures and field samp les and on the
backscattering spectra of field samples. From this analysis they propose combinations
of spectral bands (similar to that to be found on MERIS and MODIS) for determining
CDOM absorption in deltaic waters. [53], measured 119 spectra in coastal North Sea
and Dover Strait waters and were able to demonstrate that for this data set only 5
spectral bands were required to reconstruct the whole spectrum. They proved their
methodology on an independent data set taken in the Baie de Somme, a turbid tidal
estuary off the North France coast. They do admit that further investigation is required
if this method would work for algal blooms or for chlorophyll fluorescence. [54]
describe the optical properties for the Menai Strait (Wales, UK): this strait is an
excellent example of the variability in TSM and algal assemblages possible in one
water body. Overall high values of TSM varies with lunar variation due to tidal action.
The algal community composition starts off with a two-peaked spring bloom of first
diatoms and then a mixture of diatoms and phaeocystis; after the bloom the
concentrations decrease and small flagellates dominate. In summer a bloom of coccoid
cyanobacteria is possible. Kratzer tried to develop algorithms based on four spectral
channels only and concluded that this was an insufficient amount of bands to
adequately invert such variations in concentrations.
[55] present a further development of work initiated by [56] [57, 58] [59] which
relies on an iterative matrix inversion technique for measuring three optical components in coastal waters a(CDOM), phytoplankton through CHL and TSM, often augmented by one other variable such as aerosol retrieval or a backscattering coefficient of
the water. They all admit that these methods are computational intensive and probably
not very useful for operational inversion of a remotely sensed signal. [60], present a
study on detection of CHL, DOC and TSM in the estuarine mixing zone of Georgia
coastal plain rivers. The range of concentrations for these optically active constituents
was large and the source composition was highly variable ranging from terrestrial
dominated, to riverine dominated to almost ocean type waters. Of particular interest in
their study is the establishment of absence of covariance between most of these
parameters. Their results demonstrated that many empirically established relationships
would fail in remote sensing of such systems. They found poor correlation between
DOC and blue wavelength reflectance (thought to be the best spectral region for
CDOM/DOC measurements from remote sensing), but higher (inverse) correlation at
the green wavelengths. This effect may be explained through the blue reflectance signal
being confounded by multiple constituents (pigments, detritus, CDOM), whereas the
green signal was dominated by the combined effect of TSM and DOC/CDOM.
[61] discuss the inversion of a(cdom) 440 and the sediment refractive index for
nonchlorophyllous turbid coastal waters for the Rhone river mouth in France. [62]
developed a CDOM, phytoplankton and sediment bio-optical model for the Ebro River
in Spain. Forget (2000) combines these two data sets (Ebro and Rhone) and discusses
them in the context of two-flow radiative transfer modelling approach from [10]. They
specifically pay attention to the stratified system that may exist when a river plume
flows in to relatively stable coastal water. In 66% of their inversion cases they were
able to detect stratified water masses. Similar work was carried out by [63] who studied
IMAGING SPECTROMETRY OF WATER
31
the spectral re flectance and transparency of river plume waters in the Black Sea and the
Arctic Ocean using ship-borne and airborne spectroradiometric measurements. They
measured reflectance differences of a factor 10 between river plumes and the ocean
water at the other side of the front. They make a convincing case that river plumes
spreading into relatively stable ocean waters have three dimensional wedge shape that
needs to be considered when remotely sensing such phenomena. From airborne or
satellite imagery a high concentration fresh water TSM rich layer overlays a much
clearer ocean layer. [64] carried out spectral analysis of waters of the Pamlico Sound
estuary and the contributing river waters and found complex relationships relating
reflectance to TSM. Their aim was to develop an algorithm for relating Kd (PAR) to
reflectance measured by the NOAA-AVHRR 630 nm band; this was partially successful. [65] present preliminary results of a campaign intended to determine a method for
optically measuring the natural variability of TSM in a heterogeneous and dynamic
environment. For this purpose they use a suite of in situ optical properties
measurements, ferry based spectroradiometric measurements, airborne imaging
spectrometer (the EPS-A) flights and HYDROLIGHT to determine certain optical
parameters. They parameterise the [10]/ [12] model and thus are among the first to use
the more complete model presented in the section on modelling.
[66] followed the analytical approach to study an estuary in Kalimantan
(Indonesia) involving multitemporal use of SPOT and TM images. They determined
the spectral IOP’s of river, estuarine and ocean waters, developed a reflectance
simulation model, (similar to Figs 8 and 9) to estimate the full range of TSM concentrations possible in this system and subsequently derived algorithms for analytically
determining the TSM in this dynamic tidal estuarine area.
[67] present the first results of AVIRIS flights carried out in 1990 over the Tampa
Bay plume in the Florida coastal waters. Although at that time AVIRIS still had a
relatively low S:N they were able to demonstrate the use of hyperspectral remote
sensing for mapping the absorption coefficient at 415 nm and the backscattering
coefficient at 671 nm. This work was followed through by [68] who derived a model
for modelling hyperspectral remote sensing reflectance over a variety of case 2 waters
from the West Florida Shelf to the Mississippi River plume. Their model is based on
the Gordon model and takes into account CDOM , TSM, CHL, Raman scattering and
CDOM fluorescence, the only optical active variable missing is fluorescence by algal
pigments. In [69] a further development of this methodology is presented applied to
AVIRIS data of 1998 (when the S:N has increased dramatically as compared to earlier
versions). Because this paper discusses more the bathymetry effects of the substrate on
the signal, further discussion takes place in the section on optically shallow waters.
Mustard & Staid (1998) describe preliminary re sults of an AVIRIS flight over an
estuary in New England (US). They were able to accurately model AVIRIS derived
above surface reflectance. [70] followed the same methodology as [66] and were able
to closely match AVIRIS data flown over the Hudson/Raritan Estuary (New York) by
simulation using a bio-optical modelling tool parameterised for the Hudson/Raritan
estuary waters. [71] used an AISA imaging spectrometer over a coastal water area in
the Archipelago of the Baltic Sea, also using satellite and in situ data. They tested algorithms proposed for MOS, SeaWiFS and MERIS for Secchi Depth transparency, turbidity, CHL and TSM. They did not undertake any spectral measurements or modelling.
[72] used a CASI to map the sediment plume caused by dredging for a tunnel in the
Oresund, a coastal channel between Norway and Denmark. Their sediment maps were
compared with a 2-D hydrodynamic model for calibration purposes. After measuring
32
A.G. DEKKER ET AL.
the inherent optical properties, and parameterising the simplified Gordon model it was
decided to use the 544 nm spectral band as the band most suitable for determining TSM
concentrations.
A special issue of the International Journal of Remote Sensing (1999, Vol 20, no
9) is dedicated to the relevant science surrounding the preparations for the MERIS sensor on board of ENVISAT: the first dedicated imaging spectrometry type sensor explicitly designed for remote sensing of optically deep coastal waters. [73] used data from
the literature to develop an neural network based inversion model for MERIS. [74]
used the CASI airborne imaging spectrometer to develop and test algorithms for the
MERIS space sensor. [75] developed the neural network that will be implemented
operationally to derive case II water pro perties from MERIS data on a global scale.
[38] presents the research leading up the CHL fluorescence algorithm for MERIS, the
first space sensor to have dedicated fluorescence band centred at 683 nm.
3.2.2.2
Case study: The Western Scheldt Estuary
Because of the complexity, an estuary is an excellent example to demonstrate the importance of the measurement of the specific inherent optical properties to parameterise
the optical model. An estuary is a mix of two or more different water types, including
at least river water and coastal water. Additionally, different sediment types may appear due to resuspension of bottom material from tidal flats or wind and wave induced
resuspension of bed sediments. The history of wind direction and speed, tidal stages
and other meteorological conditions are thus of importance for the concentrations and
composition of suspended matter. The case considered deals with the analysis of the
typical concentration ranges and specific inherent optical properties measured in the
Western Scheldt estuary. These modelling results were applied to SPOT data for environmental baseline mapping of an intended area for construction of a [76]. Seasonal
measurements of water quality parameters show that the primary parameter describing
the optical properties in the estuary is the total suspended matter concentration (TSM).
For 1998, the TSM concentration ranged between 4 and 120 mg m-2 , whereas
chlorophyll a ranged between 2 and 20 µ g m-3 . [77] determined the mean seasonal
variation of TSM. On average, mean and standard deviation increase towards the east.
Dissolved organic carbon (DOC) concentration on the average increases from 1.5 near
the estuary’s mouth to more than 3.5 mg l-1 towards the River Scheldt. There is only
minor seasonal variability in the DOC concentration. The analytical model by [12] was
used for the light anisotropy factor f. This model (see table 6) is given by
IMAGING SPECTROMETRY OF WATER
R(0− ,λ ) = f
33
bb
( a + bb )
1
µd
1+
µu
µ d = 0.7 F + (1 − F )cos(θ 0 )
f =
(21)
µu = 0.5
θ 0 = a sin(sin(θ z ) / n)
A simulation model that calculates reflectance from the specific inherent optical
properties and the concentrations is defined as follows
*
a = aw + acdom + a*TCHL CTCHL + aTSM
CTSM
*
bb = 0.5bw + BbTSM
CTSM
acdom = acdom g440
(22)
The specific inherent optical properties (SIOP) are the absorption/scattering per unit of
mass. The SIOP are the fundamental properties required for optical models. It must be
noted here that TSM incorporates all particles not passing through a filter pore size
used to measure the CDOM concentration. Thus TSM consists of alive and dead organic particles and of inorganic particles. If possible it would be desirable to enhance the
above equation of absorption to
a = aw +a cdom+a*TCHL CTCHL + a*ISCIS+ a*d Cd
(24)
with a*IS and a* d the specific inherent absorption of inorganic suspended sediment and
detritus respectively. In the more simplified model used the main issue is to determine
the backscattering ratio for the fraction of light scattered backwards towards the upper
hemisphere. This ratio B is determined by fitting the measured reflectance spectra and
the modelled reflectance spectra using the [12] optical model, the SIOP and the in-situ
concentrations. In this way, an optical closure of the reflectance model is achieved. A
note of caution must be issued here: an error in the inherent optical property
determination or in the concentration measurements may in fluence the
parameterization of the SIOP’s and thus influence the estimation of B.
The in-situ measurements used here were taken in the scope of a project, initiated
to monitor the ecological effects due to drilling for a tunnel and the resulting sediment
dumping in the estuary using (among other techniques) remote sensing [76]. At five
stations the subsurface irradiance reflectance R(0−) was measured using a hand-held
Photo Research (Chatsworth, CA, USA) model PR-650 spectroradiometer, and water
samples were taken for subsequent laboratory analysis. Optical properties (absorption
and total scattering) were determined together with water quality parameters (TSM and
total chlorophyll (Figure 12)), resulting in the specific inherent optical properties for
each station (Figure 13).
34
A.G. DEKKER ET AL.
