# American Institute of Mathematical Sciences

• Previous Article
A study of disproportionately affected populations by race/ethnicity during the SARS-CoV-2 pandemic using multi-population SEIR modeling and ensemble data assimilation
• FoDS Home
• This Issue
• Next Article
Ensemble Kalman Inversion for nonlinear problems: Weights, consistency, and variance bounds
September  2021, 3(3): 413-477. doi: 10.3934/fods.2021001

## An international initiative of predicting the SARS-CoV-2 pandemic using ensemble data assimilation

 1 NORCE and NERSC, Bergen, Norway 2 Dept. of Meteorology University of Reading and NCEO, UK 3 CEREA, joint laboratory École des Ponts ParisTech and EDF R & D Université Paris-Est, Champs-sur-Marne, France 4 Mathematical Institute University of Utrecht, Netherlands 5 Environment and Climate Change Canada Dorval, Québec, Canada 6 Renaissance Computing Institute University of North Carolina, Chapel Hill, USA 7 Department of Geoscience and Engineering Delft University of Technology, Delft, Netherlands 8 FaCENA, UNNE and IMIT, CONICET Corrientes, Argentina

*Corresponding author: Geir Evensen

Received  June 2020 Revised  November 2020 Published  September 2021 Early access  December 2020

Fund Project: The first author is supported by NORCE

This work demonstrates the efficiency of using iterative ensemble smoothers to estimate the parameters of an SEIR model. We have extended a standard SEIR model with age-classes and compartments of sick, hospitalized, and dead. The data conditioned on are the daily numbers of accumulated deaths and the number of hospitalized. Also, it is possible to condition the model on the number of cases obtained from testing. We start from a wide prior distribution for the model parameters; then, the ensemble conditioning leads to a posterior ensemble of estimated parameters yielding model predictions in close agreement with the observations. The updated ensemble of model simulations has predictive capabilities and include uncertainty estimates. In particular, we estimate the effective reproductive number as a function of time, and we can assess the impact of different intervention measures. By starting from the updated set of model parameters, we can make accurate short-term predictions of the epidemic development assuming knowledge of the future effective reproductive number. Also, the model system allows for the computation of long-term scenarios of the epidemic under different assumptions. We have applied the model system on data sets from several countries, i.e., the four European countries Norway, England, The Netherlands, and France; the province of Quebec in Canada; the South American countries Argentina and Brazil; and the four US states Alabama, North Carolina, California, and New York. These countries and states all have vastly different developments of the epidemic, and we could accurately model the SARS-CoV-2 outbreak in all of them. We realize that more complex models, e.g., with regional compartments, may be desirable, and we suggest that the approach used here should be applicable also for these models.

Citation: Geir Evensen, Javier Amezcua, Marc Bocquet, Alberto Carrassi, Alban Farchi, Alison Fowler, Pieter L. Houtekamer, Christopher K. Jones, Rafael J. de Moraes, Manuel Pulido, Christian Sampson, Femke C. Vossepoel. An international initiative of predicting the SARS-CoV-2 pandemic using ensemble data assimilation. Foundations of Data Science, 2021, 3 (3) : 413-477. doi: 10.3934/fods.2021001
##### References:

show all references

##### References:
Flow diagram of the SEIR model
Norway: This figure summarizes scenarios related to opening up kindergartens and schools on the 20th of April. The left plots show the ensemble means and the 100 first ensemble realizations, for the number of hospitalized and the accumulated amount of deaths for different scenarios of future $R(t) = 0.8,\, 1.0$, and $1.2$. The right plots show the prior and posterior ensembles of $R(t)$. The red thin line in the plots for $R(t)$ is an indication of the value $R(t) = 1$ for easier identification
