Related Experiment Videos
Counteracting structural errors in ensemble forecast of influenza outbreaks
1Department of Environmental Health Sciences, Mailman School of Public Health, Columbia University, New York, NY, 10032, USA. sp3449@cumc.columbia.edu.
Nature Communications
|October 15, 2017
Summary
Improving influenza forecasts requires understanding nonlinear error growth. This study develops a new dynamical error correction method, significantly enhancing prediction accuracy for outbreaks up to 10 weeks ahead.
Area of Science:
- Epidemiology
- Computational modeling
- Data science
Background:
- Dynamical models for influenza forecasting face inaccuracies due to nonlinear error growth.
- Quantifying and correcting this error structure is crucial for improving forecast skill.
Purpose of the Study:
- To investigate the error growth structure in a compartmental influenza model.
- To develop an improved influenza forecasting approach by addressing nonlinear error dynamics.
Main Methods:
- Inspection of error growth within a compartmental influenza model.
- Utilizing error breeding to diagnose and counteract structural errors.
- Combining dynamical error correction with statistical filtering techniques.
Main Results:
- A robust error structure was identified, arising naturally from nonlinear model dynamics.
- The new forecast approach substantially improved accuracy for outbreak peak timing, intensity, and attack rate.
- Significant forecast skill enhancement was observed for lead times up to 10 weeks across 95 US cities (2003-2014).
Conclusions:
- The developed method effectively corrects nonlinear error growth in influenza models.
- This approach offers a generalizable strategy for enhancing the accuracy of various infectious disease dynamical models.
- Improved influenza forecasting can aid public health preparedness and response efforts.