Index | Search | About | Help
It has been the aim of this study to develop a new set of tools that describe landscape form. In particular, the thesis argued is that this can only be achieved effectively by (i) identifying spatial and scale based variation, and (ii) visualising such variation. This concluding chapter considers the success of the approach be re-evaluating the research objectives stated in the introduction, identifying weaknesses of the approach, and avenues for future research.
To assess whether a DEM alone contains useful information for geomorphological characterisation.
This research has been deliberately confined to the analysis of regular matrices of elevation values. Clearly, additional sources of information such as those provided by satellite imagery, and ground survey, provide a more detailed model of landscape. Yet it has been made clear by this work that DEMs are highly information rich descriptions of surface form. It has been argued, that many of the previous attempts to extract information from DEMs have failed to utilise fully, the scale based information contained within the model.
By adopting the parameterisation of surface form first suggested by Evans (1972) it has been possible to examine and quantify local spatial variation systematically. This local description of morphology is sufficiently exhaustive to describe geomorphological form, whilst being sufficiently simple to relate to geomorphological process. Parameterisation has been extended by modelling over a range of scales that remove the surface model from the constraints implied by the DEM resolution. Thus it has been possible to create a more useful source of information for geomorphological analysis.
To identify weaknesses in existing methods of surface characterisation and present new methods to overcome these weaknesses.
The original form of geomorphometric analysis presented by Evans (1972) involved characterisation by quantifying the frequency distribution of surface parameters. Whilst this undoubtedly reveals important diagnostic information, it does little to characterise either the spatial distribution of parameters, or their scale dependency. Quantifications of scale dependencies such as Moran's I spatial autocorrelation index provide little extra information if they are used as a global summary statistic. The solution to this problem suggested by this work is that surface analysis is most effectively carried out within an interactive visualisation context. Consequently, many of the characterisation tools developed have been orientated towards visual representation rather than numerical summary.
More recent geomorphometric analysis has been within the context of the widespread use of Geographical Information Systems. In particular, much research has been devoted to the extraction of hydrological information from DEMs. One of the most significant problems accompanying much of this research has been the apparent mismatch between our own geomorphological understanding of a landscape, and that implied by the raster data model. The multi-scaled parameterisation approach developed here reduces the impact that data model has on subsequent analysis.
To assess the effect that elevation model uncertainty has on surface characterisation.
Two reasons have been identified for the mismatch between true topographic surface form, and its representation as a DEM within a Geographical Information System. Firstly, the model itself provides some conceptual limitations. It is not possible to represent fully, a continuous, undifferentiatable surface with a discrete, finite resolution elevation model. Secondly, the process of elevation interpolation required for DEM generation can lead to model error.
It has been demonstrated that the spatial manifestation of DEM error can, in manycircumstances, be detected in much the same way as true geomorphometric variation. The visualisation of hypsometric distributions demonstrated in Chapter 3, has led to the quantification of the terracing effects resulting from contour interpolation. The visualisation of first and second derivatives of interpolated surfaces has allowed the causes of interpolation error to be hypothesised. Visualising the spatial distribution of DEM error has facilitated the development of a deterministic error model based on planimetric offset of elevation data.
These processes have demonstrated that much the same procedure can be adopted when assessing the surface form that is a function of DEM error, and surface form that is a function of geomorphological process. Yet they have also allowed the two to be separated; error possessing a 'topographic signature' in much the same way as geomorphometry. DEM error has been shown empirically to be largely a high frequency phenomenon (usually over a scale of order 1-5 cells) often possessing artifacts of the original data source. Once surface variation is considered at a scale beyond that implied by the DEM resolution, error effects are minimised.
To assess the effectiveness of visualisation as a means of conveying information on surface form, and as a methodological process.
It has been argued that visualisation has been a necessary step in generating the ideas discussed in this study. The interactive generation of images within a GIS environment has allowed ideas to be hypothesised, tested and confirmed. It was used to hypothesise sources of contour interpolation error and to develop a deterministic error model.
Visualisation has been used to illustrate statistical properties of spatial association in the form of lag diagrams. Such patterns would have been revealed by numerical summary alone. Visualisation has been used as a mechanism for conveying complex multivariate information that could not be effectively achieved by other means. The construction of animated sequences (displayed here as small multiples) of feature classification patterns changing with scale, simultaneously convey the interrelationship between four geomorphometric quantities (altitude,local relief, feature classification and scale dependency).
To produce working software that may be used in a GIS environment for surface characterisation.
The appendix to this volume includes the C source code for all software developed for this study. All modules are fully integrated with the GRASS GIS, allowing the benefits of existing GIS technology to be combined with the multiscale visualisation approach suggested here.
Although tools for geomorphometrical analysis have been developed as part of this study, they have not been used to provide a comprehensive or thorough geomorphological assessment of landscape. For the methodology suggested here to be of any use, it must allow new insights into surface form, rather than being used in a confirmatory context. This can only be achieved by considering the geomorphological context more thoroughly.
Multiscale parameterisation has been used to characterise population density surface properties (Wood et al, 1996). The use of quadratic generalisation appears particularly appropriate for such surfaces that exhibit a low spatial autocorrelation and strongly positive skew. The use of lag diagrams would be appropriate for analysing surfaces that exhibit anisotropic structural properties over a variety of scales. The rippling of fine grained channel bed form, or larger scale dune systems should be readily detectable using this method.
Work is being carried out in using multi-scaled feature classification as part of the modelling cognitive landscape evaluation. The ability to alter the scale at which features are classified makes this a particularly appropriate approach. A similar classification could be used as part of an automated cartographic generalisation process, both of landscape form and cartographic name placement.
One of the weaknesses in the use of textural lag diagrams was the difficulty in relating the measures with known geomorphological characteristics. It might be more appropriate to use the co-occurrence matrix as the basis for inferential hypothesis testing. The co-occurrence matrix has the form of a contingency table used for categorical data analysis. It would be useful to compare the matrix model with some expectation. Yet the standard Chi-squared expectation of independence is highly artificial for such a spatial distribution. Log-linear modelling provides a mechanism for alternative assumptions to be made about expectations. It would be useful to incorporate models of positive spatial autocorrelation into calculations of model expectation so that it would be possible distinguish 'expected' from 'unexpected' surface behaviour.
One of the important results to come from the feature classification process is that a simple Boolean classification of landscape features is not always appropriate. A feature membership function that describes the variation in classification with scale gives a more flexible alternative. Equally, a similar function could describe the change in any morphometric parameter with scale. It would be useful to incorporate such a function into subsequent GIS operations. Fuzzy logic would seem to provide a convenient mechanism for formalising the manipulation of such functions as part of a new type of elevation model.
An alternative elevation model could be developed based on the topological characteristics derived from multiscale quadratic modelling. In particular, the graph theoretic approach suggested by Wolf (1984) provides a parsimonious 'map' of the connectedness of surface features. It is possible to thin point features (ie pits, peaks, passes) and connect with thinned line features (ridges and valleys) as part of a weighted surface network. Such a model could itself be a useful surface characterisation, or alternatively be used for terrain generalisation.