TCHL and TSM for the Western Scheldt on 10/3/99
10
TCHL (mg m-3)
70
10.8
TCHL
57.5
8.7 53.4
TSM
8
40.1
5.4
6
4
60
8.7
50
40
27.5
30
3.6
14.7
20
2
10
0
0
TN
LA
HW
BA
TSM (g m-3)
12
WS
Figure 12. Measured water quality parameters in the Western Scheldt
For all stations absorption of coloured dissolved organic matter (CDOM ) was consid erable (a(cdom) 440 ranging from 1.4 to 1.8 m-1 ). Apart from station ‘WS’ there is an apparent increase in TSM concentration in eastward direction.
TCHL specific pigment absorption
CDOM absorption
0.1
TN
HW
LA
BA
WS
1
0
400
a)
500
600
wavelength (nm)
-1
2
0.08
2
3
a* pig (m mg )
aCDOM (m -1)
4
0.06
TN
HW
LA
BA
0.04
0.02
0
400
700
500
600
wavelength (nm)
b)
TSM specific tripton absorption
TSM specific seston scattering
0.3
0.1
0.08
TN
LA
WS
HW
BA
0.04
0
0
400
c)
TN
HW
LA
BA
WS
0.12
0.4
a*t r (m2 g- 1)
b*ses (m 2 g-1)
0.5
0.2
700
500
600
wavelength (nm)
400
700
d)
500
600
wavelength (nm)
700
Figure 13. CDOM absorption (a), specific pigment absorption (b), specific seston scattering (c) and specific
tripton absorption (d) measured for five stations in the Western Scheldt estuary.
For CDOM absorption (Figure 13), all exponential slopes are similar, except for the
‘TN’ station, which has a significant lower exponential slope than the other four. This
may be attributed to the fact that ‘TN’ is closest to the sea, and might have a different
IMAGING SPECTROMETRY OF WATER
35
(older) type of dissolved organic matter, more typical for seawater. For the other four
stations, the CDOM absorption increases towards the river mouth. A large variation (up
to a factor of 2) can be seen in the specific pigment absorption. However, because of
the low CHL concentration found in the samples, relatively large measure ment errors
are expected in the absorption measurements. No arguments could be found to explain
in detail the variation in measured specific seston scattering. A standard deviation of
10% was found over the five stations. A variation of more than 20% was found in the
specific tripton absorption. A distinct difference can be noticed between the station
nearest to the coast and the more inland stations. This is probably due to the higher
fraction of organic matter. This analysis proves that specific inherent optical properties
can differ considerably within estuaries, and that spatial gradients of water quality variables can be present. The consequences for the reflectance spectra will be discussed in
the next paragraphs.
There is no unique solution for the modelling of R(0−) , unless the SIOP for that
particular region are known. Using the specific inherent optical for stations ‘TN’, ‘HW’
and ‘BA’, and the mean SIOP of the five stations, the modelled spectra are compared to
the measured reflectance spectrum. A standard deviation of 7 to 14% was found in the
modelled R(0-) spectra, confirming that the variance in SIOP has a major impact on the
modelled reflectance spectra. This analysis demonstrates that the specific inherent optical properties within an estuary can vary considerably leading to considerable differentiation in reflectance spectra. Optical modelling shows that in this estuary R(0−) is
optically dominated by the suspended sediment concentration, because the CHL and
the CDOM concentration are relatively low.
3.2.3
Conclusions for imaging spectrometry of optically deep inland and estuarine
waters
With the advent of imaging spectrometers that are suitable for water related investigations: the PMI, the AVIRIS and the CASI in particular at the end of the eighties, this
field of research commenced. First missions and associated research were explorative
rather than operational. From the mid-nineties onwards operational examples of multitemporal deployments of airborne imaging spectrometry systems over mainly inland
water targets started to happen in The Netherlands, Germany and Scandinavia. These
studies over inland waters were able to deal with the optically deep waters quite well
and produced meaningful results for up to 5 optical water quality variables such as
CHL, CPC, TSM , Kd and transparency. The combination of these results into ecological assessment and monitoring is gathering speed rapidly. The bio-optical models for
these water are becoming more sophisticated as well as the instruments with which to
measure the IOP and AOP properties. The CASI seems to be the instrument of choice,
due to its flexibility in platform and its programmable band sets. The number of CASIs
(20 by now) also determines the availability worldwide of course. In this field we see
developments towards more complete models but also towards methods to compute an
inversion of an imaging spectrometry scene using either analytical 1 to 3 band inversions, look up tables, using matrix inversion schemes or using neural networks. For turbid estuarine remote sensing, less use has been made of airborne imaging spectrometers
due to the very dynamic nature and the often large size of the estuaries. Most of these
36
A.G. DEKKER ET AL.
studies were intended as illustrations or experiments in preparation of using satellite
sensors to monitor these systems. As spaceborne imaging spectrometers become available this field of application is likely to evolve very fast.
3.3
OPTICALLY SHALLOW WATERS
Optically shallow waters are a special case in remote sensing of aquatic systems. It involves a measurable signal from a substrate, through the water column and through the
air-water interface. In all applications where a substrate is being mapped through a
water column, a bathymetry estimate is implicitly or explicitly involved. In the case of
a bathymetry estimate from remote sensing a substrate reflectance and a water column
optical depth is implicitly or explicitly involved. The bathymetry and the water column
optical properties are thus of importance to the following remote sensing applications
in aquatic environments: submerged macrophyte mapping, seagrass and macro-algae
mapping, coral reef mapping, sand, coral rubble and mud floor mapping and benthic
micro -algae mapping. Alternatively one may want to measure the optical properties or
the concentration of substances in a water column over a visible substrate. In that case
the bathymetry signal is seen as a component that must be corrected for (as with atmospheric correction) see e.g. [78].
This section is composed as follows: first there will be a discussion of the
literature on bathymetry and substrate mapping (usually brightly reflecting bottoms),
how to simu late reflectance signals over such complex environments and what the
requirements for imaging spectrometers are. The remainder of this section will discuss
more in detail the state-of-the-art in seagrass, macrophyte and coral reef mapping, with
a case study on seagrass, macro -algae and substrate mapping in a shallow coastal
environment near Adelaide, Australia.
3.3.1
Bathymetry and bright substrate mapping
The term “measurable signal from the substrate” sums up one of the problems in performing bathymetry or measuring a substrate cover such as seagrass. It will depend on
the spectral optical depth of the water column, on the brightness and spectral contrast
of the substrate as well as the signal-to-noise performance of a remote sensor whether
bathymetry or a substrate cover can actually be determined. If there is no measurable
influence of the bottom on the remotely sensed reflectance the water is considered to be
optically deep; if there is a measurable reflectance contribution from the substrate or
plants in the water column the water is considered to be optically shallow.
In the theory section the model by [29] is presented. Equation 18 expresses the
R(0-,H) as a function of the reflectance of an infinitely deep water column, the vertical
attenuation coefficients and the irradiance reflectance of the substrate. In Figure 14 a
simulation is presented of the following case. The same bio-optical model as used in
the inland water simulations calculates the irradiance reflectance for an infinite water
column. In this case the concentrations were: CHL = 1 µg l-1 , TSM = 2.1 mg l-1 and the
absorption by CDOM is expressed as a(cdom) 440 of 0.2 m-1 .
IMAGING SPECTROMETRY OF WATER
37
R(0-) calculated over a 10 % reflecting substrate at depths of
0.5 to 6 m and a R(inf) based on :
-1
-1
-1
CHL=2 µ g l,a (cdom)=0.2 m and TSM = 2.1mg l
0.10
3.0
0.09
2.5
0.08
0.07
2.0
-1
Kd (m )
R(0,H)
0.06
0.05
1.5
0.04
1.0
0.03
0.02
H=0.5
H=1
H=2
H=3
H=4
H=5
H=6
R(inf)
Kd
0.5
0.01
0.00
400
0.0
450
500
550
600
650
700
750
wavelength (nm)
-1
Figure 14. Simulation with CHL = 1 µg l , TSM = 2.1 mg l-1 and the absorption by CDOM is expressed as
a(cdom)440 of 0.2 m -1
This same simulation software calculated the associated Kd , using the [22] model
for relating Kd to the inherent and apparent optical properties. Next a substrate reflectance of 10% was assumed (spectral neutral- dark grey). The graphs in figure 14 show
the resultant R(0-) calculated with the substrate at depths of 0.5 m, 1, m, 2m...6 m.
From the results it can be deduced that for depths of 1, 2 and 3 m it would be required
to measure R(0-) with an accuracy of 1 % in the green wavelengths in order to distinguish R(0-) differences due to the substrate at 1 m depth intervals. In terms of R(0-)
this would equate to a discretisation S:N in terms of R(0-) of 100:1, which in turn leads
to a S:N requirement of about 200:1 for R(0+) (as 48 % of upwelling irradiance just
below the water surface is reflected into the water column). The required S:N in terms
of R(0-) to distinguish the substrate between depths of 5 and 6 m quickly increases
almost a factor 5 to 500:1 (R(0-) or 1000:1 (R(0+). These latter S:N specifications are
about the maximum currently attainable by systems such as AVIRIS and CASI, flown
under ideal circumstances. It is also evident from the graph that under these simu lated
conditions most information from the substrate reaches the water surface in the wavelengths of 500 – 600 nm. In reality the situation is more complex as we now assumed a
flat water surface, a homogeneous diffuse light field and a neutral spectral behaviour of
the substrate (See Table 9 for substrate reflectances).
TABLE 9. Values of high reflectance of bottom substrate from literature.
Authors
type of substrate
RB at 400 nm
RB at 700 nm
Alberotanza et al. 1989
coral sand
28%
60%
Lee et al. 1994
offshore sed.
40%
56%
38
A.G. DEKKER ET AL.
Maritorena et al. 1994
Gould & Arnone, 19971
1
coralline sand
white sand
28%
29%
60%
41%
29 % at 412 nm and 41% at 670 nm.
For these highly reflecting bottoms in general there is an increase in reflectance towards higher wavelengths; the sharp increase of pure water absorption beyond 630 nm
will effectively attenuate this highly reflected light, both on its downward as its upward
journey through the water column. Some values for the clearest natural waters are those
given by [79] who give a maximal wavelength of penetration of 55 m for z90 at 475
nm in the Sargasso Sea.
In many coastal and inland water applications a bathymetry estimate from remote
sensing has to deal with the following factors:
• A variable substrate coverage, in the case of macrophytes/seagrasses/macroalgae/corals with a canopy height and composition as additional variable
• A variable water column composition
• A variable air/water interface
• Variable geometry of direct sunlight, diffuse skylight, wave reflections and
refraction and the sensor viewing geometry( field of view related zenith angles and
azimuth angle)
For the depth of the bottom to be retrieved accurately each of these factors has to be
determined through measurement or estimation. In principle a hyperspectral sensor
with sufficient spectral bands has the capacity to independently determine each of these
factors (Paredes & Sparo 1983), (as well as the atmosphere-which we ignore in this
discussion). Further reading on the subject of the radiative transfer aspects of bathymetry may be found in: [80] [79] [81].
Lee et al. (1999) developed a semi -analytical model, based on above surface
remote sensing reflectance Rrs, defined as the ratio of the water leaving radiance to
downwelling irradiance just above the water surface, for shallow waters. They used
HYDROLIGHT to parameterise the model. In the model downward and upward
diffuse attenuation coefficients are explicitly described as functions of the absorption a
and backscattering coefficients bb , the bottom reflectance Rb and the water column
depth z. The only assumption made is the shape, not the magnitude, of the bottom
reflectance. A set of coefficients is introduced by a computer model for a, bb , Rb and z.
