OLYMPUS-POPGEN: A synthetic population generation model to represent urban populations for assessing exposure to air quality

Abstract

Scientific question: With the new individual- and activity-based approaches to simulating exposure to air pollutants, exposure models must now provide synthetic populations that realistically reflect the demographic profiles of individuals in an urban territory. Demographic profiles condition the behavior of individuals in urban space (activities, mobility) and determine the resulting risks of exposure and environmental inequalities. In this context, there is a strong need to determine the relevance of the population modeling methods to reproduce the combinations of socio-demographic parameters in a population from the existing databases. The difficulty of accessing complete, high-resolution databases indeed proves to be very limiting for the ambitions of the different approaches.

Objective: This work proposes to evaluate the potential of a statistical approach for the numerical modeling of synthetic populations, at the scale of dwellings and including the representation of coherent socio-demographic profiles. The approach is based on and validated against the existing open databases. The ambition is to be able to build upon such synthetic populations to produce a comprehensive assessment of the risk of environmental exposure that can be cross-referenced with lifestyles, indicators of social, professional or demographic category, and even health vulnerability data.

Method: The approach implemented here is based on the use of conditional probabilities to model the socio-demographic properties of individuals, via the deployment of a Monte Carlo Markov Chain (MCMC) simulation. Households are assigned to housing according to income and house price classes. The resulting population generation model was tested in the Paris region (Ile de France) for the year 2010, and applied to a population of almost 12 million individuals. The approach is based on the use of census and survey databases.

Results: Validation, carried out by comparison with regional census data, shows that the model accurately reproduces the demographic attributes of individuals (age, gender, professional category, income) as well as their combination, at both regional and sub-municipal levels. Notably, population distribution at the scale of the model buildings remains consistent with observed data patterns.

Conclusions and relevance: The outcomes of this work demonstrate the ability of our approach to create, from public data, a coherent synthetic population with broad socio-demographic profiles. They give confidence for the use of this approach in an activity-based air quality exposure study, and thus for exploring the interrelations between social determinants and environmental risks. The non-specific nature of this work allows us to consider its extension to broader demographic profiles, including health indicators, and to different study regions.

Fig 1. Map of the Île-de-France region displaying the eight departments and secondary administrative subdivisions comprising the 5200 IRIS units.

Fig 2. Diagram of the conditional probability tree–each circle represents a characteristic associated with a household and an individual in the household.

Each geographical unit is associated with the probability of the number of individuals making up a household. This in turn determines the type of household and the position of the generated individual within the household. Age is then determined, and from this derives the other socio-demographic characteristics that characterize every individual.

Fig 3. Examples of marginal distributions for the real population of the test database and the samples generated using OLYMPUS-POPGEN for the extended attributes.

The bar charts represent distributions at regional level (a) and the scatter plots represent distributions by IRIS (b). The list of variables presented here is not exhaustive. The different colors shown in the scatter plots represent the different attribute classes possible for the corresponding variables.

Fig 4. Spatial distribution of data.

(a) Spatial distribution of households for the city of Paris. (b) Spatial distribution of households in buildings in the 9th arrondissement of Paris. (c) Spatial distribution of households by square of residence (made using census data, INSEE).