affected by the experience of the pilot and operator in collecting this type of data,
and on the wind conditions. Another factor to be aware of is that instrument
maintenance and deployment is more variable than with satellite platforms. If the
instrument is not maintained correctly dust on the sensor and other factors can
cause vertical striping in the data, and electrical noise on-board can cause horizontal striping.
For benthic mapping the instrument should have its radiometric sensitivity
adjusted to be optimal for the relatively dark sub-surface reflectance. It can be
difficult to achieve unsaturated terrestrial data from a sensor tuned for below water
applications, especially in the tropics where terrestrial surfaces may include high
reflectance coral sand. This impacts the ability to use terrestrial targets as references for atmospheric correction. Operational satellites suffer from these problems
less. Their mission requirements and instrument operational characteristics are
clearly defined and maintained with long term and multiple users in mind, rather
than for ad hoc site deployments for a single customer.
Some airborne sensors such as CASI allow for configurable band wavelengths.
In this case it is worth reviewing the literature and features of the study site to
position bands in wavelengths that are likely to provide useful information.
Table 4.2 shows a published justification of CASI band selection used in a coral
reef application (de Vries 1994). However, Table 4.2 should not be taken as
definitive and is slightly out-dated; for example near infrared bands are unlikely to
show red edge chlorophyll features due to the high absorption by water (Fig. 4.2).
The features of the site of interest and recent results on spectral discrimination
should also be considered (Hochberg et al. 2003a). Hedley and Mumby (2002)
review reef pigment spectral features and their relation to hyperspectral remote
sensing.
4.2 Hyperspectral Planning and Preprocessing
A number of distinct approaches exist for mapping or quantifying benthic composition and other biophysical parameters in shallow waters by optical remote
sensing. Almost all the described methods are equally applicable to multispectral
or hyperspectral data in a practical sense, however the quality of outputs will vary
and some approaches have been designed with hyperspectral data in mind. When
choosing a method to use the first consideration must be what data is required to be
extracted from the imagery for the given application. This must be tempered with
what is likely to be actually possible and the practical challenges in applying a
specific method. Published methods differ significantly in how complicated they
are to apply: from methods that can be applied with little or no image preprocessing and using standard software, to methods that require rigorous atmospheric
corrections and custom code. It is important to bear in mind that published results
tend to show a positive bias, results where a method performed poorly are far less
likely to be published than those where a method worked well. In addition a
86
J. D. Hedley
and on the wind conditions. Another factor to be aware of is that instrument
maintenance and deployment is more variable than with satellite platforms. If the
instrument is not maintained correctly dust on the sensor and other factors can
cause vertical striping in the data, and electrical noise on-board can cause horizontal striping.
For benthic mapping the instrument should have its radiometric sensitivity
adjusted to be optimal for the relatively dark sub-surface reflectance. It can be
difficult to achieve unsaturated terrestrial data from a sensor tuned for below water
applications, especially in the tropics where terrestrial surfaces may include high
reflectance coral sand. This impacts the ability to use terrestrial targets as references for atmospheric correction. Operational satellites suffer from these problems
less. Their mission requirements and instrument operational characteristics are
clearly defined and maintained with long term and multiple users in mind, rather
than for ad hoc site deployments for a single customer.
Some airborne sensors such as CASI allow for configurable band wavelengths.
In this case it is worth reviewing the literature and features of the study site to
position bands in wavelengths that are likely to provide useful information.
Table 4.2 shows a published justification of CASI band selection used in a coral
reef application (de Vries 1994). However, Table 4.2 should not be taken as
definitive and is slightly out-dated; for example near infrared bands are unlikely to
show red edge chlorophyll features due to the high absorption by water (Fig. 4.2).
The features of the site of interest and recent results on spectral discrimination
should also be considered (Hochberg et al. 2003a). Hedley and Mumby (2002)
review reef pigment spectral features and their relation to hyperspectral remote
sensing.
4.2 Hyperspectral Planning and Preprocessing
A number of distinct approaches exist for mapping or quantifying benthic composition and other biophysical parameters in shallow waters by optical remote
sensing. Almost all the described methods are equally applicable to multispectral
or hyperspectral data in a practical sense, however the quality of outputs will vary
and some approaches have been designed with hyperspectral data in mind. When
choosing a method to use the first consideration must be what data is required to be
extracted from the imagery for the given application. This must be tempered with
what is likely to be actually possible and the practical challenges in applying a
specific method. Published methods differ significantly in how complicated they
are to apply: from methods that can be applied with little or no image preprocessing and using standard software, to methods that require rigorous atmospheric
corrections and custom code. It is important to bear in mind that published results
tend to show a positive bias, results where a method performed poorly are far less
likely to be published than those where a method worked well. In addition a
86
J. D. Hedley