Through optimi sation the modelled and the measured spectra are matched as closely as
possible. This results in values for a, bb , Rb and z. Because the absorption coefficients
are split into absorption by phytoplankton, the sum of CDOM and detritus and pure
water absorption and the backscattering into particulate and pure water backscattering,
this methodology also derives concentrations for algal pigments, the sum of CDOM
and detritus absorption and the particulate matter concentrations (based on knowledge
or estimates of the specific inherent optical properties). The model also incorporates the
average cosines for up and downwelling radiance, the volume scattering function, and
the sensor, solar and sky radiance geometries. The results were as followed: For
computer simulated data the retrieved depth was accurate to within 5% (N=33) for a
bathymetry range of 2 to 20 m. For field data in Florida bay it was accurate to within
11% (N=37) for a range of 0.8 to 25 m. For data outside of Florida Bay it was within
8% for depth (N=33). As is the case for most remote sensing analytical or semianalytical retrieval methods the authors see most improvement possible in better
IMAGING SPECTROMETRY OF WATER
39
measurements or estimates of the inherent optical properties of the optically active
components.
[82] used data from AVIRIS flown in 1993 over Tampa Bay and in 1996 over the
Florida Keys in a neural network system to establish quantitative empirical
relationships between depth and remotely sensed spectral radiance. No use was made
of any type of physics based processing of the data to remove atmospheric or air/water
interface effects. They used an extensive set of tens of thousands of depth soundings
spanning a 1952 through to 1995 time range for Tampa Bay and for 1997 in the Florida
Keys area. Relative to these depth data sets the RMS was 0.84 m for the Tampa Bay
and 0.39 for the Florida Keys area. This RMS could have been reduced if the
bathymetry in situ data deeper than 6m had been omitted as there was no bottom
visibility beyond that depth. Also the AVIRIS was twice as sensitive in 1996 as in
1993. They also trained the neural network on both data sets which resulted in a RMS
of 0.48 m. Apparently a neural network has generalisation capacity based on this
research as the Tampa Bay and Florida Keys waters are very different with different
substrates. These results indicate the effectiveness of neural networks for fast remote
sensing data processing.
3.3.1.1
Bathymetry in inland waters
[83] report on an AVIRIS flight in 1990 over the oligotrophic Lake Tahoe. They
related the bathymetry of Lake Tahoe to upwelling spectral radiance recorded by
AVIRIS using a two-channel multiple linear regression with AVIRIS data at 490 and
560 nm. They assumed a uniform bottom reflectance and were able to map bathymetry
between the 3 and 10 meter contours.
George (1997) used a CASI to map bottom topography features in Lake Windermere in the UK. The data was flown in 1989 with one of the first versions of the CASI.
According to George the shallow water had a higher reflectance in the nearby infrared
due to increased bottom reflectance. This appears to be valid down to a depth of 20 m.
If we look back at the simple simulation in Figure 14, this seems almost impossible as
the Kd for pure water at 700 nm is already 0.57 m-1 and increases to 2.73 m-1 at the
wavelength interval of 746-759 used for this study. It is more likely that George found
a correlation between resuspended sediment and reflectance in the nearby infrared and
mis interpreted the correlation.
[84] performed experiments in mesocosms to investigate how brightness impacts
the spectral response of a water column under varied suspended sediment conditions.
They used a white aluminium panel as a bright bottom and a flat-black tank liner as the
dark bottom. 16 levels of TSM were used. The major findings were that the bright
bottom had the highest impact in the visible wavelengths. When suspended sediment
levels reached 100 mg l-1 there was no measurable signal from the bottom anymore.
They found that substrate brightness had minimal impact at wavelengths between 740
and 900 nm, suggesting that these wavelengths are best for measuring suspended
sediment for remote sensing. These findings are in direct contrast to those by George,
based on the physics the conclusions by [84] seem more valid. However they should
also have realised that the amount of passive solar irradiance at these longer
wavelengths becomes lower and the sensitivity of Silicon based CCD arrays as used by
many VIS-NIR spectrometers decreases rapidly at these wavelengths. Thus their
40
A.G. DEKKER ET AL.
conclusion that these wavelengths are best for remote sensing of TSM only applies to
the case where the water depth is low, and the NIR wavelengths cause high attenuation
due to the high water absorption.
3.3.1.2
Conclusions Bathymetry and substrate mapping
The [85] methodology seems to be the most advanced approach if the optical properties
are more or less known and there is some indication of the spectral shape of the
substrate reflectance. The neural network approach by [82] seems to be the alternative.
There are two ways of training the neural network: by use of a massive amount of in
situ data or by training it with a radiative transfer model such as described by [85] and
currently available through HYDROLIGHT. Of course the HYDROLIGHT model
requires adequate parameterisation for the water bodies under consideration. The more
empirical approaches loose their attractiveness once they are analysed with respect to
the physics involved: often the empirical applications show ambiguous results. In many
cases it is questionable whether a high signal was recorded due to substrate visibility or
due to resuspension in a shallow water area.
3.3.2
3.3.2.1
Macrophyte/seagrass and macro-algae mapping
Introduction
Aquatic macrophytes provide an important habitat in both marine and freshwater environments. Species of macrophytic algae and those of higher plants can play a major
role in providing a food resource for coastal and pelagic animals from zooplankton
through to waterfowl. In addition, they may provide important refuges or shelters from
grazing, provide nursery beds to other plant and animal communities and may help
stabilise otherwise unstable or eroded shores.
Freshwater aquatic macrophytes are typically separated into three groups based
upon their principal growth habit, a classification that can also be useful for marine
species. Some species remain principally submersed throughout their life cycles. Floating leaved plants such as water lilies and some macrophytic marine algae have leaves
or blades floating on or near the water surface. Emergent macrophytes penetrate
through the water surface. Despite their ecological importance, human disturbance in a
number of forms represents the largest threat to macrophyte-dominated communities
and the life they support ([86] [87] [88] [89].
The overall condition of macrophyte communities in both lacustrine and marine
habitats can be directly related to the quality of the waters in which it lives Nutrients
can stimulate the growth of phytoplankton and epiphytic algae that reduce light levels
necessary for seagrass survival. There is evidence to suggest that it is extremely difficult, if not impossible, for seagrass species to re-colonise areas after they have died [90,
91]. The need to map and monitor the distribution, abundance and diversity of these
areas is therefore, of prime importance in assessing the status of these coastal systems.
IMAGING SPECTROMETRY OF WATER
3.3.2.2
41
Literature review
Traditional approaches to surveying the distribution and biomass of aquatic
macrophytes have relied on potentially destructive quadrate and transect based methods
similar to those typically used for ground-based survey of plant matter in terrestrial
ecosystems. Such methods are also time consuming if the distribution of macrophytes
over large areas are to be accurately mapped [92]. For mapping distribution over large
areas remote sensing could offer a time-saving and potentially cost-saving nondestructive alternative [93] [94] [95].
A few studies have reported spectra for a range of species and growth habits (e.g.
[96] [97] [98]. Figure 15 presents reflectance spectra measured over 7 types of freshwater macrophytes in Cefni Reservoir, North Wales, UK. These spectra are similar to
those presented in [96], and both studies can be used to draw general conclusions about
the nature of reflectance from the different growing habitats for both freshwater and
marine macrophyte species. Absolute reflectances from submersed species (e.g. Isoetes
lacustris) are generally low, often lower than reflectance from deeper or background
open water areas. Reflectance generally declines further with increasing wavelength.
Red-edge increases in reflectance are generally not discernible except in the shallowest
of waters as a result of high absorption by the water itself.
Remotely sensed reflectance over submersed macrophyte communities is
influenced by both varying water column depth and turbidity. For example, Figure 16
illustrates a number of simu lated seagrass reflectance compared to measured in situ
reflectance of Posidonia around the coast of Sicily. Comparisons between the
simulated and field-measured spectra indicate the relative consistency in spectral shape
of the simulated output. However, much of the shape of these spectra is influenced
principally by the influence of the water column rather than by seagrass reflectance,
particularly the strong absorption of light in the longer wavelengths. Further
simulations to investigate the influence of water column depth indicate that much of the
useful signal reflected from submersed plant material is rapidly attenuated with
increasing depth of the water and the bottom reflectance is diminished as it is filtered
through the water column [99]. All remotely sensed measurements of reflected radiance
over submersed species will be similarly influenced by water column effects which will
ultimately affect the accuracy with which spectral classifications of individual species
can be performed [100].
Over the last two decades a number of equations have been derived which have
allowed for the either the calculation of water column depth, or in revised form the
calculation of bottom reflectance after accounting for water column depth and attenuation [101] [102] [103] [104] [30] [105] [106] [29] [107] [108]. These techniques rely
on the fact that the bottom reflected radiance is approximately a linear function of the
bottom reflectance and an exponential function of water depth.
42
A.G. DEKKER ET AL.
50
Batrachium fluitans (S)
Scirpus
40
Polygenum (F)
Reflectance (%)
Isoetes (S)
Phragmites/Solanum
30
Phragmites
Equisetum
20
Deepwater
10
0
400
500
600
700
800
900
1000
Wavelength (nm)
Figure 15. Above surface reflectance measured over different aquatic macrophyte species in Cefni Reservoir,
North Wales, UK. Measurements were made using a Spectron SE590 spectroradiometer with a 15° field of
view and referenced to above-surface measurements of downwelling irradiance obtained using a second
sensor head fitted with a cosine correcting hemispherical receptor (from [97].
1
8
0.8
0.1
6
0.6
4
0.4
2
0.2
0
400
0.2
0.3
Reflectance (field)
Reflectance (Modelled)
10
0
500
600
700
Wavelength (nm)
Figure 16. Comparison of sub-surface Posidonia reflectance (thick line) measured in Mediterranean coastal
waters around Sicily with those simulated for increasing plant biomass using a Monte Carlo reflectance
model [109]. Differences between the measured and modelled spectra are attributable to the unknown
parameters of the submersed vegetation in the field, the influence of epiphytic material and presence of dead
leaves which were not included in the model.
Although uptake in the use of these approaches has been slow, the use of water column
correction routines is now considered standard for the objective measurement of habitat
change using remote sensing for routine monitoring of bottom habitats. Such a process
has been shown to significantly improve the accuracy of classification of such habitats
[110] [81].
IMAGING SPECTROMETRY OF WATER
43
In order to address the confounding influence of the water column in optically
shallow waters, knowledge of the vertical attenuation for diffuse downwelling irra diance (Kd ); the vertical attenuation for diffuse upwelling irra diance(Ku ) and for κ the vertical attenuation coefficient for diffuse upwelling light originating in each layer of the
water column, is necessary. However, few studies have used independently acquired
estimates of attenuation, with most extracting Kd values for relevant bands directly
from their imagery in areas of uniform bottom type (e.g. sand) and known depth, assuming that these may be employed as a constant for attenuation across the image data.
