5.1.2 Water Quality
Mapping change in water quality relies on the development of a predictive model
relating the water quality variable of interest to the radiance received by the sensor.
Early work in this area was completed in the Loosdrecht Lakes in The Netherlands by
Dekker and Seyhan (1988), who used qualitative and quantitative assessment of
satellite (Landsat TM, SPOT) and airborne (CAESAR-MSS and low-altitude aerial
colour photography) remote sensing data to detect and study the temporal and spatial
variations in water quality. Further work by Lathrop et al. (1991) investigated multidate water-quality calibration algorithms for turbid inland water conditions using
Landsat TM in Green Bay, Lake Michigan to estimate absolute values and change in
total suspended solids and Secchi depth. Similar work in Egyptian lagoons using
Landsat TM and locally calibrated regression models by Dewidar and Khedr (2001)
have been similarly successful. Multitemporal classification approaches have also been
found successful in detecting water quality change. Work by Pal and Mohanty (2002),
using IRS-1B data from the Chilka Lagoon, East Coast of India, was used to predict
selected water quality parameters and lagoon modification over an inter-annual cycle.
However, the importance of accurate image calibration was demonstrated in work by
Islam et al. (2003) in Moreton Bay, Brisbane, Australia. Their estimates of total
suspended sediment and Secchi depth, based on empirical models derived from a
Landsat TM reference image, were found to differ by 35-152% when applied to
different images. They concluded that image calibration to like-values could be used to
reliably map certain water quality parameters from multitemporal TM images, as long
as the water type under study remains unchanged. To avoid the problem of multiple
calibrations, more recent semi-analytical and analytic models that account for bottom
depth have been developed (Brando and Dekker, 2003). Some, like the model of Lee
et al. (2001) have been used to derive accurate estimates of chlorophyll, dissolved
organic matter, and suspended sediment concentrations, but they rely on the availability
of calibrated hyperspectral imagery. To date, this has been difficult to obtain for multitemporal studies at regional scales. More recent work (Dekker et al., 2001b; Phinn,
2003) has addressed imitations of empirical approaches (Islam et al., 2003), and used
atmospheric and air-water interface corrected multi-date Landsat ETM data with field
measured optical properties to estimate organic and suspended matter concentrations.
5.1.3 Substrate Composition
The measurement of substrate composition and benthic cover is complicated by
the spatial variations of water depth and water quality. These variations prevent the
normalization of image data required for accurate mapping, estimation of biophysical
properties, and change detection (Phinn et al., 2000c). In this sense, the classification of
substrate composition, water depth, and the measurement of water quality parameters
are explicably linked (Lee et al., 2001). In a desktop study using a radiative transfer
code to simulate the effect of water column effects, Holden and LeDrew (2002), noted
that the classification accuracy of benthic habitat type increased significantly when the
effects of the water column were removed. Such effects can also be seen using the
interactive WASI program provided on the CD accompanying this book in association
with Chapter 4 (Gege and Albert). Image based studies confirmed the importance of
including depth effects (Mumby et al., 1998). The availability of radiative transfer
equations has allowed the development of classification approaches based on simulated
spectra derived at different depths (Louchard et al., 2003). Use of a radiative transfer
approach allows the retrieval of the seafloor reflectance, which can then be used to
classify the benthos or derive biophysical indicators of ecosystem health, such as the
234
Phinn, Joyce, Scarth and Roelfsema
Mapping change in water quality relies on the development of a predictive model
relating the water quality variable of interest to the radiance received by the sensor.
Early work in this area was completed in the Loosdrecht Lakes in The Netherlands by
Dekker and Seyhan (1988), who used qualitative and quantitative assessment of
satellite (Landsat TM, SPOT) and airborne (CAESAR-MSS and low-altitude aerial
colour photography) remote sensing data to detect and study the temporal and spatial
variations in water quality. Further work by Lathrop et al. (1991) investigated multidate water-quality calibration algorithms for turbid inland water conditions using
Landsat TM in Green Bay, Lake Michigan to estimate absolute values and change in
total suspended solids and Secchi depth. Similar work in Egyptian lagoons using
Landsat TM and locally calibrated regression models by Dewidar and Khedr (2001)
have been similarly successful. Multitemporal classification approaches have also been
found successful in detecting water quality change. Work by Pal and Mohanty (2002),
using IRS-1B data from the Chilka Lagoon, East Coast of India, was used to predict
selected water quality parameters and lagoon modification over an inter-annual cycle.
However, the importance of accurate image calibration was demonstrated in work by
Islam et al. (2003) in Moreton Bay, Brisbane, Australia. Their estimates of total
suspended sediment and Secchi depth, based on empirical models derived from a
Landsat TM reference image, were found to differ by 35-152% when applied to
different images. They concluded that image calibration to like-values could be used to
reliably map certain water quality parameters from multitemporal TM images, as long
as the water type under study remains unchanged. To avoid the problem of multiple
calibrations, more recent semi-analytical and analytic models that account for bottom
depth have been developed (Brando and Dekker, 2003). Some, like the model of Lee
et al. (2001) have been used to derive accurate estimates of chlorophyll, dissolved
organic matter, and suspended sediment concentrations, but they rely on the availability
of calibrated hyperspectral imagery. To date, this has been difficult to obtain for multitemporal studies at regional scales. More recent work (Dekker et al., 2001b; Phinn,
2003) has addressed imitations of empirical approaches (Islam et al., 2003), and used
atmospheric and air-water interface corrected multi-date Landsat ETM data with field
measured optical properties to estimate organic and suspended matter concentrations.
5.1.3 Substrate Composition
The measurement of substrate composition and benthic cover is complicated by
the spatial variations of water depth and water quality. These variations prevent the
normalization of image data required for accurate mapping, estimation of biophysical
properties, and change detection (Phinn et al., 2000c). In this sense, the classification of
substrate composition, water depth, and the measurement of water quality parameters
are explicably linked (Lee et al., 2001). In a desktop study using a radiative transfer
code to simulate the effect of water column effects, Holden and LeDrew (2002), noted
that the classification accuracy of benthic habitat type increased significantly when the
effects of the water column were removed. Such effects can also be seen using the
interactive WASI program provided on the CD accompanying this book in association
with Chapter 4 (Gege and Albert). Image based studies confirmed the importance of
including depth effects (Mumby et al., 1998). The availability of radiative transfer
equations has allowed the development of classification approaches based on simulated
spectra derived at different depths (Louchard et al., 2003). Use of a radiative transfer
approach allows the retrieval of the seafloor reflectance, which can then be used to
classify the benthos or derive biophysical indicators of ecosystem health, such as the
234
Phinn, Joyce, Scarth and Roelfsema