The effect of spatial generalisation on filtering noise for spatio-temporal analyses

Spatial data sets do not only contain true information, there is also a certain amount of `noise' associated with the data. The use of these data in spatio-temporal analyses, often results in a sub-optimal representation of reality. Generalising spatial data sets collected at different times may serve the purpose of filtering noise so that spatio-temporal change can be better elucidated. In this paper we aim to test that proposition by addressing the following questions. Does generalisation have a significant influence in the state of the noise in space-time dimensions? Can noise be filtered by a generalisation process? Does it result in a greater probability of detecting environmental variation over time? In the first part of the paper, the main aspects of the generalisation process are presented. The field representation (raster model) is described by three elements of analysis; resolution, spacing and extent. Based on these elements, five generalisation methods are analysed. Following an understanding of these methods, the generalisation process is implemented using different land-use data sets obtained from the classification of satellite images, which are then compared at two different times for spatio-temporal analysis. The observed spatio-temporal variations are presented for each method and the filtering of noise is discussed. The importance of deciding which generalisation method to use for spatio-temporal analysis is highlighted. Results show that noise filtration does occur in the generalising process. This may prove that generalising data for spatio-temporal analyses is beneficial to the quality of the results. As noise is filtered, the observed spatio-temporal variation, after the generalisation process, is probably more representative of the true spatio-temporal change in the real world