The work by [85] presented in the section on bathymetry is an example of trying to
infer IOP’s and AOP’s from the imagery itself making as little assump tions as possible.
[111] measured the bi-directional reflectance distribution function (BRDF) for varied
benthic surfaces in Bahamas and pointed out that in specific cases a specular
component does exist, while this parameter is generally assumed to be Lambertian. An
investigation into the spatial variation in attenuation in a typical tropical region was
undertaken by [112]. Measurements of gross spatial variations in downwelling
attenuation around two Caribbean islands indicate a four-fold variation in light
attenuation in shallow littoral regions alone. Spectral attenuation measurements
suggested that this variation was largely the result of scattering by particulate matter
rather than varying concentrations of dissolved organic matter. This finding suggest
that the results of studies where single measure ments of 'average' attenuation have been
used to depth-correct remotely sensed imagery should be interpreted with a high degree
of caution.
Remote sensing techniques for bottom studies can include both sub-surface and
above-surface techniques. Aerial photographic techniques have been commonly used
for assessing the extent of macrophytes in both marine and freshwater bodies (e.g[113,
114] [115, 116] [117] [118]. However, if manually undertaken, photo-interpretation of
such media is more often than not a subjective process and applied as an operational
tool can be expensive due to the time taken for effective photo-interpretation [119].
Remote discrimination of substrate types in relatively shallow coastal waters has
been limited by the strong attenuation influence of the water column [120, 121] and by
the poor spatial and spectral resolution of available sensors[122]. The principle
component analysis (PCA), a common used tool in multispectral analysis as data
reduction technique, has been successfully applied in many studies to determine the
most representative spectra [121] [123] [124]. A number of researchers have
investigated the application of airborne imaging spectrometers for mapping benthic
plant species. [125] used an airborne CASI with ground based spectroradiometry, to
test the ability of imaging radiometers to describe the principal seaweed and seagrass
beds along the coast of
Brittany (France). Their analysis showed variation in relation to pigment
characteristics, vegetation structure and environmental conditions. An algorithm was
developed to discriminate the dominant species in which visible wavelengths allowed
good discrimination between green, red and brown algae and infrared wavelengths
allowed separation of brown species, seagrasses and floating seaweed.
The spectral reflectance characteristics of features within a submerged coastal
environment are optically similar, so confusion can arise in identification. High spectral
resolution sensors are required to perceive the subtle difference, which is demonstrated
44
A.G. DEKKER ET AL.
through analysis of in situ measurements in a study by [126] Here the proportion of
correctly identified spectra using first derivatives is 75% with the main source of error
resulting from the inability to identify algae-covered surfaces. Similarly, [97] evaluated
the ability of Daedalus Airborne Thematic Mapper (ATM) imagery for mapping the
dis tribution of freshwater aquatic macrophyte species in Cefni Reservoir on the Isle of
Anglesey in the UK (see figure 15). Dis criminant analysis indicated that good
identification of macrophytes could be achieved by a combination of green, red and
near infrared wavebands. A minimum distance supervised classifier using ATM bands
3, 7 and 8 showed separation of the species surveyed. The results indicated that
airborne remotely sensed data have good potential for monitoring freshwater
macrophyte species.
Hymap Imagery has been used for the mapping of seagrass distribution in a coastal environment in the upper Spencer Golf of south Australia [122]. The data have been
atmospherically and track illumination corrected. Three feature extraction techniques
were evaluated: band ratios, PCA and spectral angle mapping (SAM). SAM was
assessed to reliably discriminate features from selected endmembers [127]. Another
technique commonly used to extract qualitative maps of submerged aquatic vegetation
is to measure reflectance spectra of different substrate in situ or in laboratory in order
to create spectral libraries of seagrass meadows and substrate [128]. Higher derivative
analysis has been found useful in resolving overlapping spectra, where the fourth
derivative peaks occur at the same wavelength as those of the original spectrum
enhancing useful information [99].
However, a radiative transfer model is an essential requirement to account for
water attenuation and multiple backscatter of a reflective sea bottom[81]. [128] used a
two-flow model to create an artificial spectral library of computer-simulated remote
sensing reflectance and compared them to in situ spectra. In this analysis they were
able to classify hyperspectral images by bottom type and bathymetry. However a twoflow model requires accurate signatures for either bottom albedo or water depths.
Spectral libraries maybe a viable alternative to the two-flow model for distinguishing
sediment from other bottom types, such as corals or seagrass from remote sensing
images. In a study in South Australia CASI imagery has been used for mapping benthic
species (see the case study at the end of this section, adapted from [129]. From the
CASI imagery it was possible to discriminate different seagrass species, as well as
various coverage of seagrass, Ulva and epiphytes within a pixel with high accuracy.
Considerable interest is being shown in hyperspectral sensor data, allowing the
ability to overcome the spectral limitations of conventional spaceborne sensors such as
Landsat TM and SPOT (e.g. [130]). Airborne hyperspectral sensors such as AVIRIS,
HYMAP and CASI show promise and the space-borne hyperspectral instrument
HYPERION is worthy of further investigation. This latter narrow-swath, 30 m spatial
resolution instrument is the forerunner to several high spectral resolution sensors which
will be launched in the near future (e.g. NEMO and ARIES ) and with significant
potential for mapping aquatic vegetation. The new genera tion of high spatial resolution
satellites such as IKONOS, also show considerable promise for monitoring
macrophytes.
In conclusion, the challenges facing digital above-surface approaches to the survey of benthic vegetation on both freshwater and marine environments, include issues
associated with spatial, spectral and radiometric resolutions, surface effects and water
column depth and turbidity. The ability of such techniques to monitor the extent and
state of seagrass and lacustrine meadows is thus dependent on an understanding of the
IMAGING SPECTROMETRY OF WATER
45
interaction between irradiance, the optically active constituents of the water column,
the vegetation and the bottom sediment. Optical modelling research is perhaps the most
effective approach in order to better understand the nature of above-surface reflectance
from macro phytic stands, in particular if quantitative information is to be derived.
Further research is required on the spectral signatures of both freshwater and marine
macrophyte species at appropriate spatial scales for the development of appropriate indices for routine application. There is evidence that it will not be possible to accurate
assess macrophytic biomass ([109], but c.f. [131] [94]. New developments in high
spatial and spectral resolution instruments show considerable promise for improved
mapping capabilities from space. The synergy between these sensors and the
information offered by subsurface acoustic instruments also warrants further
investigation.
3.3.2.3
Seagrass mapping case study
A CASI was flown, in spatial mode using 12 spectral bands (see figure 17) over a
portion of shallow coastal water near a sewage treatment plant near Adelaide, Australia
in April 1999. The spatial resolution was 0.8 m, 2.0 m and 5.0 m. At the same time
field reflectances of the benthos and estimates of the water’s inherent optical properties
were determined. Another field trip to validate the classification (see figure 18) was
undertaken in February 2000, in similar seasonal conditions.
The overall approach was to model the atmospheric and in-water radiative transfer, the
air-water interface and to use measured benthic reflectances. After atmospheric and air
water interface correction, the image data was classified. A migrating means minimum
distance classifier was seeded with 26 classes and then allowed to generate others to a
limit of 256 classes. After final iteration 99.97% of pixels were classified. Post
classification included canonical variate analyses and incremental sum of squares
analyses. The 256 basic classes spawned from the classifier were then aggregated into
46 super classes using a spectral tool (SpecTool) and statistical summaries of the class
variance. The SpecTool was used to model the field spectral data ‘through’ the optical
water column model. It uses as inputs the spectral reflectance of pure specimens collected on the ground (endmembers), water depth information, and the optical water
model, which uses the measured inherent water quality properties. Class signatures are
compared to the end member spectra adjusted for the appropriate depth [132]. In the
classification there were a number of classes that were unidentifiable using the fieldderived reflectance. These classes were labelled ‘unknown’.
46
A.G. DEKKER ET AL.
UniSpec Field Reflectances
Bolivar
5000
Reflectance *10000
4000
Heterozostera, mature
3000
Heterozostera, young
2000
Posidonia
1000
Class 24
0
400
450
500
550
600
650
700
Wavelength (nm)
750
800
850
900
Figure 17. Effects of increased water absorption due to an increased water column depth. The poor fit of the
basic Class 24, in Figure 17, to the Posidonia spectrum above 675 nm maybe attributed to increased water
absorption due to an increased water column depth . Each of the markers is the centre wavelength location of
a spectral band of the CASI.
The results of the classification are shown in Figure 18. To understand some classes that appear spectrally distinct and unmatched to library spectra a validation experiment was performed in February 2000. The region of interest was directly off the sewage treatment outlet. Random transects and point dives were carried out with differential GPS position recorded, surface and underwater photographs taken and spectral
measurements taken of the benthos using a UniSpec Radiometer equipped with an
underwater 10 meter cable. Particular attention was paid to the areas where the class
was labeled ‘unknown’. With the help of several taxonomic keys [133] [134] it was
possible to identify broad groups of seagrasses and macro-algae in the unknown areas
of the 1999 image. However, it was not possible to separate Heterozostera tasmanica
from Zostera muelleri, as the distinction is quite subtle. Because H. tasmanica is
predominantly intertidal and Z. muelleri is pre dominantly subtidal, the species could be
labelled correctly. The field-based separation between the Posidonia ostenfeldii and
sinuosa was also quite difficult, but probably involved sinuosa, as it lacked the
branching type structure. Quite obvious was P. australis, which had much broader and
thicker leaves. Amphibolis antartica was identifiable through the alternate branching
structure of the leaves.
The validation results indicate that the classification of the airborne imaging
spectrometry data correctly identified 72% of the areas surveyed 11 months after the
imagery was flown. The validation results indicated most of the ‘unknown’ classes to
be a mix of the Ulva and Heterozostera (Figure 19).
IMAGING SPECTROMETRY OF WATER
47
Figure 18. Hyperspectral Image Classification of benthic substrate species and cover for the February 1999
Bolivar Coastal site (from [129]).
3.3.2.4
Conclusions imaging spectrometry for seagrass mapping
Hyperspectral airborne data allowed discrimination of two seagrass species (Posidonia
and Heterozostera/Zostera spp) and the green algae Ulva with 72% accuracy. It was
also possible to discriminate various coverage’s of seagrasses and Ulva and epiphytic
cover within a pixel. Epiphytic infestations were found to complicate the spectral identification of the seagrasses, however linear mixing models were able to provide separation. This level of accuracy could not be obtained without modelling the atmospheric
and in-water radiative transfer, air-water interface and benthic reflectance.
Aerial photography and hyperspectral imagery quality is dependent on weather. Hyperspectral imagery offers a significant improvement of aerial photos and can be temporally compared after atmospheric correction.
48
A.G. DEKKER ET AL.
Figure 19. Intertidal zone where species are predominantly Ulva and H. tasmanica and was labeled
‘unknown’ in the classification
Hyperspectral imagery processed using this methodology can provide a much higher
level of spectral separation than colour aerial photography taken in the same area. This
methodology, once fully operational provides an objective classification tool, less
dependent of operator interpretation than aerial photography interpretation.
