Challenges of Forecasting COVID-19 During the 2nd wave

Short-term forecasts of COVID-19 case numbers for the next 1–4 weeks can be a valuable tool for deciding which measures are necessary to prevent an oversubscription of hospital beds and ICUs, but how reliable are these predictions?

 

COVID-19 weekly incident cases, Germany, 1 wk ahead, November/December 2020COVID-19 weekly incident cases, Germany, 1 wk ahead, November/December 2020
Copyright: Bracher et al., Nature Communication, DOI: 10.1038/s41467-021-25207-0

In "A pre-registered short-term forecasting study of COVID-19 in Germany and Poland during the second wave" published in Nature Communication (DOI: 10.1038/s41467-021-25207-0), scientists from JSC together with colleagues from Europe and the USA analyse their predictions collected for Germany and Poland as part of the German–Polish forecasting hub (https://kitmetricslab.github.io/forecasthub/forecast) during the second wave of COVID-19 (12 October – 19 December 2020). In the study, individual forecast performance is compared to the quality of ensemble forecasts, in which the individual forecasts are combined. The confidence interval of the ensemble forecasts provides good coverage, i.e. the 50 % confidence interval covers about 50 % of the data. The predictions of the ensemble compare well against the best single-model prediction and seem less prone to outliers.

The team’s contributions to the German–Polish forecasting hub are now also included in the European Covid-19 Forecast Hub at https://covid19forecasthub.eu/.

Contact: Dr. Jan Meinke, Jülich Supercomputing Centre, j.meinke@fz-juelich.de

also in JSC News No. 283, 24 September 2021