Analyzing the effects of observation function selection in ensemble Kalman filtering for epidemic models

Area of Science:

• Epidemiology
• Computational Science
• Data Assimilation

Background:

• The Ensemble Kalman Filter (EnKF) is widely applied in epidemiology for parameter estimation and forecasting.
• The observation function is a critical component linking model states to observed data in EnKF.
• Variations in data and modeling assumptions necessitate diverse observation functions in epidemic modeling.

Purpose of the Study:

• To computationally analyze the impact of observation function selection on EnKF performance for epidemiological models.
• To demonstrate how incorrect observation modeling affects state and parameter estimation accuracy.
• To highlight the importance of aligning observation functions with available data.

Main Methods:

• Utilized the Susceptible-Infectious-Recovered (SIR) model as the epidemiological framework.
• Implemented and evaluated four distinct, epidemiologically-inspired observation functions.
• Performed computational analysis for state estimation with known parameters and combined state-parameter estimation (constant and time-varying).

Main Results:

• Incorrect observation function choices (e.g., prevalence for incidence, ignoring under-reporting) result in inaccurate EnKF estimates and forecasts.
• The selection of an appropriate observation function significantly impacts the reliability of EnKF-derived epidemiological insights.
• Adjusting the observation noise covariance matrix can mitigate some uncertainties related to observation function misspecification.

Conclusions:

• Accurate state and parameter estimation in epidemiological studies using EnKF hinges on appropriate observation function selection.
• Modelers must carefully consider data characteristics and potential biases when choosing observation functions.
• Further research can explore advanced methods to handle observation function uncertainty within the EnKF framework.