3.3.2.5
Coral Reefs
Coral reefs are the most spectacular and divers e marine ecosystems on the planet today
and degradation of coral reefs is a major environmental problem worldwide. Coral
reefs are often located in remote areas and may have large extent. There is a strong
management need for the mapping of coral reefs and for cost-effective assessment of
the environmental health or condition of reefs.
Remote sensing could be a valuable tool for mapping and monitoring coral reefs
and related ecosystems. A satellite image covers a large geographic area and has good
temporal coverage of most areas of interest. In the past, satellite sensors had limited
spectral and/or spatial resolution. The use of hyperspectral airborne instruments has increased the number of coral reef substrates classes that can be discriminated [135] [95,
108] [136] [100, 110].
There are two different approaches in remote sensing analysis of coral reefs. The
first approach is image based. The image is divided into as many as possible classes
[137] and each class is named (i.e. mainly sand and rubble with some living corals)
after extensive ground truth measurements. This has been the only method suitable for
interpretation of satellite images because the 3 to 7 broad wavelength bands (often not
suitable for substrate discrimination) do not enable substrate mapping by their
reflectance spectra. Hyperspectral images allow the collection of endmembers for
spectral unmixing from the image itself and this has been used to increase number of
bottom classes resolvable using remote sensing [129]. The other approach is to measure
reflectance spectra of different substrates in situ or in laboratory. This allows the
IMAGING SPECTROMETRY OF WATER
49
creation of spectral libraries of coral reef substrates. Some research has been done to
collect reflectance spectra of different coral reef benthic types [29, 138, 139] [140]
[141] [128] [142] [143] [99] [144]. Most of the results in these papers have a limited
regional scope: typically 3-4 species of living corals, a few different algae species and
dead corals or sandy bottoms are measured etc. In a few cases a more systematic approach has been taken to create spectral libraries of coral reef benthic types containing
reflectance spectra of hard and soft corals, sponges, seagrasses, dead coral covered with
different algae, and different sand types ([121, 126, 138] [145-147].
3.5
3
Normalized reflectance
2.5
2
1.5
1
0.5
0
350
400
450
500
550
600
650
700
Wavelength, nm
Figure 20. Variability in reflectance spectra of different living coral species. The “average living coral”
spectrum is represented by the thick solid line. Reflectance values are normalised to a wavelength of 515 nm.
Assessing the biological state of the coral reef requires at least a capability to dis tinguish living coral (with symbiotic algae, zooxanthellae) from dead coral covered by
overgrowth of benthic algae. Changes in substrate type from living corals to benthic
algae, seagrasses, sponges or rubble can give valuable information to reef ecologists
about the state of the reef. The spectral information required to achieve this discrimination has to be obtained through water with variable depth and optical quality. Methods
have been developed to remove water column effects from remote sensing image (see
section on bathymetry). Once a hyperspectral library of substrates is available and there
is sufficient knowledge about the inherent optical properties of water under investigation it becomes possible to use spectral modelling or other analytical methods to remove water column effects and improve discrimination of different coral reef substrates. It may even make simultaneous detection of water depth and type of benthic
substrate possible from remote sensing images [103] [128]. Kutser et al. [146] [147]
have measured reflectance of about 140 different coral reef substrates in mid section of
The Great Barrier Reef. About half of the studied substrates were living Scleractinians
(reef build ing hard corals) of different species (Figure 20). Reflectance spectra of living
hard corals are variable both in shape and value. Maxima in reflectance spectra are
50
A.G. DEKKER ET AL.
typically located close to 610 nm and reflectance values at this wavelength vary
between 0.06 and 0.42. There are two “shoulders” in reflectance spectra of most
measured living hard corals located around 580 nm and 650 nm. Reflectance is
relatively lower in the shorter wavelength region of 350-500 nm. In most cases, the
between species variability in the reflectance spectrum curve is relatively small in this
spectral region. Almost all Acropora species (as well as some others) have a distinct
minimum in the UV part of the spectrum (370-380 nm).
[121] [126] [138] [141] [128] [142] have measured reflectance spectra of corals
living in other regions of the world ocean. In general the reflectance spectra are similar
to those by Kutser et al. [146, 147] from the Great Barrier Reef. Exception here are
reflectance spectra measured by [142] in the Red Sea. The Red Sea corals often show a
peak close to 570 nm and shoulders near 510 and 630 nm.
Normalised reflectance
3
2
1
0
350
400
450
500
550
600
650
700
Wavelength, nm
Figure 21. Reflectance spectra of standing dead corals and coral rubble covered with algae “turf”
(normalized to 515 nm). The average reflectance spectrum is the thick solid line.
Dead coral skeletons, rocks and even sand (in places with low wave action) are often
overgrown with filamentous algae (Figure 21). Most of these algae belong to red and
green algae that have optical properties that differ from zooxanthellae (dinoflagellates)
living in symbiosis with the coral polyps. The most distinctive spectral feature of algae
covered substrates is a double peak in the reflectance spectra at around 600-605 and
650 nm. There is tendency that long time dead coral (covered with thick layer of algae)
is darker (reflectance max 0.05-0.15) than recently dead coral (bright reflective skeleton with thin film of algae). Nevertheless, long time dead corals covered with crustose
coralline algae have reflectance values (max 0.35) close to most living corals [146,
147]. [140] have found that long time dead corals covered with coralline algae have a
higher reflectance than most living corals. This phenomenon can be explained with the
low (max 0.17) reflectance of living Porites corals studied by them. It must be noted
IMAGING SPECTROMETRY OF WATER
51
that algae invading the coral skeleton after the death of coral polyp and algae covering
long time dead corals are not the same.
Normalised reflectance
2
1.5
1
0.5
0
350
400
450
500
550
600
650
700
Wavelength, nm
Figure 22. Normalized reflectance spectra of sandy bottoms in a coral reef environment.
The reflectance of bright white coral sand is much higher than the reflectance of other
reef substrates (Figure 22). The shape of the spectrum is relatively flat with monotonous increase towards greater wavelengths [29] [138] [146, 148]. This allows bright
white sand to be relatively easily separated from other bottom types. Sand is often
covered with coral rubble (overgrown with algae) or fine silt and other sediments
(containing algae). This could explain why sandy bottom reflectances are not always
higher than living or even dead coral reflectance. There is often a double peak in sandy
bottom reflectance spectra, which is similar to the spectra of an algae-covered bottom,
indicating the presence of algae on surface of sand particles [146, 148] [128] [143].
[138] found the reflectance spectra of recently bleached corals to be higher than
those of living corals and similar to the reflectance spectra of bright white coral sand.
Kutser et al. [147] [146] [148] were able to study an Acropora hyacintus colony that
had been dead for just one month (eaten by crown-of-thorn-starfish). This was already
covered with thin film of green filamentous algae and the reflectance spectra were with
the double peak typical to dead corals. The fact that the recently dead coral spectrum
differs significantly from the living coral reflectance is important in monitoring of
bleaching events.
Reflectance spectra of soft corals are similar to those of hard corals [146, 147]
[148]. There is a minimum in the UV (370-380nm) of the reflectance spectra of most
soft corals and peak close to 610 nm. The shoulder near 650 nm tends to be lower than
the shoulder close to 580 nm in case of soft corals, whereas the shoulder reflectance
values do not differ much in case of hard corals. Thus the difference between soft and
hard coral reflectance is small. This makes differentiation between soft and hard corals
difficult using optical methods. The reflectance spectra of different macro-algae species
52
A.G. DEKKER ET AL.
are highly variable and are amongst the lowest of all substrate types. Reflectance values
do not exceed 0.09 in most cases. Some of the reflectance are similar to filamentous
algae covering dead corals (red algae, blue-green algae), some to corals (bubble algae
and brown algae) and some differ from all other reef substrate types corals [147] [146]
[148]. This makes reliable discrimination of macro algae from other substrate types
difficult. [99] have measured reflectance of fleshy and coralline red algae and brown
algae in the Caribbean. Their results are similar to Kutser et al. [147] [146, 148]: red
algae have double peak in their reflectance spectra (near 605 and 650 nm) and brown
algae reflectance spectra are similar to living corals reflectance (peak near 605 and
shoulders near 580 and 650 nm). [29] measured the reflectance of five algae species on
coral reefs in French Polynesia. Their results agree well with above mentioned authors:
brown algae (Sargassum sp. and Turbinaria sp.) are spectrally similar to living corals,
green algae (Boodlea sp.) reflectance is different from all other substrate types and is
similar to reflectance of Chlorodesmis measured by Kutser et al. [147] [146] [148] in
GBR, and red encrusting algae (Porolithon onkodes and an unidentified algae) spectra
are with the double peak (near 605 and 650 nm). [129] found that some seagrasses
(Heterozostera, Posidonia) have reflectance spectra similar to Boodlea and
Chlorodesmis (very low values below 500 nm, maximum near 550). They have also
found that seagrasses covered with an epiphyte (such as Gifforedia) have reflectance
spectra with the double peak similar to dead coral covered with filamentous algae. The
number of reflectance of different sponges available in literature is too small to draw
any conclusion about reflectance of sponges. Some of the spectra are close to living
coral spectra, some to dead coral spectra. So, presence of sponges may decrease the
level of confidence in remote sensing estimation of living and dead coral cover.
3.3.3
Conclusions imaging spectrometry of optically shallow waters
In the optically shallow waters developments have been somewhat different as the biooptical or physical model describing the interaction of light in the water column and on
the substrate is more complex than for optically deep waters and is less easily inverted.
This inversion is required to produce meaningful maps of water variables or substrate
variables. Similar to the developments in the inland waters, but increasingly complex
due to the effect of the substrate, more sophisticated inversion schemes are being proposed. For bathymetry assessment over sandy bottoms and for seagrass mapping several successful examples are discussed or presented. Coral reefs are being studied increasingly as the environmental concern for coral reef health increases. Most work has been
done to characterise the IOP’s and especially the AOP’s by establishing spectral libraries of coral reef reflectance. Very few imaging spectrometry data sets are available
that were analysed for coral reef cover, health or species dis crimination. Most applications were of a qualitative nature.
4
Conclusions
The field of imaging spectrometry is currently at a cross-roads: up till now all results
for imaging spectrometry came from data gathered by airborne systems, as there were
no civilian ima ging spectrometry sensors in space. The successful launch of the EO-1
IMAGING SPECTROMETRY OF WATER
53
platform by TRW/NASA in November 2000, with the imaging spectrometer HYPERION on board marks the dawn of a new era- imaging spectrometry from space. This
book and chapter is therefore marks the state-of-the-art for aquatic systems imaging
spectrometry till the new space sensors results start arriving in the literature. With the
advent of imaging spectrometers that were suitable for water related investigations: the
PMI, the AVIRIS and the CASI in particular at the end of the eighties, this field of
research commenced. First missions and associated research were explorative rather
than operational. From the mid-nineties onwards, operational examples of multitemp oral deployments of airborne imaging spectrometry systems over, mainly, inland water
targets started to happen in The Netherlands, Germany and Scandinavia. These studies
over inland waters were able to deal with the optically deep waters quite well and produced meaningful results for up to 5 optical water quality variables such as CHL, CPC,
TSM, Kd and transparency. The combination of these results into ecological assessment
and monitoring is gathering speed. The bio-optical models for these water are becoming more sophisticated as well as the instruments with which to measure the IOP and
AOP properties. The CASI seems to be the instrument of choice, due to its flexi bility in
platform and its programmable band sets. The number of CASI’s (20 by now) also
determines the availability worldwide of course.
