Near-real-time sensing and modelling of human activity at geographic feature-level

There is growing interest in the estimation of detailed small area population estimates at different times of day and different day types for a wide range of applications, ranging from planning for emergencies to planning convenience retail provision. This proposal concerns development of a conceptual framework and practical application to harvest real time and near-real time human activity profiles associated with Ordnance Survey (OS) map features for dynamic population mapping. It builds on previous ESRC-funded work by the potential supervisors and a previous OS-sponsored EPSRC iCase studentship, which have established a broad modelling framework for estimating time-specific populations at high spatial resolution, but which do not presently consume real time data.

Related previous work provides a substantial body of knowledge on which to build, but has involved intensive manual data preparation, and has thus not been directly scalable or readily updatable. Recent work has established the value of classifying addresses by activity type and the feasibility of allocating activity time profiles to addresses and small areas.

A key objective of the proposed research is to develop and implement an architecture for real-time harvesting of human activity data into a dynamic model that is directly scalable and updatable. This would be based on a combination of online data feeds (e.g. traffic sensors, popular times, etc.) and their association with appropriate OS object identifiers, as well as time signatures obtained from previously harvested activity data (e.g. retail footfall sensors), from which it is possible to analyse recorded activity patterns by time of day, day of week, etc. The emphasis is on establishing a scalable protocol for integration of these sources that may be used in combination with more conventional population data. It is not anticipated that this research will specifically focus on social media activity, which has relatively small population and geocoding coverage.

The research will require further development of both concepts and software tools, including a linkage model that classifies cartographic features to which directly sensed live data can be applied in real time, and a design for population activity harvesting on a large scale that could continue to be delivered in future. The potential supervisors already work with a range of industry stakeholders interested in these topics.

The candidate will have a first or strong upper second class degree in a quantitative social science subject such as Geography or Social Statistics, or Computer Science and will have programming experience and a strong interest in big data and mapping.