RSS .92| RSS 2.0| ATOM 0.3
  • Home
  • About
  •  

    Notes from EEO talk on population modelling with GIS

    March 22nd, 2010

    David Martin spoke in the EEO seminar series last Friday. Here are my notes:

    In the last decades we have become “sophisticated in our tools, but our fundamental techniques and results aren’t very different”. Census data is not the same as demographic data, however census approaches to modelling population have become dominant – a “long-term reliance on census-based shaded area map to inform spatial decision-making.

    Importance of small area population mapping for policy – resource allocation and site location decisions, calculation of prevalence rates. “Who is present in a small area, and what characteristics do they have”. A house or flat becomes a “proxy” for a person, who is tied to the space.

    This doesn’t give a clear usage picture, specifically it is night-time activity rather than day time which has very different patterns of repetition and variation of movement.

    More general problems with census-taking –

    • underenumeration
    • infrequency
    • spatially concentrated error

    “We could cut the city differently and produce variations in the pattern” – research in automated generation of census zones, looking for areas with social homogeneity, size, population, based on previous samplings.

    “Population distribution is not space-filling but is quasi-continuous”.

    “Interest in surfaces, grids and dasymetric approaches”. Using a grid to slice and visualise population data. The grid gives us a finer grained depiction of actualy activity.

    Interestingly, shift in government policy regarding census taking. Rapid development of space, and new tech, cause problems – people are more mobile, with multiple bases; concerns about data privacy are more mainstream.
    The US Census Bureau has dropped the “long-form” return which used to go to one in six recipients. In France the idea of a periodic census has been dropped completely, they now conduct a “rolling census” compiled from different data sources.

    “Register-based sources” – e.g. demographic data is held by health services, local government, transport providers, business associations, communications companies. It’s possible to “produce something census-like”, but richer, by correlating these sources.

    Also the cross-section of other sources gives an idea of where census records are flawed and persistently inaccurate, e.g. council tax records not corresponding to where people claim they live.

    Towards new representations of time-space

    Temporal issues still neglected by geodata specialists, in fact some of the issues are gnarlier and trickier than spatial representation is.

    space–time specific population surface modelling.

    Dr Martin identified “emergent issues” affecting this practise- “Spatial units, data sources as streams, representational concepts”. His group has a some software in development to document the algorithm for gridding data space – I wanted to ask whether the software and implicitly the algorithm would be released as open source.

    A thought about gridded data is that it’s straightforward to recombine (given grid cells for different sources are the same size). Something like OGC WCS but much, simpler.