3.2.2 Image Processing
The general procedures for image preprocessing (i.e., geometric correction,
radiometric correction, atmospheric correction, and sunglint removal) were
introduced in Chap. 1 and are summarized in Green et al. (2000). This section is
focused on selected examples for the processing steps that follow the initial preprocessing, specifically atmospheric correction, water column correction, image
classification and contextual editing.
Atmospheric correction. The total signal received by satellite sensors is dominated by radiance contributed through atmospheric scattering. Thus, atmospheric
correction is essential to retrieve signals from the sea. The simplest such procedure
is dark pixel subtraction. Based on an assumption that somewhere in the image is a
pixel with zero reflectance, which means the radiance recorded by the sensor is
solely from the atmospheric scattering, the minimum pixel value is subtracted
from all other pixels to remove the radiance derived from atmospheric scattering.
Because images with coral reefs normally contain ocean areas, pixels from deep
ocean areas are often used for this correction. Modeling radiative transfer in the
atmosphere is another more sophisticated option for correction. Mishra et al.
(2005) described a procedure for first-order atmospheric correction for IKONOS
imagery based on radiative transfer theory. Radiative transfer codes such as 6S
(Vermote et al. 1997) and MODTRAN (http://www.modtran.org) may also be
used, and some image processing software has optional modules for atmospheric
correction (e.g., FLAASH for ENVI and ATCOR for ERDAS Imagine).
Water column correction. A fundamental challenge for remote sensing of coral
reefs is the existence of the water column above the bottom features of interest
(i.e., the benthic habitat). Light intensity decreases exponentially with increasing
depth. This is known as attenuation, and the degree of attenuation is a function of
wavelength, water depth and water properties. Water strongly absorbs wavelengths
longer than 700 nm (near infrared), thus visible regions (blue, green and red) are
used for mapping bottom features and infrared regions are used for extracting
shorelines and emergent vegetation.
Rigorous removal of the water influence on bottom reflectance requires
knowledge or estimations of water depth and attenuation characteristics at every
pixel. Calculations using these parameters can be effectively used to estimate reeftop reflectance (‘‘reef-up’’ approach) for substrate mapping (Purkis 2005). In cases
where ground truth data (e.g., light attenuation coefficients, water depth, and
bottom albedo) are available, wavelength-dependency of light penetration offers
high ability to estimate bathymetry of reefs (e.g., Philpot 1989; Stumpf et al.
2003). Recently, Mishra et al. (2005) proposed a method to estimate water depths
and attenuation characteristics using IKONOS data, with minimum requirements
for ground truth data. They identified a ratio of wavebands (blue and green) that is
constant for all bottom types, and developed a polynomial equation to estimate
water depths from the ratio. Attenuation characteristics were then estimated by
3 Multispectral Applications
59
The general procedures for image preprocessing (i.e., geometric correction,
radiometric correction, atmospheric correction, and sunglint removal) were
introduced in Chap. 1 and are summarized in Green et al. (2000). This section is
focused on selected examples for the processing steps that follow the initial preprocessing, specifically atmospheric correction, water column correction, image
classification and contextual editing.
Atmospheric correction. The total signal received by satellite sensors is dominated by radiance contributed through atmospheric scattering. Thus, atmospheric
correction is essential to retrieve signals from the sea. The simplest such procedure
is dark pixel subtraction. Based on an assumption that somewhere in the image is a
pixel with zero reflectance, which means the radiance recorded by the sensor is
solely from the atmospheric scattering, the minimum pixel value is subtracted
from all other pixels to remove the radiance derived from atmospheric scattering.
Because images with coral reefs normally contain ocean areas, pixels from deep
ocean areas are often used for this correction. Modeling radiative transfer in the
atmosphere is another more sophisticated option for correction. Mishra et al.
(2005) described a procedure for first-order atmospheric correction for IKONOS
imagery based on radiative transfer theory. Radiative transfer codes such as 6S
(Vermote et al. 1997) and MODTRAN (http://www.modtran.org) may also be
used, and some image processing software has optional modules for atmospheric
correction (e.g., FLAASH for ENVI and ATCOR for ERDAS Imagine).
Water column correction. A fundamental challenge for remote sensing of coral
reefs is the existence of the water column above the bottom features of interest
(i.e., the benthic habitat). Light intensity decreases exponentially with increasing
depth. This is known as attenuation, and the degree of attenuation is a function of
wavelength, water depth and water properties. Water strongly absorbs wavelengths
longer than 700 nm (near infrared), thus visible regions (blue, green and red) are
used for mapping bottom features and infrared regions are used for extracting
shorelines and emergent vegetation.
Rigorous removal of the water influence on bottom reflectance requires
knowledge or estimations of water depth and attenuation characteristics at every
pixel. Calculations using these parameters can be effectively used to estimate reeftop reflectance (‘‘reef-up’’ approach) for substrate mapping (Purkis 2005). In cases
where ground truth data (e.g., light attenuation coefficients, water depth, and
bottom albedo) are available, wavelength-dependency of light penetration offers
high ability to estimate bathymetry of reefs (e.g., Philpot 1989; Stumpf et al.
2003). Recently, Mishra et al. (2005) proposed a method to estimate water depths
and attenuation characteristics using IKONOS data, with minimum requirements
for ground truth data. They identified a ratio of wavebands (blue and green) that is
constant for all bottom types, and developed a polynomial equation to estimate
water depths from the ratio. Attenuation characteristics were then estimated by
3 Multispectral Applications
59