Modelling air pollution for epidemiologic research – Part II: Predicting temporal variation through land use regression

Over recent years land use regression (LUR) has become a frequently used method in air pollution exposure studies, as it can model intra-urban variation in pollutant concentrations at a fine spatial scale. However, very few studies have used the LUR methodology to also model the temporal variation in air pollution exposure. The aim of this study is to estimate annual mean NO2 and PM10 concentrations from 1996 to 2008 for Greater Manchester using land use regression models. The results from these models will be used in the Manchester Asthma and Allergy Study (MAAS) birth cohort to determine health effects of air pollution exposure.The Greater Manchester LUR model for 2005 was recalibrated using interpolated and adjusted NO2 and PM10 concentrations as dependent variables for 1996-2008. In addition, temporally resolved variables were available for traffic intensity and PM10 emissions. To validate the resulting LUR models, they were applied to the locations of automatic monitoring stations and the estimated concentrations were compared against measured concentrations.The 2005 LUR models were successfully recalibrated, providing individual models for each year from 1996 to 2008. When applied to the monitoring stations the mean prediction error (MPE) for NO2 concentrations for all stations and years was -0.8 μg/m³ and the root mean squared error (RMSE) was 6.7 μg/m³. For PM10 concentrations the MPE was 0.8 μg/m³ and the RMSE was 3.4 μg/m³.These results indicate that it is possible to model temporal variation in air pollution through LUR with relatively small prediction errors. It is likely that most previous LUR studies did not include temporal variation, because they were based on short term monitoring campaigns and did not have historic pollution data. The advantage of this study is that it uses data from an air dispersion model, which provided concentrations for 2005 and 2010, and therefore allowed extrapolation over a longer time period.

Research highlights
► Few studies have used LUR to model temporal variation of air pollution.
► Temporal recalibration of an existing LUR model seems to be the most suitable methodology.
► The 2005 LUR models for Greater Manchester were successfully recalibrated, providing models for each year from 1996 to 2008 .
► Validation against automatic urban monitoring stations showed relatively low prediction errors.