For imaging spectrometry of aquatic ecosystems developments are towards more
complete physics-based models but also towards methods to compute an inversion of
an imaging spectrometry scene using either analytical 1 to 3 band inversions, look up
tables, using matrix inversion schemes or using neural networks. For turbid estuarine
remote sensing, less use has been made of airborne imaging spectrometers due to the
very dynamic nature and the often large size of the estuaries. Most of these studies
were intended as illustrations or experiments in preparation of using satellite sensors to
monitor these systems. As spaceborne imaging spectrometers become available this
field of application is likely to evolve very fast.
In the optically shallow waters developments have been somewhat different as the
bio-optical or physical model describing the interaction of light in the water column
and on the substrate is more complex than for optically deep waters and is less easily
inverted. This inversion is required to produce meaningful maps of water variables or
substrate variables. Similar to the developments in the inland waters, but increasingly
complex due to the effect of the substrate, more sophisticated inversion schemes are
being proposed. For bathymetry assessment over sandy bottoms and for seagrass mapping several successful examples are discussed or presented. Coral reefs are being studied increasingly as the environmental concern for coral reef health increases. Most
work has been done here to characterise the IOP’s and especially the AOP’s by establish
ing spectral libraries of coral reef reflectance. Very few imaging spectrometry
data sets are available that were analysed quantitatively for coral reef cover, health or
species discrimination.
The last 10 years have seen a development towards more understanding of the
physics of how spectral irradiance interacts with the water column and the substrate.
Simultaneously the remote sensing sensors have developed towards systems with
higher sensitivity and more (flexible) spectral bands available. A concurrent increase in
computing power has led to a situation, (together with space imaging spectrometers
being launched successfully) where we anticipate an fast expanding field of research,
development, demonstration, operationalisation and commercialisation of imaging
spectrometry of aquatic ecosystems.
54
A.G. DEKKER ET AL.
As bio-optical and physics based models become more accurate, inversion schemes can become more sophisticated, enabling near real time processing of imaging
spectrometry data. Simulations of the reflectance spectrum from waters, several examples given in this chapter, reduce the requirement for in situ measurements drastically
in the long term. They also enable pre-flight determination of optimal spectral band
configurations for specific tasks. As in situ detection and monitoring becomes more
expensive (due to rising labour costs) and is shown to be less exact, a remote sensing
based approach is beginning to make more and more economical sense. It will be necessary to have available local airborne imaging spectrometry systems or the availability of data from space sensors. The fact that the water column is an ever changing me dium, both in space and time as well as in reflectance signature, indicates that imaging
spectrometry will be the remote sensing instrument of choice for the future for accurate
detection and monitoring of optical water quality and substrate variables.
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
Doerffer, R., Imaging spectroscopy for detection of chlorophyll and suspended
matter, in Imaging spectroscopy: fundamentals and prospective applications,
F. Toselli and J. Bodechtel, Editors. 1992, Kluwer: Dordrecht. p. 215-257.
Dekker, A.G., T.J.M. Malthus, and H.J. Hoogenboom, The remote sensing of
inland water quality, in Advances in Environmental Remote Sensing, F.M.
Danson and S.E. Plummer, Editors. 1995, John Wiley & Sons, UK.:
Chichester. p. 123-142.
Kondratyev, K.Y., D.V. Pozdnyakov, and L.H. Pettersson, Water quality
remote sensing in the visible spectrum. International Journal of Remote
Sensing, 1998. 19(5): p. 957-979.
Lindell, L.T., et al., Manual for monitoring European lakes using remote
sensing techniques. 1999, Luxembourg: European Communities. 164.
Durand, D., et al., Characterisation of inland and coastal waters with space
sensors. 1999, European Commission-Centre for Earth Observation (CEO):
Solheimsviken, Norway. p. 1-179.
IOCCG, Remote sensing of ocean colour in coastal, and other opticallycomplex waters. 2000, International Ocean-Colour Coordinating Group:
Dartmouth. p. 140.
Bricaud, A., A. Morel, and V. Barale, MERIS potential for ocean colour
studies in the open ocean. International Journal of Remote Sensing, 1999. 20:
p. 1757-1769.
Mobley, C.D., Light and water; Radiative transfer in natural waters. 1994,
London: Academic Press. 592.
Mobley, C.D. and L.K. Sundman, Hydrolight 4.1 technical documentation.
2000, Sequoia Scientific, Inc: WA. p. 76.
Aas, E., Two-stream irradiance model for deep waters. Appl.Opt., 1987. 26:
p. 2095-2101.
Mobley, C.D., The optical properties of water, in Handbook of Optics, I. Bass,
Editor. 1995, McGraw-Hill and Optical Society of America: New York. p. 56.
Walker, R.E., Marine light field statistics. Wiley series in pure and applied
optics, ed. J.W. Goodman. 1994, New York: Wiley. 675.
Dekker, A.G., et al. Angular scattering functions of algae and silt: an analysis
of backscattering to scattering fraction. in Ocean Optics XIII. 1997.
Bellingford, USA: SPIE.
IMAGING SPECTROMETRY OF WATER
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
55
Gordon, H.R., O.B. Brown, and M.M. Jacobs, Computed relationships
between the inherent and apparent optical properties of a flat homogeneous
ocean. Appl.Opt., 1975. 14(2): p. 417-427.
Krijgsman, J., Optical remote sensing of water quality parameters;
interpretation of reflectance spectra. 1994, Delft University of Technology:
Delft, The Netherlands. p. 1-198.
Gordon, H.R., Can the Lambert-Beer law be applied to the diffuse attenuation
coefficient of ocean water? Limnol.Oceanogr., 1989. 34(8): p. 1389-1409.
Stavn, R.H. and A.D. Weidemann, Shape factors, two-flow models, and the
problem of irradiance inversion in estimating optical parameters.
Limnol.Oceanogr., 1989. 34(8): p. 1426-1441.
Wilson, W.H. and D.A. Kiefer, Reflectance spectroscopy of marine
phytoplankton. Part 2 A simple model of ocean color. Limnol.Oceanogr.,
1979. 24(4): p. 673-682.
Kirk, J.T.O., Volume scattering function, average cosines, and the underwater
light field. Limnol.Oceanogr., 1991. 36(3): p. 455-467.
Duntley, S.Q., Light in the sea. J.Opt.Soc.America, 1963. 53 : p. 214-233.
Kirk, J.T.O., Dependance of relationship between inherent and apparent
optical properties of water on solar altitude. Limnol.Oceanogr., 1984. 29(2):
p. 350-356.
Kirk, J.T.O., Characteristics of the light field in highly turbid waters: a Monte
Carlo study. Limnol.Oceanogr., 1994. 39(3): p. 702-706.
Morel, A. and L. Prieur, Analysis of variations in ocean colour.
Limnol.Oceanogr., 1977. 22(4): p. 709-722.
Kirk, J.T.O., Light & photosynthesis in aquatic ecosystems. 2 ed. 1994,
Cambridge, UK: University Press. 1-509.
Gordon, H.R., et al., A semianalytical model of ocean colour. J.Geophys.Res.,
1988. 93(D9): p. 10909.
Morel, A. and B. Gentili, Diffuse reflectance of oceanic waters. II.
Bidirectional aspects. Appl.Opt., 1993. 32(33): p. 6864-6879.
Dekker, A.G., Detection of optical water quality parameters for eutrophic
waters by high resolution remote sensing, in Faculty of Earth Sciences. 1993,
Vrije Universiteit: Amsterdam. p. 1-240.
Whitlock, C.H., et al., Comparison of reflectance with backscatter and
absorption parameters for turbid waters. Appl.Opt., 1981. 20(3): p. 517-522.
Maritorena, S., A. Morel, and B. Gentili, Diffuse reflectance of oceanic
shallow waters: influence of water depth and bottom albedo.
Limnol.Oceanogr., 1994. 39(7): p. 1689-1703.
Kirk, J.T.O., The upwelling light stream in natural waters. Limnol.Oceanogr.,
1989. 34(8): p. 1410-1425.
Philpot, W.D. and A. Vodacek, Laser-induced fluorescence: limits to the
remote detection of hydrogen ion, aluminium, and dissolved organic matter.
Remote Sens.Environ., 1989. 29: p. 51-65.
Plass, G.N. and G.W. Kattawar, Monte Carlo calculations of radiative
transfer in the Earth's atmosphere-ocean system. I. Flux in the atmosphere
and ocean. J. Phys. Oceanogr, 1972. 2 : p. 139-145.
Estep, L. and R.A. Arnone, Effect of whitecaps on determination of
chlorophyll concentration from satellite data. Remote Sensing of
Environment, 1994. 50: p. 328-334.
56
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
A.G. DEKKER ET AL.
Gordon, H.R. and M.H. Wang, Retrieval of water-leaving radiance and
aerosol optical-thickness over the oceans with Seawifs - a preliminary
algorithm. Applied Optics, 1994. 33: p. 443-452.
Kirk, J.T.O., Light and photosynthesis in aquatic ecosystems. 1983:
Cambridge University Press.
Hilton, J., Airborne remote sensing for freshwater and estuarine monitoring.
Water Res., 1984. 18(10): p. 1195-1223.
Bukata, R.P., et al., Optical properties and remote sensing of inland and
coastal waters. 1995, Boca Raton, USA: CRC Press. 1-362.
Gower, J.F.R., R. Doerffer, and G.A. Borstad, Interpretation of the 685 nm
peak in water-leaving radiance spectra in terms of fluorescence, absorption
and scattering, and its observation by MERIS. International Journal of Remote
Sensing, 1999. 20: p. 1771-1786.
Jupp, D.L.B., J.T.O. Kirk, and G.P. Harris, Detection, identification and
mapping of cyanobacteria-using remote sensing to measure the optical quality
of turbid inland waters. Australian Journal of Marine and Freshwater
Research, 1994. 45: p. 801-828.
Hoogenboom, H.J., A.G. Dekker, and J.F. De Haan, Retrieval of chlorophyll
and suspended matter in inland waters from CASI data by matrix inversion.
Canadian Journal of Remote Sensing, 1998. 24(2): p. 144-152.
Dekker, A.G., et al., Remote sensing for inland water quality detection and
monitoring: State-of-the-art application in Friesland waters, in GIS and
Remote Sensing techniques in Land- and Water-management, A. van Dijk and
M.G. Bos, Editors. 2001, Kluwer Academic Publishers: Netherlands. p. 17-38.
Olbert,
C.,
Bestimmung
raumlicher
Verteilungsmuster
von
Wasserinhaltstoffen in ausgewahlten Berliner und Brandenburger Gewassern
mit Methoden der Fernerkundung, in Institut fur WeltraumwissenchaftenStrahlung in Atmosphare und Ozean. 2000, Freie Universitat Berlin: Berlin. p.
172.
Fischer, J., Fernerkundung von Schwebstoffen im Ozean. 1983, University of
Hamburg: Hamburg, Germany.
