District-Based Coronavirus Forecasts

Neuroinformaticians at Osnabrück University and the Jülich Supercomputing Centre (JSC) are providing new model results for predicting COVID-19 infections on a daily basis. The results include daily updated estimates of reported new infections and a 5-day forecast for each German district. It can be accessed via an interactive dashboard hosted on JSC's OpenStack HDF Cloud. The forecasts are based on data from the Robert Koch Institute that are statistically analyzed on JSC’s JUSUF computing cluster using a new, probability-weighted model developed by the Osnabrück neuroinformaticians.

The "COVID-19 Bayesian Modelling for Outbreak Detection", or BSTI model for short, has two essential features that distinguish it from other methods. Firstly, the new method provides a prediction horizon that makes it possible to assess the reliability of the forecasts. Furthermore, it takes into account the influence of the locally adjacent occurrence of infection. This also allows us to evaluate the dynamics of the spread. Furthermore, it is also possible to assess the situation while accounting for statistical uncertainties, which can provide helpful insights even with low case numbers.

Visitors to the website https://covid19-bayesian.fz-juelich.de can interactively view 5-day forecasts for various available districts or compare the current reporting data from the Robert Koch Institute with estimations of actual new infections. Due to delays in data transmission, the reported figures sometimes differ significantly from the actual number of new cases. A so-called "nowcast" aims to first estimate the current figures using statistical analyses. A "forecast" provides an estimate of the development for another five days.

This is one of the projects that has accepted JSC's offer of computing time and support to help advance coronavirus-related research.

Contact: Jens Henrik Göbbert, j.goebbert@fz-juelich.de

from JSC News No. 276, 31 October 2020