Modelling the distribution of population using ordnance survey vector data sets
The UK Census of Population, collected once every ten years, forms the main data source for demographic analysis. However, pre-2001, the smallest units at which this data was publicly available, known as enumeration districts in England, Wales and Northern Ireland, were irregular and diverse, and these often did not correspond to other survey data sets. For this reason, attempts to integrate these data sources have proved difficult, thus necessitating the urgent development of new methods capable of modelling population distribution with greater accuracy. Within recent years an explosion of georeferenced data has become readily available to both the academic and commercial world, including vector data sets from the Ordnance Survey (OS), specifically - OS Land-Line.Plus and OS MasterMap. These highly detailed vector-based data sets collected and available at scales from 1:1, 250 to 1:10,000, contain the location of many features of the built and natural environment. These features which include road centrelines and building outlines may prove to be appropriate ancillary data sets to better inform the distribution of population. Five regression-based dasymetric models were constructed using combinations of these vector data sets to predict population at a variety of spatial scales and locations in an area of south Leicestershire, UK. This study area, which was split into a source zone of known population and a target zone of unknown population, was chosen due to its diverse mix of both urban and rural areas. Results using the vector-based models were subsequently compared statistically to those from a Landsat dasymetric model. This research not only reveals the varying degrees of success achieved by the newly developed methods, but perhaps more notably so, the superior results obtained by the more complex models in relation to those from the already extensively research Landsat based approach or the simple areal interpolation method.