The utilisation of geographic information systems and digital image processing techniques for the construction of digital landscape models necessitate for a reconsideration of the classical concepts for landscape ecological mapping. In this thesis, some methods are presented for the spatial modelling and monitoring of natural landscapes based upon digital workflow information.
In spatial information processing, two major approaches for the conceptual representation of spatial features are distinguished, the field and spatial object respectively. A field is a feature which is contiguously distributed over space and time. The object approach, on the contrary, assumes that the earth´s surface is populated with spatially interacting discrete units. Because natural landscapes often show both continuous and discrete variation in space and time, a hybrid terrain description is proposed, denoted as ´spatial object with nested field´. In this hybrid approach the discrete landscape patterns are described by spatial objects, while the internal spatial variability within an spatial object is represented by a field.
Classification is applied during the construction of the spatial objects and nested fields, because it is acknowledged to be a powerful technique to extract essential information from the background of infinite complexity. Crisp classification yields discrete attribute values and is therefore suitable for the definition and construction of spatial objects. The representation of continuously varying terrain features requires a continuous type of classification, i.e. fuzzy classification. Throughout this thesis, fuzzy classification is applied to construct fields.The concepts for spatial modelling, that were introduced above, were used in three cases resulting from the landscape management practice in the Amsterdam Waterworks Dunes:
Spatio-temporal mapping of the vegetation structure from high resolution CIR-images
Two radiometrically corrected, digital colour infrared orthophotos from the summer of 1990 and 1995 with a resolution of 0.25 metre were semi-automatically interpreted. Crisp and fuzzy classification techniques were applied to construct the spatial objects and their nested fields, representing the vegetation structure of the test site. Compared to manual photo-interpretation, the semi-automatic interpretation of vegetation structure results in a more realistic, more detailed and less subjective digital representation of the landscape.
Subsequently, the vegetation structural dynamics were explored on the basis of this multi-temporal data set. Methods are presented to answer two primary questions relevant to nature managers, regarding the turnover between cover types and the changes in the spatial structure. It appeared necessary to aggregate the spatial objects provided by the image interpretation to composite objects prior to the spatio-temporal analysis, because thematic and geometric inaccuracies in the data can yield faulty analysis results.
Estimation of the spatial distribution of vegetation communities from environmental data
In addition to information about vegetation structural dynamics, there is a need for information on changes in the vegetation composition. This information can be provided by a multi-temporal map of vegetation communities. An experiment was conducted to estimate the presence of vegetation communities from environmental data, including vegetation structural data.
A reliable procedure for the automatic mapping of vegetation communities from environmental data requires a fine tuned definition of the vegetation communities and a powerful explanatory model. The first condition was met by applying the concept of fuzzy vegetation communities. After the optimisation of the degree of vagueness, the fuzzy vegetation community types show a closer resemblance with the vegetation abundance data in the relevees compared to the classical crisp vegetation classes. The second objective, i.e. the construction of a powerful explanatory model, was not successfully achieved. Clearly, vegetation structural data are not sufficient to map the vegetation composition in dunes. Additional information regarding some abiotic site characteristics seem indispensable to improve the explanatory power of the model.
Fuzzy ecohydrological expert modelling of dune slacks
The thesis also describes the fuzzy ecohydrological expert model ECOMOD. The model predicts the primary parameters for the specification of ecotopes, i.e. vegetation structure, moisture content, nutrient availability and acidity of the soil. ECOMOD is calibrated for dune slacks by a team of experts. These experts estimated the parameters of the membership functions that quantify the fuzzy classes and fuzzy relations constituting the model. Experts apply vagueness or fuzziness in order to quantify the continuous turnover between classes and to quantify the uncertainty related to their knowledge. Therefore, fuzzy expert models like ECOMOD enable the robust modelling of complex systems, even if relatively little data is available. Models like this are particularly suited for scenario studies. ECOMOD is a generic ecohydrological model that can be adapted and calibrated for other ecohydrological systems.