Ensemble modeling and statistical mapping of airborne radioactivity

Abstract

Manufacturing and processing of chemicals and other substances takes place in densely populated areas world wide. Whilst this kind of industry is essential for society, the proximity to populated areas introduces great risks. Accidents such as Chernobyl underline the importance of minimizing the risk that such an accident occurs and when it occurs to take effective countermeasures. The main goal of this thesis was to explore two ways of combining the many sources of information that are available to decision makers in the event of an accident. Note that in this study we focused on accidents involving radioactive material. Our results showed we could effectively design an automatic interpolation system for background radiation levels. A second major contribution was the use of rainfall intensity maps derived from rainfall radar images to improve the interpolated maps of radiation level. Furthermore, this thesis illustrated the use of web-services to effectively distribute the results of the interpolation system through a Web Map Service and discussed the potential use of other web service standards such as a Web Processing Service. Modeling of a tracer dataset showed the great potential of ensemble modeling combined with data assimilation. However, the results also showed that there still was much to gain in modeling actual radiation level patterns. This could be attributed to the scale difference between the model and the observations used for data assimilation. Finally, we concluded that ensemble modeling was preferable to deterministic modeling. First, ensemble modeling quantifies the uncertainty associated with the model predictions, which is essential for decision making. Second, ensemble modeling provides a formal way through data assimilation to use observation to improve the physical model, i.e. data assimilation. The conclusion of thesis was that each of these methods has its own merits and is useful for decision makers. The main advantage of geostatistics is its speed of calculation. A disadvantage is the lack of being able to use interpolation for forecasting and the poor performance under emergency conditions. Physical modeling on the other hand performs well under emergency conditions and explicitly incorporates a lot of the underlying process which causes the spatial pattern in radiation level. However, run times are quite large compared to geostatistics. Both methods quantify the uncertainty associated with their prediction. This allowed us to construct prediction intervals and classify them relative to threshold values. These exceedance maps, which incorporate the uncertainty of the prediction, are of great benefit to decision makers. Although we focused on radioactivity in this thesis, we believe that the results are also useful for other environmental issues such as ground water pollution.