Norway (base case): For the two cases (left and right plots), the only difference is the prior-guess for $R(t)$ after starting the interventions. For the first two rows of plots, we show the posterior ensemble means and the 100 first realizations of the posterior solution from ESMDA. The blue lines are the total number of cases, while the gray lines give the number of active cases. The red curves denote the number of hospitalized, and green lines show the total number of deaths. The upper plot uses a log $y$-axis. The second row is a zoom of the upper plot using a linear $y$-axis. The lower plots show the corresponding prior and posterior estimates of $R(t)$ for the two cases. The red thin line in the plots for $R(t)$ is an indication of the value $R(t) = 1$ for easier identification
England: The plots show the data available for assimilation. Three different agencies report the number of deaths, the UK government press conference publishes the number of people in hospitals, CHESS reports the daily hospital admissions, and PHE presents the number of new cases
England. Left panels: ESMDA posterior estimates of the accumulated number of deaths (green), the daily number of hospitalizations (red), active (gray), and total cases (blue). Observations are displayed in black. Right panels: Prior and ESMDA posterior estimates of the effective reproduction number $R(t)$. Ensemble members (thin lines) and ensemble means (thick lines). Assimilation experiment D (top panels), DH (mid panels) and DHC (bottom panels). The red thin line in the plots for $R(t)$ is an indication of the value $R(t) = 1$ for easier identification
Québec: From top to bottom, the plots show the results from the three assimilation experiments DHC, DH, and D. The left column presents the accumulated number of deaths and the number of hospitalizations, and the right column shows corresponding the reproductive number $R(t)$ for the experiments. Time is since the start of the epidemic on March 8th. Observations are indicated with points when used to obtain the model fit. The solid lines are for the ensemble mean posterior estimates. After May 28th, we kept the realizations of $R(t)$ constant and equal to the latest values for the remainder of the simulation. The red thin line in the plots for $R(t)$ is an indication of the value $R(t) = 1$ for easier identification
Québec: The plots show verification of retroactive week-two forecasts for experiments DHC, DH, and D. For the predictions, issued one week apart, we show the mean estimate with the full line, and the dashed lines give the mean value plus or minus one standard deviation. We indicate the reported values with crosses
Québec: The plots present verification of the week-two forecasts (retroactively) issued on April 1st for experiment D (left) and the experiment DHC (right). The solid line denotes the mean prediction. The reported values used to fit the model parameters are indicated with circles, while the triangles are the values used for verification
The Netherlands: The plots show the results from the Cases 1I, 1H, 2H, and 3H, from top to bottom. The left plots include model estimates of the number of hospitalized patients and dead, in addition to the total number of cases as well as the number of active cases. The right plots show the corresponding estimates of $R(t)$. The red thin line in the plots for $R(t)$ is an indication of the value $R(t) = 1$ for easier identification
France: The plot shows official data curves for France from Santé Publique France up to May, 31
France: The figure shows the reference case (left) and case with an unknown intervention (right). The upper plots show the number of deaths at hospital, hospitalized patients, the total number of cases, and the currently active cases. The lower plots show the ensemble of effective reproduction number $R(t)$. The red thin line in the plots for $R(t)$ is an indication of the value $R(t) = 1$ for easier identification
