Grid models of population: temporal comparison by fixing the geography David Martin, School of Geography University of Southampton
Overview Problems of comparison over time Merits of using gridded data Data sources Grid modelling principles Grid modelling resources Example: Southampton Conclusions
Problems of comparison over time Population changes location and composition Boundaries and definitions are recast between data collection events; measurement success varies Cannot disentangle true change from representation and measurement If geographical units remained the same, one part of the problem would be addressed
Southampton 1991 EDs N = 417 mean address count = 228 Population 197,000
Southampton 2001 OAs (test data) N = 762 mean address count = % overall population increase
Intersection of EDs and OAs N = 1771 mean address count = 54 Where was the increase? (and ED to ED was no better)
Merits of using gridded data Grid cell data comparable through time Settlement geography is retained No boundary data required for geographical comparison Readily combined with gridded environmental models
Data sources Great Britain: 1971 census gridded data published but not available Northern Ireland: grid square data published No general GB grid counts Spatial modelling required population-weighted centroid locations published Postcode locations can be treated as centroids
Grid modelling principles (1) Population-weighted centroid(s) as summary points of local distribution Locally adaptive kernel estimation based on inter-centroid distances Redistribution of centroid counts into grid Could use dasymetric and other methods This is NOT interpolation
Grid modelling principles (2) Weighting Distance w ij d ij k Centroid
Grid modelling principles (3) Centroids and boundaries Gridded population model
NI surfaces 1991 Construct surfaces from NI 1991 ED centroids Experiment with dispersion, postcode centroids, directional interpolators Selective comparison with grid square data Martin, Tate and Langford (2000)
Standard model Postcode centroids Empirical kernel width Directional kernels
Grid modelling resources NI grid square data are true count data for download SURPOP provides 1981 and 1991 ready-made surfaces SurfaceBuilder allows construction of own surfaces from centroid data (URLs on final slide…)
Surpop Requires census data registration 200m cells 1981 or 1991 Select variable Extract window Export to GIS
SurfaceBuilder Download and install VB program Download X,Y,Z centroid data file Specify surface parameters Run and preview model Export to GIS
SurfaceBuilder sequence
Example: Southampton x 25km region centred on City of Southampton Population change from census and postcode data 1980s: large-scale greenfield development at urban fringe 1990s: combined with brownfield redevelopment within urban area
1981 population count 0.2m
1991 population count 0.2m
2001 population count 0.2m
population diff 0.2m
population diff 0.2m
change, 0.2km
t2-t1(t2-t1)/(t2+t1) change, 1km
Contemporary change: postcodes All Fields Postcode Directory providing quarterly postcode record Downloadable from UKBORDERS Very big files! Postcodes as centroids with household counts can be modelled onto grid Census counts could be reallocated
change, 0.2km (AFPD households)
Conclusions Grid-based modelling offers specific advantages over small areas Longer-scale neighbourhood change Monitoring contemporary change Practical modelling advantages Problem of noise due to centroid relocation (requires smoothing) Challenge of definitional drift and measurement error remains
Key URLs Northern Ireland grid square data 001/ni/GRID/index.htm 001/ni/GRID/index.htm Surpop urpop/ urpop/ SurfaceBuilder /martindj/davehome/software.htm /martindj/davehome/software.htm