PhD Defense | Jens Peter Frankemölle | Modelling of near-range atmospheric dispersion and the ambient dose equivalent

Modelling of near-range atmospheric dispersion and the ambient dose equivalent: Building blocks for a consistent Bayesian inverse modelling framework

In the event of a nuclear or radiological emergency, there is a need for suitable models and measurements to determine the impact to humans and the environment. Often, such emergencies involve airborne releases of radionuclides. Therefore, models that can predict the spread of radionuclides through the atmosphere are key to estimate the radiological impact. We do not only rely on models, however. Many nuclear facilities are equipped with early warning systems, which monitor the environmental radiation level (i.e., the ambient dose equivalent rate) and which can sound the alarm if some threshold is exceeded. 

In real accidents, detailed information is often lacking in the early stages. For instance, the radiological source term—the number of emitted radionuclides—is often not well known but it is a crucial parameter to estimate the radiological impact. In this thesis, we show how measurements by an early warning system and a near-range dispersion model can be combined to better detect and quantify anomalous releases of radioactivity. 

To that end, we have first developed the ADDER model, which combines a fast Gaussian plume type dispersion model with an algorithm that computes the external radiation from airborne or deposited radionuclides. We have validated ADDER using a variety of real-world case studies—ranging from routine and incidental releases from a nuclear research reactor to a full-scale experiment involving the detonation of an RDD—as well as compared its outputs with various other models. We have compared the predicted concentration and deposition of radionuclides with actual measurements, and put specific emphasis on reproducing environmental radiation measurements.

Leveraging the ADDER model, we have then developed an inverse model which, rather than ingesting input parameters to predict a resulting increase in the environmental radiation level, ingests the measured environmental radiation level to predict the most appropriate input parameters. Since the measured environmental radiation level also includes a background component, which is not the result of an accidental release but is rather caused by naturally occurring radionuclides, we have developed a model to separately account for that background contribution. Relying on Bayesian inference, the combined inverse model can be used to predict not only the radiological source term but can also be used to estimate, e.g., the most likely wind direction and wind speed during the release. In this way, measurements by an early warning network can say much more than simply that a release is occurring.

Promoter:

Prof. dr. ir. Johan Meyers, KU Leuven

SCK CEN mentors:

Prof. dr. Johan Camps

Dr. Pieter De Meutter