France: The plot presents forecasts of the reference case with three distinct scenarios after intervention setting the reproduction number to $R_3 = 0.75,\, 0.85,\, 1.00$. We have plotted the posterior values for the number of deaths at hospitals (green lines), and the hospitalizations (red lines), together with the first hundred realizations for each case
Brazil: The plots illustrate the evolution of SARSCoV-2 in terms of the number of confirmed cases, deaths, and mortality rate (CFR), starting from the day of the first reported death due to COVID-19. The reported states are São Paulo (SP), from the Southeast geopolitical region; Bahia (BA), from the Northeast; Para (PA), North; Rio Grande do Sul (RS), South; and Goiás (GO), Mid West. Each state represents a different evolution of the disease. We have shifted the curves in time to correspond to the first confirmed case.
Brazil: The figure shows the simulation of the SARS-CoV-2 evolution in the {São} Paulo Brazilian State. In the left plots, we show a neutral case where the reproductive number $R(t)\sim 1.0$ with a standard deviation of $0.1$ for the prediction. The right plots present a stable situation where $R(t)\sim 0.6$ with a standard deviation of 0.06, after a second intervention. The red thin line in the plots for $R(t)$ is an indication of the value $R(t) = 1$ for easier identification
Brazil: The plots compare simulations considering relative observation errors of 5% and 20%. The upper and lower rows represent a neutral scenario with $R_3 = 1.0$. They differ in the standard deviations or $R(t)$ (0.1 and 0.2). The middle plots show a stable situation with $R_3 = 0.6$ and a standard deviation of 0.06, following a second assumed intervention. The left plots show the deaths, and to the right, we present $R(t)$ using different measurement errors
Argentina: The dots are the observations. In Case DC (left plots) both the accumulated deaths and the estimated number of cases were conditioned on, while in Case D (right plots) we only conditioned on the total number of deaths. The red thin line in the plots for $R(t)$ is an indication of the value $R(t) = 1$ for easier identification
Argentina: Same as Figure 16 but for an experiment (Exp. 2) focused on a probabilistic prediction in which three scenarios, with different effective reproductive numbers imposed from June 1st, $R(t) = 1.7$, 1.3, and 0.9. The red thin line in the plots for $R(t)$ is an indication of the value $R(t) = 1$ for easier identification
US: Mobilites for each of the states considered
US Case 1: Large uncertainty in $R(t)$ The red thin line in the plots for $R(t)$ is an indication of the value $R(t) = 1$ for easier identification
US Case 2: Assuming voluntary federal guidance was immediately observed by the citizens on March 16th. The red thin line in the plots for $R(t)$ is an indication of the value $R(t) = 1$ for easier identification
US Case 3: A gradual step down in $R(t)$ with the first and intermediate values chosen so that the prior mean closely follows the data until the time for which $R(t)$ is guessed to be one. The red thin line in the plots for $R(t)$ is an indication of the value $R(t) = 1$ for easier identification
US: Forecasts for Case 3: Here we also show the prior forecast mean to highlight the difference between the idealized scenario and after analysis. In this case, only New York was able to achieve a trajectory that predicts fewer deaths than the idealized scenario. Furthermore, New York also is the only state in this study to make an end of the outbreak by late August
The table gives a set of first-guess model parameters. As we could not find scientific estimates of these parameters, we set their values based on available information from the internet and initial model-tuning experiments. We leave it to the data assimilation system to fine-tune the parameter values