Fell, F. and J. Fischer, Numerical simulation of the light field in the
atmosphere-ocean system using the matrix-operator method. J. Quant. Spec.
Radiat. Transfer, in press.
Fell, F., Validierung eines Modells zur Simulation des Strahlungstransportes
in Atmosphäre und Ozean. 1997, Freie Universitat: Berlin.
Babey, S.K. and R.J. Soffer, Radiometric calibration of the compact airborne
spectrographic imager (CASI). Can.J.Remote Sensing, 1992. 18(4): p. 233242.
Olbert, C., Atmospheric correction for CASI data using an atmospheric
radiative transfer model. Can.J.Remote Sensing, 1998. 24(2): p. 114-127.
Bishop, C.M., Neural Networks for Pattern Recognition. 1995, New York:
Oxford University Press.
Cracknell, A.P., Remote sensing techniques in estuaries and coastal zones - an
update. International Journal of Remote Sensing, 1999. 20: p. 485-496.
Ferrier, G. and J.M. Anderson, The application of remotely sensed data in the
study of tidal systems in the Tay Estuary, Scotland, U. K. International Journal
of Remote Sensing, 1997. 18(9): p. 2035-2065.
IMAGING SPECTROMETRY OF WATER
51.
52.
53.
54.
55.
56.
57.
58.
59.
60.
61.
62.
63.
64.
65.
66.
57
Siegel, H. and M. Gerth, Satellite-Based Studies of the 1997 Oder Flood Event
in the Southern Baltic Sea. Remote Sensing of Environment, 2000. 73(2): p.
207-217.
Ferrari, G.M., N. Hoepffner, and M. Mingazzini, Optical properties of the
water in a deltaic environment: prospective tool to analyze satellite data in
turbid waters. Remote Sensing of Environment, 1996. 58(1): p. 69-80.
Wernand, M.R., S.J. Shimwell, and J.C.D. Munck, A simple method of full
spectrum reconstruction by a five-band approach for ocean colour
applications. International Journal of Remote Sensing, 1997. 18(9): p. 19771986.
Kratzer, S., D. Bowers, and P.B. Tett, Seasonal changes in colour ratios and
optically active constituents in the optical Case-2 waters of the Menai Strait,
North Wales. International Journal of Remote Sensing, 2000. 21(11): p. 22252246.
Frette, O., J.J. Stamnes, and K. Stamnes, Optical remote sensing of marine
constituents in coastal waters: a feasible study. Appl.Opt., 1998. 37(36): p.
8318-8326.
Jain, S.C. and J.R. Miller, Subsurface parameters: optimization approach to
their determination from remotely sensed water color data. Appl.Opt., 1976.
15(4): p. 886-890.
Doerffer, R. and J. Fischer, Concentrations of chlorophyll, suspended matter
and gelbstoff in case II waters derived from satellite coastal zone color
scanner with inverse modeling methods. Journal of Geophysical Research,
1994. 99(c4): p. 7457-7466.
Doerffer, R. and H. Schiller. Inverse modelling for retrieval of ocean color
parameters in Case II coastal waters: an analysis of the minimum error. in
Ocean Optics XII. 1994. Bellingham, Washington, USA: SPIE.
Tassan, S., Local algorithms using SeaWIFS data for the retrieval of
phytoplankton, pigments, suspended sediment, and yellow substance in coastal
waters. Appl.Opt., 1994: p. 2369-2378.
Schalles, J.F., et al., Estimation of Chlorophyll a from time series
measurements of high spectral resolution reflectance in an eutrophic lake.
Journal of Phycology, 1998. 34: p. 383-390.
Forget, P., et al., Inversion of reflectance spectra of nonchlorophyllous turbid
coastal waters. Remote Sensing of Environment, 1999. 68(3): p. 264-272.
Lahet, F., S. Ouillon, and P. Forget, A three-component model of ocean color
and its application in the Ebro River mouth area. Remote Sensing of
Environment, 2000. 72(2): p. 181-190.
Vasilkov, A.P., V.I. Burenkov, and K.G. Ruddick, The spectral reflectance
and transparency of river plume waters. International Journal of Remote
Sensing, 1999. 20: p. 2497-2508.
Woodruff, D.L., et al., Remote Estimation of Water Clarity in Optically
Complex Estuarine Waters. Remote Sensing of Environment, 1999. 68(1): p.
41-52.
Hoogenboom, H.J. and J.R. Roberti. An integral approach to retrieve total
suspended matter concentrations in the Marsdiep (Netherlands). in Ocean
Optics XV. 2000. Monaco.
Dekker, A.G., et al. Analytical processing of multitemporal SPOT and
Landsat images for estuarine management in Kalimantan Indonesia. in 18th
58
67.
68.
69.
70.
71.
72.
73.
74.
75.
76.
77.
78.
79.
80.
A.G. DEKKER ET AL.
EARSeL Symposium on Operational Remote Sensing for Sustainable
Development. 1999. Enschede, The Netherlands: Balkema, Rotterdam, The
Netherlands.
Carder, K.L., et al., Aviris calibration and application in coastal oceanic
environments. Remote Sens.Environ., 1993. 44: p. 205-216.
Lee, Z., et al., Model for the interpretation of hyperspectral remote-sensing
reflectance. Applied Optics, 1994. 33(24): p. 5721-5732.
Lee, Z., K.L. Carder, and R.F. Chen. Environmental properties observed from
AVIRIS data. in The 9th JPL Airborne Earth Science Workshop. 2000.
California: NASA/JPL.
Bagheri, S., et al. Comparison of field spectroradiometers in preparation for
bio-optical modelling. in 9th JPL Airborne Earth Science Workshop. 2000.
California.
Eloheimo, K., et al. Coastal monitorng using satellite, airborne and in situ
measurement data in the Baltic Sea. in 5th International Conference on
Remote Sensing for Marine and Coastal Environments. 1998. San Diego,
California: ERIM-Veridian.
Jørgensen, P.V. and K. Edelvang, CASI data utilized for mapping suspended
matter concentrations in sediment plumes and verification of 2-D
hydrodynamic modelling. International Journal of Remote Sensing, 2000.
21(11): p. 2247-2258.
Buckton, D., E. Omongain, and S. Danaher, The use of Neural Networks for
the estimation of oceanic constituents based on the MERIS instrument.
International Journal of Remote Sensing, 1999. 20: p. 1841-1851.
Moore, G.F., J. Aiken, and S.J. Lavender, The atmospheric correction of
water colour and the quantitative retrieval of suspended particulate matter in
Case II waters: application to MERIS. International Journal of Remote
Sensing, 1999. 20: p. 1713-1733.
Schiller, H. and R. Doerffer, Neural network for emulation of an inverse
model - operational derivation of Case II water properties from MERIS data.
International Journal of Remote Sensing, 1999. 20: p. 1735-1746.
Pasterkamp, R., S.W.M. Peters, and A.G. Dekker. Environmental baseline
mapping of total suspended matter concentrations in the Western Scheldt
Estuary supported by the use of SPOT images. in Sixth International
Conference on Remote Sensing for Marine and Coastal Environments. 2000.
Charleston, South Carolina, US: Veridian ERIM International.
Maldegem, D.C.v., De slibbalans van het Schelde-estuarium. 1992,
Rijkswaterstaat, DGW: Middelburg, The Netherlands.
Gould, R.W. and R.A. Arnone, Remote sensing estimates of inherent optical
properties in a coastal environment. Remote Sens.Environ., 1997. 61: p. 209301.
Pozdnyakov, D.V. and K.Y. Kondratev, Remote sensing of natural waters in
the visible band of the spectrum. 3. Some special questions. Earth Observation
& Remote Sensing, 1999. 15: p. 365-385.
Kohler, D.D.R. and W.D. Philpot. Comparing in situ and remotely sensed
measurements in optically shallow waters. in Ocean Optics XV. 2000.
Monaco.
IMAGING SPECTROMETRY OF WATER
81.
82.
83.
84.
85.
86.
87.
88.
89.
90.
91.
92.
93.
94.
95.
96.
97.
59
Durand, D., J. Bijaoui, and F. Cauneau, Optical remote sensing of shallowwater environmental parameters. Remote Sensing of Environment, 2000.
73(2): p. 152-161.
Sandidge, J.C. and R.J. Holyer, Coastal bathymetry from hyperspectral
observations of water radiance. Remote Sensing of Environment, 1998. 65(3):
p. 341-352.
Hamilton, M.K., et al., Estimating chlorophyll content and bathymetry of Lake
Tahoe using AVIRIS data. Remote Sensing of Environment, 1993. 44: p. 217230.
Tolk, B.L., L. Han, and D.C. Rundquist, The impact of bottom brightness on
spectral reflectance of suspended sediments. International Journal of Remote
Sensing, 2000. 21(11): p. 2259-2268.
Lee, Z., et al., Hyperspectral remote sensing for shallow waters: 2. deriving
bottom depths and water properties by optimization. Applied Optics, 1999.
38(18): p. 3831-3843.
Wilcox, D.A., Wetland and aquatic macrophytes as indicators of
anthropogenic hydrologic disturbance. Natural Areas Journal, 1995. 15(3): p.
240-248.
Short, F.T. and S. Wyllie -Echeverria, Natural and human-induced disturbance
of seagrasses. Environmental Conservation, 1996. 23: p. 17-27.
Meinesz, A., Killer algae: the true tale of a biological invasion. 1999,
Chicago: University of Chicago Press. 360.
Pasqualini, V., et al., Contribution of side scan sonar to the management of
Mediterranean littoral ecosystems. International Journal of Remote Sensing,
2000. 21(2): p. 367-378.
Hick, P., Determination of water column chaeracteristics in coastal
environments using remote sensing, in School of Surveying and Land
Information. 1997, Curtin University of Technology: Perth. p. 227.
Jernakoff, P. and P. Hick. Spectral measurement of marine habitat:
simultaneous field measurements and CASI data. in 7th Australasian Remote
Sensing Conference. 1994. Melbourne, Australia.
Zhang, X., On the estimation of biomass of submerged vegetation using
satellite thematic mapper (TM) imagery: a case study of the Honghu , PR
China. International Journal of Remote Sensing, 1998. 19(1): p. 11-20.
Lennon, P. and P. Luck, Seagrass mapping using Landsat TM data: a case
study in southern Queensland. Asian-Pacific Remote Sensing Journal, 1990.
2(2): p. 6.
Mumby, P.J., et al., Measurement of seagrass standing crop using satellite
and digital airborne remote sensing. Marine Ecology-Progress Series, 1997.
159: p. 51-60.
Clark, C.D., et al., Mapping and measurement of tropical coastal
environments with hyperspectral and high spatial resolution data.
International Journal of Remote Sensing, 1997. 18(2): p. 237-242.
Penuelas, J., et al., Assessing Community Type, Plant Biomass, Pigment
Composition, and Photosynthetic Efficiency of Aquatic Vegetation From
Spectral Reflectance. Remote Sensing of Environment, 1993. 46(2): p. 110118.
Malthus, T.J. and D.G. George, Airborne remote sensing of macrophytes in
Cefni Reservoir, Anglesey, UK. Aquatic Botany, 1997. 58(3-4): p. 317-332.
