Abstract
One of the key challenges in environmental epidemiology is the exposure assessment of large populations. Spatial exposure models have been developed that predict exposure to the pollutant of interest for large study sizes. However, the validity of these exposure models is often unknown. In this thesis, we present various aspects
... read more
that are important for quantifying the accuracy and uncertainty of model predictions. We focus on modelling radiofrequency electromagnetic fields (RF-EMF) from mobile phone base stations. Accurate positional information is essential for both model validation and model input, especially for exposures with high spatial variability such as RF-EMF. GPS-device are increasingly used to obtain positional information, but little is known about their performance for epidemiological studies. We assessed the outdoor accuracy of GPS-devices and found that GPS-errors were on average a few meters large, but over 14% of the errors were larger than 10 m. This potential displacement should be considered when using GPS-devices to obtain positional information. Next, the validity and uncertainty in RF-EMF exposure modelling with the radio wave propagation model NISMap was assessed. The outdoor performance was validated with repeated continuous measurements along predefined ~2 km long paths in five different areas, and the indoor performance with 15-minute spot measurements in 263 rooms in 101 primary schools and 30 private homes in Amsterdam, the Netherlands. We found a Spearman correlation between modelled and measured total downlink RF-EMF of 0.69 for outdoor, and 0.73 for indoor model predictions. The model was not able to accurately predict absolute RF-EMF levels, as model predictions were often a few factors off from measured values. As in many epidemiological studies not all input data are available, we compared models with various levels of input data detail with our outdoor measurement data set. Results showed that the model was able to rank exposure when 3D-building data and basic antenna information were available: Spearman correlations were generally larger than 0.6. The model performance was not sensitive to changes in building damping parameters. Antenna specific information about down-tilt, type and output power did not significantly improve model performance compared with using average down-tilt and power values, or assuming one standard antenna type. Exposure models are also dependent on the accuracy of the input data. With Monte Carlo simulation we assessed the effect of input uncertainty on modelled RF-EMF. The uncertainty was large with a median coefficient of variation of 1.5. The largest uncertainties in model output came from those inputs that influence whether there is an unblocked line-of-sight between antenna and receptor site: the antenna location, building height and receptor site x, y, z location. Epidemiologists should be aware of uncertainty in the prediction of exposure models. By quantifying the accuracy and uncertainty of the exposure model predictions, both the potential and limitations of the exposure assessment can be judged, improving the reliability of environmental epidemiological studies.
show less