 Parameter First guess Description $\tau_ \rm{inc}$ 5.5 Incubation period $\tau_ \rm{inf}$ 3.8 Infection time $\tau_ \rm{recm}$ 14.0 Recovery time mild cases $\tau_ \rm{recs}$ 5.0 Recovery time severe cases $\tau_ \rm{hosp}$ 6.0 Time until hospitalization $\tau_ \rm{death}$ 16.0 Time until death $p_ \mathrm{f}$ 0.009 Case fatality rate $p_ \mathrm{s}$ 0.039 Hospitalization rate (severe cases) $p_ \mathrm{h}$ 0.4 Fraction of fatally ill going to hospital
 Parameter First guess Description $\tau_ \rm{inc}$ 5.5 Incubation period $\tau_ \rm{inf}$ 3.8 Infection time $\tau_ \rm{recm}$ 14.0 Recovery time mild cases $\tau_ \rm{recs}$ 5.0 Recovery time severe cases $\tau_ \rm{hosp}$ 6.0 Time until hospitalization $\tau_ \rm{death}$ 16.0 Time until death $p_ \mathrm{f}$ 0.009 Case fatality rate $p_ \mathrm{s}$ 0.039 Hospitalization rate (severe cases) $p_ \mathrm{h}$ 0.4 Fraction of fatally ill going to hospital
The $p$ numbers indicate the fraction of sick people in an age group ending up with mild symptoms, severe symptoms (hospitalized), and fatal infection. The population-weighted averages (for the Norwegian population) of the case-fatality rate is 0.0090, and the rate of severe (hospitalized) cases is 0.039
 Age group 1 2 3 4 5 6 7 8 9 10 11 Age range 0–5 6–12 13–19 20–29 30–39 40–49 50–59 60–69 70–79 80–89 90–105 Population 351159 451246 446344 711752 730547 723663 703830 582495 435834 185480 45230 p–mild 1.0000 1.0000 0.9998 0.9913 0.9759 0.9686 0.9369 0.9008 0.8465 0.8183 0.8183 p–severe 0.0000 0.0000 0.0002 0.0078 0.0232 0.0295 0.0570 0.0823 0.1160 0.1160 0.1160 p–fatal 0.0000 0.0000 0.0000 0.0009 0.0009 0.0019 0.0061 0.0169 0.0375 0.0656 0.0656
 Age group 1 2 3 4 5 6 7 8 9 10 11 Age range 0–5 6–12 13–19 20–29 30–39 40–49 50–59 60–69 70–79 80–89 90–105 Population 351159 451246 446344 711752 730547 723663 703830 582495 435834 185480 45230 p–mild 1.0000 1.0000 0.9998 0.9913 0.9759 0.9686 0.9369 0.9008 0.8465 0.8183 0.8183 p–severe 0.0000 0.0000 0.0002 0.0078 0.0232 0.0295 0.0570 0.0823 0.1160 0.1160 0.1160 p–fatal 0.0000 0.0000 0.0000 0.0009 0.0009 0.0019 0.0061 0.0169 0.0375 0.0656 0.0656
Norway: This $\hat{ {\bf{R}}}$-matrix increases transmissions among children after opening kindergartens and schools on April 20th. We chose the numbers ad-hoc to give a qualitative impact of opening kindergartens and schools. To estimate these transmissions' correct values, we will need access to additional data that are not yet available
 Age groups 1 2 3 4 5 6 7 8 9 10 11 1 $\bf 3.3$ 1.8 1.8 1.3 1.3 1.0 0.9 0.9 0.9 0.9 0.9 2 1.8 $\bf 3.3$ 1.8 1.3 1.3 1.3 0.9 0.9 0.9 0.9 0.9 3 1.8 1.8 $\bf 0.9$ 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 4 1.3 1.3 0.9 $\bf 0.9$ 0.9 0.9 0.9 0.9 0.9 0.9 0.9 5 1.3 1.3 0.9 0.9 $\bf 0.9$ 0.9 0.9 0.9 0.9 0.9 0.9 6 1.0 1.3 0.9 0.9 0.9 $\bf 0.9$ 0.9 0.9 0.9 0.9 0.9 7 0.9 0.9 0.9 0.9 0.9 0.9 $\bf 0.9$ 0.9 0.9 0.9 0.9 8 0.9 0.9 0.9 0.9 0.9 0.9 0.9 $\bf 0.9$ 0.9 0.9 0.9 9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 $\bf 0.9$ 0.9 0.9 10 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 $\bf 0.9$ 0.9 11 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 $\bf 0.9$
 Age groups 1 2 3 4 5 6 7 8 9 10 11 1 $\bf 3.3$ 1.8 1.8 1.3 1.3 1.0 0.9 0.9 0.9 0.9 0.9 2 1.8 $\bf 3.3$ 1.8 1.3 1.3 1.3 0.9 0.9 0.9 0.9 0.9 3 1.8 1.8 $\bf 0.9$ 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 4 1.3 1.3 0.9 $\bf 0.9$ 0.9 0.9 0.9 0.9 0.9 0.9 0.9 5 1.3 1.3 0.9 0.9 $\bf 0.9$ 0.9 0.9 0.9 0.9 0.9 0.9 6 1.0 1.3 0.9 0.9 0.9 $\bf 0.9$ 0.9 0.9 0.9 0.9 0.9 7 0.9 0.9 0.9 0.9 0.9 0.9 $\bf 0.9$ 0.9 0.9 0.9 0.9 8 0.9 0.9 0.9 0.9 0.9 0.9 0.9 $\bf 0.9$ 0.9 0.9 0.9 9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 $\bf 0.9$ 0.9 0.9 10 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 $\bf 0.9$ 0.9 11 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 $\bf 0.9$
England: The contact matrix $\hat{ {\bf{R}}}_1$ used to describe the transmission between different age groups in England before the enforced lockdown on March 23rd. The same contact matrix is used for the prediction from June 1st. See the right panel of Figure 2A in [42] for a heat-map representation of the original matrix
