Population Surface Generation
The initial focus has been on generation of 1km grid cells using Census data for 1971 to 2011. After experimentation using land use data and postcodes to inform reallocation of counts to grid cells, we have developed an approach to grid creation based on the overlay of source zones (enumeration districts or outputs areas) and grids representing densities of postcodes. Code has been written in the R programming language to automate generation of population grids for any available variables and for any Census between 1971 and 2011 inclusive. The grid generation process works as follows:
- Postcode density is computed for a 1km grid using kernel estimation
- This postcode density grid is overlaid with the source zones
- The population of the areas of overlap is computed using a weighted sum (areas with large postcode densities receive proportionately more people)
- The overlapping areas are aggregated by grid code to give population estimates for 1km cells
- The cell values are smoothed so as to make neighbouring cells values more alike
The inputs are EDs or OAs for the 1971, 1981, 1991, 2001 and 2011 Censuses in Britain. Postcode locations for 1981, 1991, 2001 and 2011 were used to compute postcode densities. For 1971, postcodes are not available and so the 1981 postcodes were used in that case.
Fig.1 and Fig.2 are example grids generated using postcode densities (but without the smoothing step).
Please check back soon – we are currently in the process of generating 1km grids for variables including (where available for each Census from 1971 to 2011 inclusive) age, sex, ethnicity, country of birth, health status, employment status, housing tenure, qualifications, and car and van access. R code will also be provided to allow users to generate grids for any additional variables they are interested in. The development of 100m grids is also in progress, although using a slightly different (less memory-heavy and computationally intensive) approach than that used to generate 1km grids.