Comparison of ensemble-based data assimilation techniques for epidemiological forecasting and parameter estimation in an age-based compartmental SEIR model

JUAN RUIZ; PULIDO, MANUEL; SANTIAGO ROSA; TADEO JAVIER COCUCCI

Bergen

Workshop; EnKF workshop 2021; 2021

Ensemble-based data assimilation techniques are coupled with an age-based compartmental SEIR epidemiological model for the estimation of critical parameters like a time-dependent effective reproductive number, mortality rate, hospitalization rate, the subdetection rate, and the components of the contact matrix among different population groups. Three different techniques are evaluated: an iterative ensemble Kalman smoother, a lag-Kalman smoother, and the Kalman filter. Experiments are conducted to compare the performance of these techniques in estimating the value of the mentioned epidemiological parameters and in providing a short-range forecast of the evolution of the number of infections. Comparison is done assuming age-dependent observations of cases, death, and hospitalizations in the context of observation simulation experiments where the true parameter values are known. Different scenarios are considered including the presence of model errors and systematic observation errors. An experiment using age-structured observations in Argentina was conducted and will be also discussed.