Cellular distance to or from
a set areas classified as built-up
ArcView was used to assign values to each cell corresponding with the distance
to the nearest built-up area, the view above shows a classification of
these distance values.
European 1-Decimal-Minute digital map data
The map shows that in England the distance from a built up areas is quite
small and many areas are classed as built up compared with the Mediterranean
region of the EU in general. It seems likely that there is a significant
variation in the interpretation of built-up areas and the resulting classification
across Europe in the source data set. An example of an area which
maybe expected to be classified as a built-up is Bordeux in France, especially
when you consider that York in England is.
The variation in the way built up has been defined in different regions
(an inconsistency in the data source) causes problems. One way to
reduce the effect of the inconsistency in later models (developed in an
iterative manner) is to add new areas likely to be built-up that are currently
missing to enrich the data, for example, one could add built-up areas based
on a classification of population density itself derived from previous
neural network output.
Actual distance on a geographical projection is distorted some what.
People used to maps of the UK based on the Great Britain grid might think
this area appears a bit short and fat compared to usual on the above map.
The measure of distance really being substituted as a proxy for actual
distance is based on the cell distance, where the cells are not equidistant.
For the initial development of the SPS it was thought best to simply use
a geographical projection as a compromise of areal density, distance, location
and direction distortions as the common spatial framework. The geographical
projection compromises the distortions caused by projecting a surface onto
a flat map to some degree. In this case, it would have been better
to reproject the source data into a projection and coordinate system which
minimised distance distortions, then calculate cell distances and then
reproject back again so the data was consistent with the main chosen analytical