Disaggregating heterogeneous agent attributes and location

• Disaggregate individual agents using aggregate data and empirical research.
• Both attribute and location of agents are generated as a GIS layer for ABMs.
• Developed a shared tool for agent disaggregation in the virtual and real cities.
• The approach applies to conditions with no micro-data.

The use of micro-models as supplements for macro-models has become an accepted approach into the investigation of urban dynamics. However, the widespread application of micro-models has been hindered by a dearth of individual data, due to privacy and cost constraints. A number of studies have been conducted to generate synthetic individual data by reweighting large-scale surveys. The present study focused on individual disaggregation without micro-data from any large-scale surveys. Specifically, a series of steps termed Agenter (a portmanteau of “agent producer”) is proposed to disaggregate heterogeneous agent attributes and locations from aggregate data, small-scale surveys, and empirical studies. The distribution of and relationships among attributes can be inferred from three types of existing materials to disaggregate agent attributes. Two approaches to determining agent locations are proposed here to meet various data availability conditions. Agenter was initially tested in a synthetic space, then verified using the acquired individual data, which were compared to results generated using a null model. Agenter generated significantly better disaggregation results than the null model, as indicated by the proposed similarity index (SI). Agenter was then used in the Beijing Metropolitan Area to infer the attributes and location of over 10 million residential agents using a census report, a household travel survey, an empirical study, and an urban GIS database. Agenter was validated using micro-samples from the survey, with an average SI of 72.6%. These findings indicate the developed model may be suitable for using in the reproduction of individual data for feeding micro-models.