Multi-Model Nowcasting and Short-Term Forecasting of Infectious Disease Spread

• Applicant: Professor Dr. Johannes Bracher
• Subject Area: Epidemiology and Medical Biometry/Statistics
• Term: since 2022
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 512483310

During outbreaks of infectious diseases, probabilistic short-term forecasts for horizons of several days or weeks contribute substantially to situational awareness. They complement so-called nowcasts, which serve to predict occurred but not yet reported events, thus correcting preliminary values of epidemiological indicators. In order to provide decision makers with reliable statements on expected trends and the remaining uncertainty, it is increasingly common to combine predictions from different models are into so-called ensemble forecasts. Compared e.g., to weather forecasting, however, this is complicated by more limited data availability and the occurrence of structural breaks, for instance following changes in control measures. Improving the quality of probabilistic forecasts is crucial for the further management of the COVID-19 pandemic, but also with respect to seasonal infectious diseases or future outbreaks of other pathogens. The proposed junior research group will address central methodological challenges concerning the combination, creation and statistical evaluation of disease forecasts and nowcasts. In order to improve combined ensemble forecasts, the team will develop novel methods to reflect complex dependency structures and to account for reporting delays. As the quality of ensemble forecasts hinges on their member models, the group will moreover be concerned with the improvement of two fundamental forecasting approaches. Firstly, new strategies to reflect contemporaneous correlations in spatio-temporal models of disease spread will be developed. This serves to improve regional-level forecasts. Secondly, the group will build up a system for human crowd forecasts, which compared to computational models offer advantages in their capacity to adapt quickly e.g. to changing control measures. Here, the methodological focus will be on statistical post-processing of human forecasts. To enhance the relevance of statistical forecast evaluations, my group will assess how performance measures can be adjusted to the priorities of forecast users from public health and society. Moreover, new methods will be developed in order to compare different forecasting approaches in settings where these have not been applied to the exact same set of prediction tasks. My team will apply the developed forecasting methods to several infectious diseases in real time and compare them to other approaches within international collaborative platforms. A special focus will be put on a consistent open science approach, with scientific transparency ensured via pre-registrations, public accessibility of forecasts, and open-source software.

• DFG Programme: Independent Junior Research Groups