 Age groups 1 2 3 4 5 6 7 8 9 10 11 1 $\bf 2.0$ 1.5 1.5 1.0 1.5 0.5 0.5 0.5 0.4 0.4 0.4 2 0.5 $\bf 8.0$ 6.0 2.0 2.5 2.5 1.5 1.4 0.9 0.9 0.9 3 0.5 6.0 $\bf 8.0$ 2.0 2.5 2.5 1.5 1.4 0.9 0.9 0.9 4 0.5 2.5 2.5 $\bf 6.0$ 2.0 2.0 1.9 1.5 0.9 0.9 0.9 5 1.2 2.5 2.5 2.0 $\bf 3.0$ 2.0 1.9 1.8 0.5 0.5 0.5 6 0.5 2.3 2.3 2.0 2.0 $\bf 3.0$ 1.9 1.5 1.4 1.4 1.4 7 0.5 2.0 2.0 1.5 1.5 1.5 $\bf 2.0$ 1.5 0.9 0.9 0.9 8 0.5 1.9 1.9 1.0 1.2 1.2 1.9 $\bf 1.5$ 0.9 0.9 0.9 9 0.5 1.5 1.5 0.9 0.9 1.2 1.0 1.5 $\bf 1.5$ 1.5 1.5 10 0.4 1.0 1.0 0.9 0.7 1.2 1.0 1.0 1.5 $\bf 1.5$ 1.5 11 0.4 0.9 0.9 0.9 0.7 1.2 1.0 1.0 1.5 1.5 $\bf 1.5$
 Age groups 1 2 3 4 5 6 7 8 9 10 11 1 $\bf 2.0$ 1.5 1.5 1.0 1.5 0.5 0.5 0.5 0.4 0.4 0.4 2 0.5 $\bf 8.0$ 6.0 2.0 2.5 2.5 1.5 1.4 0.9 0.9 0.9 3 0.5 6.0 $\bf 8.0$ 2.0 2.5 2.5 1.5 1.4 0.9 0.9 0.9 4 0.5 2.5 2.5 $\bf 6.0$ 2.0 2.0 1.9 1.5 0.9 0.9 0.9 5 1.2 2.5 2.5 2.0 $\bf 3.0$ 2.0 1.9 1.8 0.5 0.5 0.5 6 0.5 2.3 2.3 2.0 2.0 $\bf 3.0$ 1.9 1.5 1.4 1.4 1.4 7 0.5 2.0 2.0 1.5 1.5 1.5 $\bf 2.0$ 1.5 0.9 0.9 0.9 8 0.5 1.9 1.9 1.0 1.2 1.2 1.9 $\bf 1.5$ 0.9 0.9 0.9 9 0.5 1.5 1.5 0.9 0.9 1.2 1.0 1.5 $\bf 1.5$ 1.5 1.5 10 0.4 1.0 1.0 0.9 0.7 1.2 1.0 1.0 1.5 $\bf 1.5$ 1.5 11 0.4 0.9 0.9 0.9 0.7 1.2 1.0 1.0 1.5 1.5 $\bf 1.5$
England: The contact matrix $\hat{ {\bf{R}}}_2$ used to describe the transmission between different age groups in England during the lockdown from March 23rd to May 31st. See left hand panel of Figurek__ge 2A in [42] for a heat-map representation
 Age groups 1 2 3 4 5 6 7 8 9 10 11 1 $\bf 1.0$ 0.9 0.9 0.8 1.0 0.5 0.5 0.4 0.3 0.3 0.3 2 0.5 $\bf 2.0$ 1.5 0.9 1.0 1.0 0.5 0.4 0.3 0.3 0.3 3 0.5 1.5 $\bf 2.0$ 0.9 1.0 1.0 0.5 0.4 0.3 0.3 0.3 4 0.5 1.0 1.0 $\bf 1.2$ 1.0 1.0 0.9 0.5 0.4 0.3 0.3 5 0.8 1.0 1.0 0.9 $\bf 1.1$ 0.9 0.9 0.5 0.4 0.3 0.3 6 0.5 1.0 1.0 1.0 1.0 $\bf 1.1$ 0.9 0.5 0.4 0.3 0.3 7 0.5 0.6 0.6 0.9 0.9 0.9 $\bf 1.0$ 0.7 0.5 0.5 0.5 8 0.5 0.6 0.6 0.8 0.9 1.0 1.0 $\bf 1.0$ 0.5 0.5 0.5 9 0.5 0.6 0.6 0.6 0.5 1.0 0.9 0.9 $\bf 1.1$ 1.1 1.1 10 0.5 0.6 0.6 0.6 0.5 1.0 0.9 0.9 1.1 $\bf 1.1$ 1.1 11 0.5 0.6 0.6 0.6 0.5 1.0 0.9 0.9 1.1 1.1 $\bf 1.1$
 Age groups 1 2 3 4 5 6 7 8 9 10 11 1 $\bf 1.0$ 0.9 0.9 0.8 1.0 0.5 0.5 0.4 0.3 0.3 0.3 2 0.5 $\bf 2.0$ 1.5 0.9 1.0 1.0 0.5 0.4 0.3 0.3 0.3 3 0.5 1.5 $\bf 2.0$ 0.9 1.0 1.0 0.5 0.4 0.3 0.3 0.3 4 0.5 1.0 1.0 $\bf 1.2$ 1.0 1.0 0.9 0.5 0.4 0.3 0.3 5 0.8 1.0 1.0 0.9 $\bf 1.1$ 0.9 0.9 0.5 0.4 0.3 0.3 6 0.5 1.0 1.0 1.0 1.0 $\bf 1.1$ 0.9 0.5 0.4 0.3 0.3 7 0.5 0.6 0.6 0.9 0.9 0.9 $\bf 1.0$ 0.7 0.5 0.5 0.5 8 0.5 0.6 0.6 0.8 0.9 1.0 1.0 $\bf 1.0$ 0.5 0.5 0.5 9 0.5 0.6 0.6 0.6 0.5 1.0 0.9 0.9 $\bf 1.1$ 1.1 1.1 10 0.5 0.6 0.6 0.6 0.5 1.0 0.9 0.9 1.1 $\bf 1.1$ 1.1 11 0.5 0.6 0.6 0.6 0.5 1.0 0.9 0.9 1.1 1.1 $\bf 1.1$
England: Prior and posterior mean and standard deviation for the time independent parameters estimated with ESDMA in the experiments D, DH and DHC
 Parameters Prior Posterior D Posterior DH Posterior DHC $I_0$ 59.97 (6.06) 61.32 (6.01) 61.95 (6.01) 60.08 (5.98) $E_0$ 240.64 (24.17) 246.68 (23.85) 251.56 (23.62) 239.43 (23.46) $\tau_ \rm{inf}$ 3.80 (0.50) 2.73 (0.33) 3.13 (0.33) 2.83 (0.28) $\tau_ \rm{inc}$ 5.50 (0.50) 4.62 (0.40) 4.79 (0.40) 5.14 (0.33) $\tau_ \rm{recm}$ 13.98 (0.49) 13.99 (0.49) 13.94 (0.49) 13.94 (0.49) $\tau_ \rm{recs}$ 4.99 (0.41) 4.98 (0.40) 4.13 (0.31) 3.57 (0.31) $\tau_ \rm{hosp}$ 5.99 (0.51) 5.54 (0.49) 5.35 (0.48) 4.79 (0.39) $\tau_ \rm{death}$ 15.99 (0.50) 15.64 (0.50) 15.13 (0.47) 14.43 (0.44) $p_ \mathrm{f}$ 0.009 (0.001) 0.009 (0.001) 0.014 (0.0009) 0.014 (0.0002) $p_ \mathrm{s}$ 0.039 (0.004) 0.039 (0.003) 0.011 (0.002) 0.015 (0.002)
