The land covermodel was applied in stages. First,10% of the gridded points were randomly selected and assigned a land cover class based on the allocation probability. A further 10% of gridded points were then randomly selected and the adjusted probabilities of class membership calculated by weighting the allocation probability with the stress measure. This process was repeated until all gridded points were attributed with a land cover class. To reduce granularity and ensure that grid cells cluster into realistically sized land cover patches, a majority filter was applied to smooth the land cover surface. Tests were carried out to establish the optimal filter size with regard to shape and size of the resulting land cover class patches and the overall target size for the land cover class. For this purpose, the SIENA land cover model was repeated for the inland cities Leicester and Reading. Filters of various sizes, ranging from 3 3 neighbourhoods to 11 11 neighbourhoods, were applied to the land cover model.