Spatio-temporal generation of large-scale hazardous events that may cause flooding

Diederen, Dirk (2022). Spatio-temporal generation of large-scale hazardous events that may cause flooding. PhD thesis The Open University.



This thesis treats the generation of weather scenarios that may cause flooding, generally referred to as events, within the context of flood risk. A key concept for the risk approach is that not only hazardous events are generated, but that realistic probabilities, or frequencies, are connected to the events. The purpose of this thesis is to provide some methodological improvements on flood hazard generators and to bring into discussion some useful concepts with regards to spatio-temporal challenges.

The flood risk approach is largely data driven. The methods used in this thesis start from obtained data sets, provided by others. General computer algorithms and statistical models are used to generate hypothetical, or `synthetic' data sets. These synthetic data sets are investigated and analysed using computational code.

The starting point is the generation of a large synthetic set of pan-European river discharge events, spread out over several hundreds of locations in Europe. Some methodological advances are provided, which allow moving discharge waves to be tracked throughout all major river basins in Europe. A key point in the used methodology is to capture the spatial dependence between events occurring at different locations, which will be referred to as the static spatio-temporal approach. Compared to a local approach, where each location is considered individually, the gains of considering spatial dependence seem rather clear. However, it appears that the static spatio-temporal approach does not work well with an event-based approach, since this methodology implicitly puts boundaries on the procedure of spatio-temporal event identification.

Therefore, the next step was the development of a generator that could provide large synthetic sets of precipitation events, over the entire Atlantic sea and Europe. A key point in the methodology used here is that not only the events are dynamic, as their movement is tracked, but also the event descriptors are dynamic. Hence, here the dynamic spatio-temporal approach is introduced, which implies the generator methodology moved beyond generation at a fixed set of locations. The methodological framework is established, which allows a first version of such a dynamic generator, with the potential to be applied globally. And some exploration is provided of methodological extensions that allow to treat multiple variables concurrently in a compound framework.

With the newly introduced complexity of the dynamic generator, the final step was to start generating big synthetic data sets and to thoroughly test what the generator produced. A comprehensive sensitivity analysis provides some first insights into the behaviour of the generator, which allow to understand two key concepts. First, with the dynamic spatio-temporal approach, spatial coherence of extremes comes naturally. This contrasts with the static spatio-temporal approach where spatial coherence of extremes has to be assumed prior to the modelling and is typically done by the application of a spatial process. Second, for any location, the dynamic spatio-temporal approach allows to more directly include events occurring in the area surrounding a location to compute extremes at that particular location. Relatively short data records are a standard limitation for the risk approach, whereby this `dynamic expansion of information' may be of help.

The provided methodological advances and new concepts may help the way forward to the generation of global hazard events that may cause flooding. Large-scale, coherent event sets allow interactions and system behaviour to be studied, which is a main requirement to be able to compute the system risk. In addition, an interesting outlook is provided for future research. The dynamic spatio-temporal approach may in the future be able to provide not only spatially coherent extremes, but also temporally coherent extremes. This could be a first step towards credible global flood risk time series, which could be a very useful tool, in the current predicament of global climate change.

Viewing alternatives

Download history


Public Attention

Altmetrics from Altmetric

Number of Citations

Citations from Dimensions

Item Actions