1 Spatial Modelling of the Terrestrial Environment - Georeferencial

134 Spatial Modelling of the Terrestrial Environment
and also through to estimates of volumes of change. This showed that these errors were
reduced to such a level that the DEMs could be used to detect erosion and deposition
patterns and to determine volumes of material eroded and deposited. Table 6.6 also shows
the theoretical estimates of precision that the design of this study was based upon, using the
scale of imagery acquired and the scanning resolution used to create digital imagery from
diapositives. This shows how the theoretical precision is seriously downgraded in practice
as a result of the effects of surface structure upon the data collection process. If a certain data
quality is critical, room for degradation of the theoretical precision is required. Table 6.6
also shows how the final data quality improved more than expected when the image scale
was increased as compared with the improvement in theoretical precision. This emphasizes
the sensitivity of the data collection method used in this study to surface texture, and this
will apply to any study where stereo-matching is used for surface representation.
6.7 Conclusion
This chapter has described the way in which we developed systems for managing error in
a complex digital photogrammetric project. The need to develop these systems arose from
an increasingly common problem for remote sensing in general: the growth of automated
methods for data generation reduce the traditional interpretation of individual data points;
whilst this benefits data collection processes by allowing data to be acquired over a much
larger area and much more frequently, problems arise if there is degradation in the quality
of results obtained. In relation to environmental modelling, this is of particular concern,
because the parameters that determine earth surface processes tend to be derivatives of a
surface (e.g. slope, difference between two surfaces, surface curvature), and not the surface
itself. As equations (3)–(5) show, errors will propagate into these derivatives so that
subtle errors can have quite severe consequences at the point at which remotely sensed
data is used for process modelling. It also follows that process modellers must become
closely involved in the data generation and interpretation process in order to make sure that
avoidable decisions that affect data quality adversely (e.g. certain types of point filtering
algorithms) are not taken.
The process of error management we used in the Waimakariri case example was based
upon the identification of three types of error: systematic error, blunders and random error.
The work shows that these sorts of errors are manifest and need to be treated in different
ways. Most notably, error correction methods needed to reflect: (1) how the method used
to generate the data introduces error (e.g. the banding that emerged during DEM stitching,
associated with uncertainties in sensor position and orientation); and (2) how the surface
determines the nature of error that arises (e.g. problems associated with sudden breaks of
slope close to the water edge). In this case, the most important progress was associated with
developing methods for finding spatially-localized error using a form of smoothing based
upon the stereo-matching algorithm adopted. This approach has the potential to be of more
widespread use in terms of other surface types and terrain collection methods, wherever
there are a relatively small number of seriously erroneous data points within a surface that
is generally good. In principle, this means that an approximate surface representation may
be achieved as a means of identifying isolated data points that are in error. In practice,
research is required to assess the best means of obtaining both an approximate surface