STEAM: Comparative evaluation of Spatial-Temporal Exposure Assessment Methods for estimating the health effects of air pollution

STEAM is led by King's College London. St George's University of London, Imperial College, the National and Kapodistrian University of Athens and Prof. Joel Schwartz from Harvard University are Partners. Thus, to achieve its objectives STEAM takes advantage of the expertise and synergies within the MRC Centre for Environment and Health (Imperial College, King's College London, St George's and Public Health England) and draws upon the expertise of pioneering international groups.

Background and Objectives

The investigation of health effects of exposure to air pollution has, in the past, relied on measurements of air pollutants concentrations by fixed monitoring sites. Did these represent well the population exposure and the variability between individual exposures? Certainly, the representation was suboptimal...More recently, the development of models has attempted to improve exposure assessment and enabled the investigation of a broader range of pollutants. However, several types of models have been developed which vary both in the scientific methods and in the details and no systematic validation and comparison of their performance has been undertaken.
The FIRST OBJECTIVE OF STEAM addresses the systematic comparison of various modelling methods and the development of the best models to address the issues related to estimating the effects of air pollution on health. The project will include ambient particulate matter of different sizes (PM10, PM2.5) and gaseous pollutants (NO2, O3). Studies have linked long- and short-term outdoor air pollution exposures to increased risks of death and morbidity. The design of studies assessing acute effects and that of studies assessing chronic effects differ with the former exploiting the temporal variability in pollution concentrations and the latter exploiting spatial heterogeneity in annual air pollution concentrations. How do these adverse health effects overlap and compare in significance and magnitude?
Our models will provide spatially and temporally (daily) resolved estimates of pollutant concentrations to enable, as SECOND OBJECTIVE, an integrated assessment of health effects arising from both long- and short-term exposures to the above pollutants. The models rely on expensive and technically demanding measurements to test their validity and robustness. But how many measurements are really adequate?
STEAM, as a THIRD OBJECTIVE, will use the London area extensive monitoring system to test how many monitors would be adequate to support a model that takes advantage of many sources of information (such as satellite, land use, traffic data) and thus provide valuable knowledge for planning surveillance systems in other parts of the world.

Methods and implications

STEAM will include the development and validation of dispersion models, statistical models (termed "Land Use Regression" models) and models using information from remote sensing to compare and combine modeling approaches. For this purpose, it will use simulated data (where the "real" effect of short and long-term air pollution is known) and data from real health data bases. Through these methods it aims to provide information on:
1. the advantages and disadvantages of each exposure model to inform both model choice and the optimum extent and density of monitoring systems. 2. the simultaneous estimation of the effects of short and long-term exposures to serve as a basis for the balance between emergency short-term action and long-term pollution management interventions. 3. the development of modelling in areas of Europe and the World where monitoring is not so dense. 4. the comparative importance of short vs long-term effects.

Stakeholder involvement

STEAM is inviting stakeholders to express their views on the scope and usefulness of the project and contribute their ideas and suggestions

Study outputs

Within the STEAM project we used a machine learning method to enhance the PM2.5 data base of daily values. This was done by modelling relationships between PM10 and PM2.5 where both pollutants were measured, then utilizing this relationship to ‘convert’ the more widespread PM10 measurements to derived PM2.5. A full explanation of the methodology is included here. We think that this enhanced data base may be useful for other researchers or policy makers. If you are interested in using it please send a request, briefly explaining your objectives to or

© Imperial College London, Environmental Research Group. All rights reserved 2020.