My main reserach interests are in the field of data assimilation, atmospheric modeling and spatial statistics. All these topics share the common goal of understanding natural systems using a combination of physical or statistical models and observations. From the models and observations we can make a prediction of the future state of the systems, for example predicting the spread of radioactivity following a nuclear release.
BSc and MSc
I took my first steps on the academic path at Utrecht University where I studied Earth Sciences. I took a broad range of courses during this study ranging from topics such as hydrology, stochastic processes to remote sensing. For my bachelor thesis I optimized the spatial pattern of hydraulic conductivity in a catchment using a genetic algorithm. At the end of my masters, I wrote my masters thesis about extracting watercontent of the canopy from remote sensing images. I did that by inverting a radiative transfer model to try and reconstruct reflection spectra measured by the airborne sensor, HyMap in this case.
After finishing my MSc, I continued with a PhD at the same faculty, Earth Sciences. My PhD was focused on detecting and explaining increased levels of radiation. I did this work in coorperation wit the National Institute of Public Health and the Environment (RIVM). In this work I used both observations from a monitoring network and atmospheric transport models.
Interpolation of monitoring data
I compared two methods of using these sources of information for nuclear decision managment. The first method was focused on interpolation of the observations using spatial statistics. The interpolated map provides a comprehensive overview of radiation level, including non-observed locations.
The previous method method works well in situations with background radiation levels, or mild elevations above that level. However, during an accidental nuclear release the statistical model underlying the interpolation becomes invalid because of the large elevations in radiation level, i.e. stationarity does not hold. This brings us to the second method of combining the available sources of information: atmospheric modeling combined with data assimilation. In this approach we generate an ensemble of possible model realizations by drawing from probaility distribution functions (PDFs) that we assigned to the most important model parameters in the atmospheric transport model. This can be for example wind direction or wind speed. We run the ensemble of model realizations forward in time until we have observations of radiation level available. At that time, we use the observations to determine how well each of the model realizations performs. Well performing realizations are cloned and poorly performing realizations are eliminated. This data assimilation algorithm is known as a particle filter. In contrast to the first method, interpolation, this method performs well in the case of a nuclear release.
The conlusion in my PhD thesis was that both methods have their use in the practice of nuclear decision managment. Interpolation works well in background situations, the particle filter in case of a nuclear release.
After successfully defending my PhD thesis, I started working as a post-doc at the Royal Netherlands Meteorological Institute (KNMI). My research there is within the EU FP7 project SUMO and is focused developing a novel computational strategy to improve climate simulations.