– In the course of the pandemic, studying the daily updated infection numbers of the Robert Koch Institute (RKI) has become a routine breakfast sideline. However, the job of the data is not finished with its publication. They are processed into a forecast in the “Jureca” supercomputer in a joint project between the Neuroinformatics working group at the University of Osnabrück and the Jülich Supercomputing Center.
The evaluation there currently shows for the district of Roth and the city of Schwabach in unison, in the language of probability mathematicians, “a clear trend towards increasing numbers of infections with a probability of greater than 95%”. For comparison: In Cuxhafen, which has been spared from the virus, Jureca speaks of “a tendency towards increasing numbers of infections with a probability of greater than 75%”.
You could come to such conclusions yourself through pure observation on site, but there are so many recorded numbers, assumptions and model calculations in the data from the supercomputer that you actually need such a mega-computer for processing. In addition to the number of infections recorded by the RKI via the health authorities, “reported and unreported infections” up to a period of five days are taken into account, as are dependencies on the day of the week (decrease in reports on the weekend). Then over 1000 different models are calculated, the probable courses and their limits of reliability (these are called quantiles).
Prediction and Reality
If you want to show 90 percent reliability, you use the fairly widely spread quantile of 5 to 95 percent. With the reference point from last Friday, Jureca accurately predicted the skipping of the 500 incidence in the district of Roth, but in the further course it went up a little steeper than in reality. For Wednesday, for example, the computer set the model value to 703, the RKI determined 538.7. For Schwabach, on the other hand, the Thursday value was almost a direct hit with a prediction of the seven-day incidence of 655 and the RKI value of 628.
To save the honor for the Jülich Recheneisen: The range for the seven-day incidence with a 90 percent probability was in Roth on Wednesday between 317 and 1324. In addition, the method was not developed for Covid-19, but for Campylobacteriosis, rotavirus – and Lyme disease infections. In any case, the results should get better and better as the amount of data increases.
This is what the forecast course of the number of cases looks like for the district of Roth. The points, that is the numbers determined by the RKI, are pretty close to the green line of the model calculation. The light green area shows the leeway the statisticians have to calculate if they want to achieve a reliability of 50 percent (light gray 90 percent).
© Jülich Supercomputing Centre, NN
–
–
The model value of an incidence of 1000 for the district of Roth would be reached on Monday, by the way, for Schwabach it would then be 800. In terms of the number of cases, Jureca is again quite close to the values determined by the RKI (see graphic).
This is what the forecast course of the number of cases for the city of Schwabach looks like. The points, that is the numbers determined by the RKI, are pretty close to the green line of the model calculation. The light green area shows the leeway the statisticians have to calculate if they want to achieve a reliability of 50 percent (light gray 90 percent).
© Jülich Supercomputing Centre, NN
–
–
What’s happening now: A further increase in incidences is accompanied by three coronavirus-related deaths. The RKI reports one from Schwabach, two from the Roth district.
Detailed information at https://covid19-bayesian.fz-juelich.de/
–
