Help from an unexpected quarter: geosciences data techniques can help predict corona spread

News - 11 June 2020 - Webredactie

An international team of scientists is studying the possibility of using data assimilation, a data technique from geosciences, to predict the spread of coronavirus and the efficacy of certain measures. TU Delft geoscientists on the team think that data assimilation might prove a useful tool for the RIVM epidemiologists. A paper on the subject has been submitted to scientific journal Foundations of Data Science.

Wider range of applications
‘In geosciences we typically predict things like subsurface behaviour, or the weather, based on uncertain data. The mathematical method we use for this, and have done successfully for decades, is data assimilation,’ Femke Vossepoel of the TU Delft’s Civil Engineering and Geosciences department explains. ‘It’s a method that can be applied much more widely, to include any problem that can be modelled, really. That struck me, even before corona, every time I read something in the paper about models for, for example, ecosystems or financial markets.’

So when Norwegian fellow researcher Geir Evensen approached her to take a closer look at the spread of coronavirus, Vossepoel leapt at the chance. With 11 colleagues from seven different countries, including Brazil, Argentina, the UK and France, she got to work. ‘We have spent the last couple of weeks modelling the spread of the virus. The fact that so many countries have become involved makes it pretty unique.’

Hospital admissions
‘Our model, which is fairly simple, is based on the numbers available for each age category and shows the evolution of the number of patients, the number of hospital admissions, the number of ICU patients and the number of coronavirus deaths. We start with a wide range of possible values to determine the model’s parameters, for instance the now well-known R number (the number of people that one infected person will pass the virus on to, on average). These values are then calibrated with the real data so the predictions of the model correspond with the actual measurements, taking into account their uncertainty.’

The team used the model to look at the datasets of different countries, each with a very different development of the outbreak. ‘We can model the outbreak of corona in the various countries very precisely. This results in new and better parameter values, such as R, including, and this is very important, the associated uncertainties. With this improved set of model parameters we can then make reasonably accurate short-term predictions, typically for a period of two weeks,’ Vossepoel says.

Scenarios for the longer term can be calculated as well, which might include measures such as closing or opening up schools. ‘For longer-term predictions we depend on the accuracy of the medical and epidemiological assumptions. We are well aware that that is not our area of expertise but that of institutes like RIVM. We think that we can really make a difference if we can team up with them to make their predictions of the spreading of the virus even better.’