Post-processing and weighted combination of infectious disease nowcasts.

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Infectious disease surveillance relies on incidence data, often plagued by reporting delays and corrections.
• Accurate real-time trend assessment is crucial but challenging due to data limitations.
• Probabilistic nowcasting methods aim to correct these biases.

Purpose of the Study:

• To enhance nowcasts using statistical post-processing techniques.
• To evaluate weighted ensemble nowcasts as an extension of unweighted ensembles.
• To address challenges in post-processing and ensemble building with revised data.

Main Methods:

• Applied statistical post-processing methods, similar to weather forecasting, to nowcasting models.
• Investigated weighted combinations of different probabilistic nowcasts (weighted ensembles).
• Utilized COVID-19 hospitalization data from Germany for model evaluation.

Main Results:

• Post-processing significantly improved individual model performance in scores and forecast interval coverage.
• Weighted ensemble methods showed modest score improvements for some approaches but decreased performance for most.
• Weighted ensembles consistently improved forecast interval coverage compared to unweighted ensembles.

Conclusions:

• Statistical post-processing is a valuable tool for improving individual nowcasting models.
• Weighted ensembles present challenges, with limited gains in accuracy but improved interval coverage.
• Findings offer insights for refining real-time infectious disease surveillance systems.