 Parameters Prior Posterior D Posterior DH Posterior DHC $I_0$ 59.97 (6.06) 61.32 (6.01) 61.95 (6.01) 60.08 (5.98) $E_0$ 240.64 (24.17) 246.68 (23.85) 251.56 (23.62) 239.43 (23.46) $\tau_ \rm{inf}$ 3.80 (0.50) 2.73 (0.33) 3.13 (0.33) 2.83 (0.28) $\tau_ \rm{inc}$ 5.50 (0.50) 4.62 (0.40) 4.79 (0.40) 5.14 (0.33) $\tau_ \rm{recm}$ 13.98 (0.49) 13.99 (0.49) 13.94 (0.49) 13.94 (0.49) $\tau_ \rm{recs}$ 4.99 (0.41) 4.98 (0.40) 4.13 (0.31) 3.57 (0.31) $\tau_ \rm{hosp}$ 5.99 (0.51) 5.54 (0.49) 5.35 (0.48) 4.79 (0.39) $\tau_ \rm{death}$ 15.99 (0.50) 15.64 (0.50) 15.13 (0.47) 14.43 (0.44) $p_ \mathrm{f}$ 0.009 (0.001) 0.009 (0.001) 0.014 (0.0009) 0.014 (0.0002) $p_ \mathrm{s}$ 0.039 (0.004) 0.039 (0.003) 0.011 (0.002) 0.015 (0.002)
Québec: The set of prior model parameters and their standard deviations (a zero std dev denotes that the parameter is kept fixed). Columns DHC, DH and D show posterior values for, respectively, experiments DHC, DH and D. Note that $p_{h}$ was supplied externally. The curves for $R_1$ and $R_2$ are shown in Figure 6
 Parameters Prior(Std Dev) DHC DH D $R_1$ 3.0(0.6) - - - $R_2$ 1.0(0.5) - - - $I_0$ 100.0(20.0) 67 98 96 $E_0$ 240.0(48.0) 167 204 235 $\tau_ \rm{inc}$ 5.5(1.0) 5.2 3.4 3.8 $\tau_ \rm{inf}$ 3.8(0.6) 1.8 1.9 2.7 $\tau_ \rm{recm}$ 14.0(2.0) 14.1 12.8 14.8 $\tau_ \rm{recs}$ 5.0(1.0) 6.9 6.8 5.5 $\tau_ \rm{hosp}$ 6.0(1.2) 5.9 5.8 6.7 $\tau_ \rm{death}$ 10.0(2.0) 5.8 3.4 10.4 $p_ \mathrm{f}$ 0.020(0.004) 0.020 0.021 0.023 $p_ \mathrm{s}$ 0.039(0.006) 0.040 0.047 0.038 $p_{h}$ 0.5(0) - - -
 Parameters Prior(Std Dev) DHC DH D $R_1$ 3.0(0.6) - - - $R_2$ 1.0(0.5) - - - $I_0$ 100.0(20.0) 67 98 96 $E_0$ 240.0(48.0) 167 204 235 $\tau_ \rm{inc}$ 5.5(1.0) 5.2 3.4 3.8 $\tau_ \rm{inf}$ 3.8(0.6) 1.8 1.9 2.7 $\tau_ \rm{recm}$ 14.0(2.0) 14.1 12.8 14.8 $\tau_ \rm{recs}$ 5.0(1.0) 6.9 6.8 5.5 $\tau_ \rm{hosp}$ 6.0(1.2) 5.9 5.8 6.7 $\tau_ \rm{death}$ 10.0(2.0) 5.8 3.4 10.4 $p_ \mathrm{f}$ 0.020(0.004) 0.020 0.021 0.023 $p_ \mathrm{s}$ 0.039(0.006) 0.040 0.047 0.038 $p_{h}$ 0.5(0) - - -
The Netherlands: The table gives values of the parameters used in Case 1, for the parameters that are different from those indicated in Table 1. The starting date of the simulations is February 20th, 2020
 Parameter First guess Std. Dev. Description $E_0$ 500.0 50.0 Initially exposed $I_0$ 400.0 40.0 Initially infectious $R_1$ 3.8 0.05 Reproduction number before interventions (Case 1DH and Case 1DI) $R_1$ 0.8 0.01 Reproduction number after first nation-wide intervention (Case 1DH and Case 1DI) $R_1$ 1.0 0.75 Reproduction number (Case 2DH) $p_ \mathrm{s}$ 0.010 0.0001 Hospitalization rate (Case 1DI) $p_ \mathrm{s}$ 0.039 0.0039 Hospitalization rate (Case 1DH, Case 2DH and Case 3DH) $p_ \mathrm{h}$ 0.5 Fraction of fatally ill going to hospital (Case 1DI) $p_ \mathrm{h}$ 0.6 Fraction of fatally ill going to hospital (Case 1DH, Case 2DH and Case 3DH)
 Parameter First guess Std. Dev. Description $E_0$ 500.0 50.0 Initially exposed $I_0$ 400.0 40.0 Initially infectious $R_1$ 3.8 0.05 Reproduction number before interventions (Case 1DH and Case 1DI) $R_1$ 0.8 0.01 Reproduction number after first nation-wide intervention (Case 1DH and Case 1DI) $R_1$ 1.0 0.75 Reproduction number (Case 2DH) $p_ \mathrm{s}$ 0.010 0.0001 Hospitalization rate (Case 1DI) $p_ \mathrm{s}$ 0.039 0.0039 Hospitalization rate (Case 1DH, Case 2DH and Case 3DH) $p_ \mathrm{h}$ 0.5 Fraction of fatally ill going to hospital (Case 1DI) $p_ \mathrm{h}$ 0.6 Fraction of fatally ill going to hospital (Case 1DH, Case 2DH and Case 3DH)
The Netherlands: overview of cases
 Case Assimilated data Description 1DI deaths, ICU patients prior $R(t)$ equals 3.8 before intervention, 0.8 after 1DH deaths, hospitalized prior $R(t)$ equals 3.8 before intervention, 0.8 after 2DH deaths, hospitalized prior $R(t)$ equals 1.0 3DH deaths, hospitalized prior $R(t)$ equals 1.8 at start of simulation and gradually ramps down to 0.8
 Case Assimilated data Description 1DI deaths, ICU patients prior $R(t)$ equals 3.8 before intervention, 0.8 after 1DH deaths, hospitalized prior $R(t)$ equals 3.8 before intervention, 0.8 after 2DH deaths, hospitalized prior $R(t)$ equals 1.0 3DH deaths, hospitalized prior $R(t)$ equals 1.8 at start of simulation and gradually ramps down to 0.8
