• Are adequate personnel resources available to process the data? Bear in mind
simply acquiring the data may be a small fraction of the work that needs to be
done.
• What long term resources can be allocated to the dataset? For the very best
value, datasets should be treated like ‘living’ entities that receive constant use
and attention. The longer a dataset is left unused the less likely it will ever be
used again.
What other data beyond the imagery itself will be required or useful?
• For spatial ground truth of benthic composition, how much, what spatial sampling strategy and what level of ‘descriptive resolution’ is needed?
• Are spectral libraries of benthic type reflectance needed? These are required as
input for some methods, but of little practical use for others.
• Are in-water measurements of light fields, constituents or optical properties
needed?
• If atmospheric correction is required, above water in situ reflectances measured
concurrently with image acquisition for vicarious calibration may be very
useful.
What is the status of the method in terms of acceptance and use by
practitioners?
• How ‘tried and tested’ is the technique? Will the wider community find the
results easy to accept and interpret? Is this a ‘standard procedure’ that has been
used many times and is well understood?
• How is accuracy reported? Is this easy to interpret and appropriate for the
intended application?
• What is the transferability of a method? A method may have been successfully
demonstrated at one site but will it work elsewhere if a site has different
composition or bio-optical properties?
4.2.2 Preprocessing Considerations
To understand what types of preprocessing are required for different analysis
techniques it is helpful to distinguish between two types of image processing
algorithms. The first, ‘data manipulation’, simply performs calculations on the
image data without incorporating additional information that varies from pixel-topixel. For example, subtracting a dark pixel spectrum from every pixel in an image
to perform a basic atmospheric correction, or generating a new data layer as the
ratio of the values in two other bands. Clearly this type of processing does not add
any information to the image on a pixel-by-pixel basis. The new data values are
simply a transform of the existing data at each pixel. In fact, information may be
lost, for example a band ratio transform discards information about the differing
88
J. D. Hedley
simply acquiring the data may be a small fraction of the work that needs to be
done.
• What long term resources can be allocated to the dataset? For the very best
value, datasets should be treated like ‘living’ entities that receive constant use
and attention. The longer a dataset is left unused the less likely it will ever be
used again.
What other data beyond the imagery itself will be required or useful?
• For spatial ground truth of benthic composition, how much, what spatial sampling strategy and what level of ‘descriptive resolution’ is needed?
• Are spectral libraries of benthic type reflectance needed? These are required as
input for some methods, but of little practical use for others.
• Are in-water measurements of light fields, constituents or optical properties
needed?
• If atmospheric correction is required, above water in situ reflectances measured
concurrently with image acquisition for vicarious calibration may be very
useful.
What is the status of the method in terms of acceptance and use by
practitioners?
• How ‘tried and tested’ is the technique? Will the wider community find the
results easy to accept and interpret? Is this a ‘standard procedure’ that has been
used many times and is well understood?
• How is accuracy reported? Is this easy to interpret and appropriate for the
intended application?
• What is the transferability of a method? A method may have been successfully
demonstrated at one site but will it work elsewhere if a site has different
composition or bio-optical properties?
4.2.2 Preprocessing Considerations
To understand what types of preprocessing are required for different analysis
techniques it is helpful to distinguish between two types of image processing
algorithms. The first, ‘data manipulation’, simply performs calculations on the
image data without incorporating additional information that varies from pixel-topixel. For example, subtracting a dark pixel spectrum from every pixel in an image
to perform a basic atmospheric correction, or generating a new data layer as the
ratio of the values in two other bands. Clearly this type of processing does not add
any information to the image on a pixel-by-pixel basis. The new data values are
simply a transform of the existing data at each pixel. In fact, information may be
lost, for example a band ratio transform discards information about the differing
88
J. D. Hedley