130
Chapter 14
Secondly, there will always be errors associated with the geo-registration
of remotely sensed data. These errors are encountered, irrespective of
whether the image is registered to a map base (which will itself contain
error) or to a second image; they tend to be at a minimum in flat terrain. In
consequence, it will never be possible to overlay two images precisely so
that the pixelation matches exactly; a proportion of the apparent differences
between images will therefore be due to mis-registration, rather than to real
changes on the ground. These effects are more fully explored in Townshend,
Justice et al. (1992), while Fuller, Barr and Wyatt (1998) demonstrate the
dramatic consequences for classification accuracy of displacements of the
order of a pixel or less.
5.
TEMPORAL ASPECTS
It is also important to reconcile the time and frequency of acquisition of
remotely sensed data with the rate of change in the features of interest.
Changes in land use and land cover occur at a range of speeds. Users with an
interest in the progress and condition of agricultural crops require data at key
intervals in the growing season. Other changes in land cover, deforestation
or flooding, for example, may take place almost instantaneously and nearsynchronous data are required if the precise timing of the event is of interest.
Many change processes take place only slowly, over years or even decades.
In the case of optical imagery, the incidence of cloud is a major constraint on
the frequency with which it is possible to acquire data (Legg 1991, Fuller et
al. 1994). As a result, remotely sensed surveys over spatially extensive areas
may derive from data acquired over extended periods, often of many months,
and it is difficult to establish a precise baseline against which future change
may be measured.
6.
RADIOMETRIC CONSIDERATIONS
The successful detection of land cover from remote sensing is dependent
on the land units of interest exhibiting distinct radiometric responses
(spectral signatures) in the spectral regions covered by the sensor system.
The same consideration applies both to optical systems and to SAR imagery,
though the physical principles are very different. Many changes induce
gradual, rather than sudden radiometric change, which renders their
detection problematic, particularly at intermediate stages. The difficulty is
especially acute in the case of qualitative changes, for example, in drainage,
nutrition or species composition. The advent of hyper-spectral data sources
Chapter 14
Secondly, there will always be errors associated with the geo-registration
of remotely sensed data. These errors are encountered, irrespective of
whether the image is registered to a map base (which will itself contain
error) or to a second image; they tend to be at a minimum in flat terrain. In
consequence, it will never be possible to overlay two images precisely so
that the pixelation matches exactly; a proportion of the apparent differences
between images will therefore be due to mis-registration, rather than to real
changes on the ground. These effects are more fully explored in Townshend,
Justice et al. (1992), while Fuller, Barr and Wyatt (1998) demonstrate the
dramatic consequences for classification accuracy of displacements of the
order of a pixel or less.
5.
TEMPORAL ASPECTS
It is also important to reconcile the time and frequency of acquisition of
remotely sensed data with the rate of change in the features of interest.
Changes in land use and land cover occur at a range of speeds. Users with an
interest in the progress and condition of agricultural crops require data at key
intervals in the growing season. Other changes in land cover, deforestation
or flooding, for example, may take place almost instantaneously and nearsynchronous data are required if the precise timing of the event is of interest.
Many change processes take place only slowly, over years or even decades.
In the case of optical imagery, the incidence of cloud is a major constraint on
the frequency with which it is possible to acquire data (Legg 1991, Fuller et
al. 1994). As a result, remotely sensed surveys over spatially extensive areas
may derive from data acquired over extended periods, often of many months,
and it is difficult to establish a precise baseline against which future change
may be measured.
6.
RADIOMETRIC CONSIDERATIONS
The successful detection of land cover from remote sensing is dependent
on the land units of interest exhibiting distinct radiometric responses
(spectral signatures) in the spectral regions covered by the sensor system.
The same consideration applies both to optical systems and to SAR imagery,
though the physical principles are very different. Many changes induce
gradual, rather than sudden radiometric change, which renders their
detection problematic, particularly at intermediate stages. The difficulty is
especially acute in the case of qualitative changes, for example, in drainage,
nutrition or species composition. The advent of hyper-spectral data sources