France: The table gives a set of "calibrated" first-guess (i.e., prior) model parameters and their standard deviations used for France. $p_{\mathrm h}$ is set ot $1$ to inform the model that care homes deaths are excluded from the death numbers. All other parameter settings are unchanged as compared to the ones given in Table 1
 Parameter First guess Std. Dev. Description $t_0$ February 16th - Start date of simulation $t_1$ March 17th - Start date of intervention $t_2$ May 11th - End of lockdown $R_1$ 3.5 0.20 $R(t)$ prior before intervention $R_2$ 0.65 0.20 $R(t)$ prior during lockdown $R_3$ 0.85 0.20 $R(t)$ prior after full lockdown $E_0$ 500 500 Initial Exposed $I_0$ 200 200 Initial Infectious $\tau_ \rm{recs}$ 20 2 Recovery time severe cases $\tau_ \rm{hosp}$ 6 0.5 Time until hospitalization $\tau_ \rm{death}$ 7 1 Time until death $p_ \mathrm{f}$ 0.02 0.02 Case fatality rate $p_ \mathrm{s}$ 0.039 0.03 Hospitalization rate for severe cases $p_{\mathrm h}$ $1$ - Fraction of ${\bf Q}_\mathrm{f}$ that go to hospital
 Parameter First guess Std. Dev. Description $t_0$ February 16th - Start date of simulation $t_1$ March 17th - Start date of intervention $t_2$ May 11th - End of lockdown $R_1$ 3.5 0.20 $R(t)$ prior before intervention $R_2$ 0.65 0.20 $R(t)$ prior during lockdown $R_3$ 0.85 0.20 $R(t)$ prior after full lockdown $E_0$ 500 500 Initial Exposed $I_0$ 200 200 Initial Infectious $\tau_ \rm{recs}$ 20 2 Recovery time severe cases $\tau_ \rm{hosp}$ 6 0.5 Time until hospitalization $\tau_ \rm{death}$ 7 1 Time until death $p_ \mathrm{f}$ 0.02 0.02 Case fatality rate $p_ \mathrm{s}$ 0.039 0.03 Hospitalization rate for severe cases $p_{\mathrm h}$ $1$ - Fraction of ${\bf Q}_\mathrm{f}$ that go to hospital
Brazil: The table gives the parameter values used in the SARS-CoV-2 simulations for São Paulo, Brazil. All other parameters were kept unchanged as compared to the ones given in Table \protect1
 Parameter Initial value Std dev $t_0$ March 10th - Start date of simulation $t_1$ March 23rd - Start date of interventions $t_2$ June 1st - Start date of prediction $E_0$ 656.0 65.6 Initial Exposed $I_0$ 164.0 16.4 Initial Infectious $R_1$ 4.0 0.4 Prior $R(t)$ during spinup $R_2$ 1.0 0.1 Prior $R(t)$ during interventions $R_3$ 1.0 0.1 Prior $R(t)$ in prediction phase $p_ \mathrm{h}$ 0.2 - Ratio of fatally sick hospitalised $p_ \mathrm{f}$ 0.065 0.0065 Case fatality rate (CFR)
 Parameter Initial value Std dev $t_0$ March 10th - Start date of simulation $t_1$ March 23rd - Start date of interventions $t_2$ June 1st - Start date of prediction $E_0$ 656.0 65.6 Initial Exposed $I_0$ 164.0 16.4 Initial Infectious $R_1$ 4.0 0.4 Prior $R(t)$ during spinup $R_2$ 1.0 0.1 Prior $R(t)$ during interventions $R_3$ 1.0 0.1 Prior $R(t)$ in prediction phase $p_ \mathrm{h}$ 0.2 - Ratio of fatally sick hospitalised $p_ \mathrm{f}$ 0.065 0.0065 Case fatality rate (CFR)
US: The parameters used in our experiments, the different values for the reproductive number at intervention steps are described in the sections for each case. Values without state indications are the same for all states. Any parameters not listed are the same as in Tab 1
 Param Prior Std. Dev. Description $T_s$ 20/2-2020 - Start Date $E_0$ 50(NY, CA), 10(AL), 20(NC) 10(NY, CA, NC), 2.0(AL) Initial Exposed $I_0$ 10(NY, CA), 1(AL), 2(NC) 5(NY, CA, NC), 1.0(AL) Initial infectious $p_ \mathrm{f}$ 0.18(NY), 0.009(CA, NC, AL) 0.001 CFR
 Param Prior Std. Dev. Description $T_s$ 20/2-2020 - Start Date $E_0$ 50(NY, CA), 10(AL), 20(NC) 10(NY, CA, NC), 2.0(AL) Initial Exposed $I_0$ 10(NY, CA), 1(AL), 2(NC) 5(NY, CA, NC), 1.0(AL) Initial infectious $p_ \mathrm{f}$ 0.18(NY), 0.009(CA, NC, AL) 0.001 CFR
