Analysing spatial patterns and representations of agricultural land use data

This thesis aims to propose new methodologies for spatial data analysis that are relevant to agricultural land use modelling. Spatial land use data are most commonly represented in raster format, but the real landscape is not composed of regular spatial units. Thus, when a landscape is rasterized at different spatial resolutions it is important to know the reliability and stability of the resulting datasets. Large changes in cell size are known to have an effect on the accuracy of raster maps and the information they contain. The work presented here, however, shows that even small changes in cell size can substantially affect raster representations of landscapes. The instability of raster maps raises doubts about the results of land use analyses that use spatial data at a single resolution. The CORINE and PELCOM datasets, for example, are single resolution datasets that are commonly used in land use studies. Both CORINE and PELCOM, however, were found to strongly overestimate arable land use and neither was able to correctly represent minor land use classes. Rather than use such general datasets at a coarse resolution, the aggregation of fine resolution spatial data is recommended. A new method to assess the quality of raster maps, using a parcel, rather than a pixel-centred approach, allows parcels or zones that are insensitive to changes of scale to be identified and mapped. Thus, the appropriateness of raster data for different spatial analysis and modelling approaches can be evaluated. In an analysis of the factors determining spatial structure within the agricultural landscape of the study area, the relative location of fields to farms was shown to be very important. Surprisingly, however, farmer imitation was not found to affect land use patterns. These results have important implications for the modelling of land use patterns.