A modified SEIR model to predict the COVID-19 outbreak in Spain: simulating control scenarios and multi-scale epidemics

Leonardo R Lopez; Xavier Rodo

Because the strength of this type of model, both to adjust and to predict future dynamics, depends on the accuracy in case reporting, it is important to analyze the uncertainties in the adjustment and evaluate the degree of variation of the parameter estimates. To get a better idea on its performance, the model was fitted with successively longer data series, adding a new day to the data window with which the fitting was computed. The shortest time series used was of 5 days from February 29 to March 3 and the longest covered up to March 31. The temporal span of the time series used can seriously affect the value of the fitted parameters. This is especially true at the beginning of the epidemic when the relationship between infected, dead and recovered is not entirely clear. For example if the number of confirmed cases is small, it is difficult to ascertain whether the quarantine has been rigorously applied or not. In addition, such under-representation would suggest that the number of infectious people is much larger than the number of confirmed cases. Results of the different dynamics and the parameter space can be seen in Figure 8 . Those parameters most affected are the ones directly related with the infection rate and the recovery rate, as well as the protection rate α.