60
98.
99.
100.
101.
102.
103.
104.
105.
106.
107.
108.
109.
110.
111.
A.G. DEKKER ET AL.
Alberotanza, L., et al., Eutrophication and environmental recovery of the
Orbetello Lagoon, in Structure and processes in the Mediterranean
ecosystems, F.M. Faranda, L. Guglielmo, and G. Spezie, Editors. 1999,
Springer.
Wittlinger, S. and R. Zimmermann. Hyperspectral remote sensing of subtidal
macroalgal assemblages in optically shallow waters. in Ocean Optics XV.
2000. Monaco.
Mumby, P.J., et al., Benefits of water column correction and contextual
editing for mapping coral reefs. International Journal of Remote Sensing,
1998. 19(1): p. 203-210.
Lyzenga, D.R., Remote sensing of bottom reflectance and water attenuation
parameters in shallow water using aircraft and Landsat data. International
Journal of Remote Sensing, 1981. 2 : p. 71-82.
Moussa, H.B., M. Viollier, and T. Belsher, Remote-sensing of macrophytic
algae in the Molene Archipelago: ground-radiometry and application to
SPOT satellite-data. International Journal of Remote Sensing, 1989. 10: p. 5369.
Bierwirth, P.N., T.J. Lee, and R.V. Burne, Shallow sea-floor reflectance and
water depth derived by unmixing multispectral imagery. ISPRS Journal of
Photogrammetry and Remote Sensing, 1993. 59(3): p. 331-338.
Jupp, D.L.B., K.K. Mayo, and e. al., Remote sensing for planning and
managing the Great Barrier Reef of Australia. Photogrammetria, 1985. 40: p.
21-42.
Maritorena, S., Remote sensing of water attenuation in coral reefs: case study
in French Polynesia. International Journal of Remote Sensing, 1996. 17(1): p.
155-166.
Maritorena, S. and N. Guillocheau, Optical properties of water and spectral
light absorption by living and non-living particles and by yellow substances in
coral reef waters of French Polynesia. Marine Ecol.Progr.Ser., 1996. 131: p.
245-255.
Tassan, S., modified Lyzenga's method for macroalgae detection in water with
non- uniform composition. International Journal of Remote Sensing, 1996.
17(8): p. 1601-1607.
Pratt, P., K.L. Carder, and D.K. Costello. Remote sensing reflectance
algorithms developed to correct underwater coral imagery for the effect of
optical thickness to assist in benthic classification. in 4th Remote Sensing
Conference for Marine and Coastal Environments. 1997. Orlando, Florida:
ERIM.
Malthus, T.J., et al. Can biophysical properties of submersed macrophytes be
determined by remote sensing? in Fourth International Conference on Remote
Sensing for Marine and Coastal Environments: Technology and Applications.
1997. Orlando, Florida.: ERIM-Veridian, Michigan.
Mumby, P.J., et al., Digital analysis of multispectral airborne imagery of
coral reefs. Coral Reefs, 1998. 17(1): p. 59-69.
Kenneth, J., H. Zhang, and A. Chapin. Bi-directional reflectance distribution
functions (BRDF) of benthic surfaces in the littoral zone. in Ocean Optics XV.
2000. Monaco, France: SPIE.
IMAGING SPECTROMETRY OF WATER
112.
113.
114.
115.
116.
117.
118.
119.
120.
121.
122.
123.
124.
125.
126.
61
Karpouzli, E., et al., Underwater light characterisation for correction of
remotely sensed images. International Journal of Remote Sensing, 2001. In
press.
Meulstee, C., P.H. Nienhuis, and H.T.C. van Stokkom, Biomass assessment of
estuarine macrophytobenthos using aerial photography. Marine Biology,
1986. 91: p. 331-5.
Malthus, T.J.M., E.P.H. Best, and A.G. De kker, An assessment of the
importance of emergent and floating-leaved macrophytes to trophic status in
the Loosdrecht lakes (The Netherlands). Hydrobiologia, 1987. 191: p. 257263.
Ferguson, R.L., L.L. Wood, and D.B. Graham, Monitoring spatial change in
seagrass habitat with aerial- photography. Photogrammetric Engineering and
Remote Sensing, 1993. 59(6): p. 1033-1038.
Marshall, T.R. and P.F. Lee, Mapping aquatic macrophytes through digital
image-analysis of aerial photographs - an assessment. Journal of Aquatic
Plant Management, 1994. 32: p. 61-66.
Green, E.P., et al., A review of remote-sensing for the assessment and
management of tropical coastal resources. Coastal Management, 1996. 24: p.
1-40.
Rutchey, K. and L. Vilchek, Air photointerpretation and satellite imagery
analysis techniques for mapping cattail coverage in a northern Everglades
impoundment. Photogrammetric Engineering and Remote Sensing, 1999. 65:
p. 185-191.
Jensen, J.R., et al., Remote sensing inland wetlands: a multispectral approach.
Photogrammetric Engineering and Remote Sensing, 1986. 52: p. 87-100.
Holden, H. and E. LeDrew. Spectral discrimination of bleached and healthy
submerged corals based on principal component analysis. in 4th International
Conference on Remote Sensing for Marine Coastal Environments. 1997.
Orlando, Florida: Waterloo laboratory for Earth Observations, University of
Waterloo.
Holden, H. and E. LeDrew, Spectral Discrimination of Healthy and NonHealthy Corals Based on Cluster Analysis, Principal Components Analysis,
and Derivative Spectroscopy. Remote Sens.Environ., 1998. 65 : p. 217-224.
Dunk, I. and M.R. Lewis. Seagrass and shallow water feature discrimination
using Hymap Imagery. in 10th Australian Remote Sensing and
Photogrammetry Conference. 1999. Australia: University of Adelaide.
Pasqualini, V., C. Pergent-Martini, and C. Fernandez, The use of airborne
remote sensing for benthic cartography: advantages and reliability.
International Journal of Remote Sensing, 1997. 18(5): p. 1167-1177.
Bajjouk, T., J. Populus, and B. Guillaumont, Quantification of subpixel cover
fractions using principal component analysis and a linear programming
method: application to the coastal zone of Roscoff (France). Remote Sensing
of Environment, 1998. 64(2): p. 153-165.
Bajjouk, T., B. Guillaumont, and J. Populus, Application of airborne imaging
spectrometry system data to intertidal seaweed classification and mapping.
Hydrobiologia, 1996. 327: p. 463-471.
Holden, H. and E. LeDrew. Hyperspectral identification of coral reef features
in Fiji and Indonesia. in 5th international conference on remote sensing for
marine and coastal environments. 1998. San Diego, California.
62
127.
128.
129.
130.
131.
132.
133.
134.
135.
136.
137.
138.
139.
140.
141.
142.
A.G. DEKKER ET AL.
Alberotanza, L., et al., Hyperspectral aerial images. A valuable tool for
submerged vegetation recognition in the Orbetello Lagoons, Italy.
International Journal of Remote Sensing, 1999. 20(3): p. 523-533.
Louchard, E.M., R.P. Reid, and C.F. Stephens. Classification of sediment
types and estimation of water depth using spectral libraries. in Ocean Optics
XV. 2000. Monaco.
Anstee, J.M., et al. Use of hyperspectral imaging for benthic species mapping
in South Australian coastal waters. in 10th Australian Remote Sensing and
Photogrammetry Conference. 2000. Adelaide, Australia: The Remote Sensing
& Photogrammetry Association of Australia.
Jakubauskas, M., et al., Close-range remote sensing of aquatic macrophyte
vegetation cover. International Journal of Remote Sensing, 2000. 21(18): p.
3533-3538.
Armstrong, R.A., Remote sensing of submerged vegetation canopies for
biomass estimation. International Journal of Remote Sensing, 1993. 14(3): p.
621-627.
Jupp, D.L.B., et al., Port Philip Bay Benthic Habitat mapping project Task
G2.2. 1996, CSIRO Land&Water: Melbourne. p. 40.
Phillips, R.C. and Menez, Seagrasses, in Smiths. Contr. Mar. Sci. 1988:
Washington, D.C.
Kuo, J. and A.J. Mc Comb, A treatise on the biology of seagrass with special
reference to the Australian region, in Biology of Seagrass., A.W.D. Larkum,
McComb, A.J.Shepherd,S.A., Editor. 1989: Elsevier, Amsterdam. p. 6-73.
Borstadt, G.A., et al. Towards a management plan for a tropical reef-lagoon
system using airborne multispectral imaging and GIS. in Fourth International
Conference on Remote Sensing for Marine and Coastal Environments. 1997.
Orlando, Florida: ERIM.
Holasek, R., et al. Coral and substrate mapping in Kaneohe Bay, Oahu,
Hawaii using the advanced airborne hyperspectral imaging system (AAHIS).
in Fifth International Conference on Remote Sensing for Marine and Coastal
Environment. 1998. San Diego, California.
Mumby, P.J. and A.R. Harborne, Development of a systematic classification
scheme of marine habitats to facilitate regional management and mapping of
Caribbean coral reefs. Biological Conservation, 1999. 88: p. 155-163.
Holden, H. and E. Ledrew, Hyperspectral identification of coral reef features.
International Journal of Remote Sensing, 1999. 20: p. 2545-2563.
Myers, M.R., et al., Optical spectra and pigmentation of Caribbean reef
corals and macroalgae. Coral Reefs, 1999. 18: p. 179-186.
Clark, C.D., et al., Spectral discrimination of coral mortality states following
a severe bleaching event. International Journal of Remote Sensing, 2000.
21(11): p. 2321-2327.
Hochberg, E.J. and M.J. Atkinson, Spectral discrimination of coral reef
benthic communities. Coral Reefs, 2000. 19: p. 164-171.
Schalles, J.F., et al. Close range, hyperspectral reflectance measurements of
healthy Indo-Pacific and Carribean Corals. in Sixth International Conference
on Remote Sensing for Marine and Coastal Environments. 2000. Charleston:
ERIM -Veridian.
IMAGING SPECTROMETRY OF WATER
143.
144.
145.
146.
147.
148.
63
Stephens, A., et al. Effects of micro-algal communities on radiance reflectance
of carbonate sediments in optically shallow marine environments. in Ocean
Optics XV. 2000. Monaco: SPIE.
Zimmermann, R. and S. Wittlinger. Hyperspectral remote sensing of
submerged aquatic vegetation in optically shallow waters. in Ocean Optics
XV. 2000. Monaco.
Holden, H. and E. LeDrew. Hyperspectral versus multispectral imaging for
submerged coral detection. in 6th International Conference on Remote
Sensing for Marine and Coastal Environments. 2000. Charleston, NC.
Kutser, T., et al. Hyperspectral detection of coral reef bottom types. in Ocean
Optics XV. 2000. Monaco.
Kutser, T., et al., A hyperspectral model for interpretation of passive optical
remote sensing data from turbid lakes. The Science of the Total Environment,
2001. 268(1-3): p. 47-58.
Kutser, T., et al. Hyperspectral detection of coral reef health. in 10th
Australasian Remote Sensing Conference. 2000. Adelaide, Australia.