An international group of geoscientists analyzed the evolution of the COVID-19 pandemic. As a forecasting model, a compartimental model of the SEIR type is used. The basic model has compartments for the susceptible (S), exposed (E), infected (I) and recovered (R) population. We add compartments Q, H and D for the population in quarantine (Q), hospital (H) or deceased (D) and we have compartments by age group. There are observations for the population with confirmed positive tests, which is in hospital or deceased.

Ensemble based assimilation methods use the observations to estimate parameters of the compartimental model for COVID-19 as well as the reproductive number and also the fractions of the exposed and infected population at the start of the simulation. Subsequently, the SEIR model can be used with these values to produce short-range forecasts with an estimate of the uncertainty.

The situation in Quebec is used to illustrate the situation and test the methodology. Each week of the epidemic, we do a retroactive forecast two weeks into the future. This gives us 10 cases for verification. We will discuss options to improve the quality of these epidemiological forecasts.

Pieter Houtekamer is a senior scientist at Environment and Climate Change Canada. After gaining his doctorate from Wageningen University in the Netherlands he moved to Montreal to develop a global medium-range ensemble prediction system for the Atmospheric Environment Service. Currently, he investigates if the experience with Numerical Weather Prediction can also serve towards ensemble prediction for the COVID-19 epidemic.

Reference for the use of ensemble methods in the atmospheric sciences: https://doi.org/10.1175/MWR-D-15-0440.1

Reference for COVID-19 work: https://www.medrxiv.org/content/10.1101/2020.06.11.20128777v1

Short professional information: https://profils-profiles.science.gc.ca/en/profile/pieter-l-houtekamer