 [1] Emmanuel Fleurantin, Christian Sampson, Daniel Paul Maes, Justin Bennett, Tayler Fernandes-Nunez, Sophia Marx, Geir Evensen. A study of disproportionately affected populations by race/ethnicity during the SARS-CoV-2 pandemic using multi-population SEIR modeling and ensemble data assimilation. Foundations of Data Science, 2021, 3 (3) : 479-541. doi: 10.3934/fods.2021022 [2] Jiangqi Wu, Linjie Wen, Jinglai Li. Resampled ensemble Kalman inversion for Bayesian parameter estimation with sequential data. Discrete & Continuous Dynamical Systems - S, 2021  doi: 10.3934/dcdss.2021045 [3] Azmy S. Ackleh, Jeremy J. Thibodeaux. Parameter estimation in a structured erythropoiesis model. Mathematical Biosciences & Engineering, 2008, 5 (4) : 601-616. doi: 10.3934/mbe.2008.5.601 [4] Débora A. F. Albanez, Maicon J. Benvenutti. Continuous data assimilation algorithm for simplified Bardina model. Evolution Equations & Control Theory, 2018, 7 (1) : 33-52. doi: 10.3934/eect.2018002 [5] Joshua Hudson, Michael Jolly. Numerical efficacy study of data assimilation for the 2D magnetohydrodynamic equations. Journal of Computational Dynamics, 2019, 6 (1) : 131-145. doi: 10.3934/jcd.2019006 [6] Alex Capaldi, Samuel Behrend, Benjamin Berman, Jason Smith, Justin Wright, Alun L. Lloyd. Parameter estimation and uncertainty quantification for an epidemic model. Mathematical Biosciences & Engineering, 2012, 9 (3) : 553-576. doi: 10.3934/mbe.2012.9.553 [7] John Maclean, Elaine T. Spiller. A surrogate-based approach to nonlinear, non-Gaussian joint state-parameter data assimilation. Foundations of Data Science, 2021, 3 (3) : 589-614. doi: 10.3934/fods.2021019 [8] Gianni Gilioli, Sara Pasquali, Fabrizio Ruggeri. Nonlinear functional response parameter estimation in a stochastic predator-prey model. Mathematical Biosciences & Engineering, 2012, 9 (1) : 75-96. doi: 10.3934/mbe.2012.9.75 [9] Matthew Gardner, Adam Larios, Leo G. Rebholz, Duygu Vargun, Camille Zerfas. Continuous data assimilation applied to a velocity-vorticity formulation of the 2D Navier-Stokes equations. Electronic Research Archive, 2021, 29 (3) : 2223-2247. doi: 10.3934/era.2020113 [10] Adam Larios, Yuan Pei. Approximate continuous data assimilation of the 2D Navier-Stokes equations via the Voigt-regularization with observable data. Evolution Equations & Control Theory, 2020, 9 (3) : 733-751. doi: 10.3934/eect.2020031 [11] Alexandre J. Chorin, Fei Lu, Robert N. Miller, Matthias Morzfeld, Xuemin Tu. Sampling, feasibility, and priors in data assimilation. Discrete & Continuous Dynamical Systems, 2016, 36 (8) : 4227-4246. doi: 10.3934/dcds.2016.36.4227 [12] Sebastian Springer, Heikki Haario, Vladimir Shemyakin, Leonid Kalachev, Denis Shchepakin. Robust parameter estimation of chaotic systems. Inverse Problems & Imaging, 2019, 13 (6) : 1189-1212. doi: 10.3934/ipi.2019053 [13] Jochen Bröcker. Existence and uniqueness for variational data assimilation in continuous time. Mathematical Control & Related Fields, 2021  doi: 10.3934/mcrf.2021050 [14] Simon Hubmer, Andreas Neubauer, Ronny Ramlau, Henning U. Voss. On the parameter estimation problem of magnetic resonance advection imaging. Inverse Problems & Imaging, 2018, 12 (1) : 175-204. doi: 10.3934/ipi.2018007 [15] Robert Azencott, Yutheeka Gadhyan. Accurate parameter estimation for coupled stochastic dynamics. Conference Publications, 2009, 2009 (Special) : 44-53. doi: 10.3934/proc.2009.2009.44 [16] Laurent Devineau, Pierre-Edouard Arrouy, Paul Bonnefoy, Alexandre Boumezoued. Fast calibration of the Libor market model with stochastic volatility and displaced diffusion. Journal of Industrial & Management Optimization, 2020, 16 (4) : 1699-1729. doi: 10.3934/jimo.2019025 [17] Issam S. Strub, Julie Percelay, Olli-Pekka Tossavainen, Alexandre M. Bayen. Comparison of two data assimilation algorithms for shallow water flows. Networks & Heterogeneous Media, 2009, 4 (2) : 409-430. doi: 10.3934/nhm.2009.4.409 [18] Yuan Pei. Continuous data assimilation for the 3D primitive equations of the ocean. Communications on Pure & Applied Analysis, 2019, 18 (2) : 643-661. doi: 10.3934/cpaa.2019032 [19] Juan Carlos De los Reyes, Estefanía Loayza-Romero. Total generalized variation regularization in data assimilation for Burgers' equation. Inverse Problems & Imaging, 2019, 13 (4) : 755-786. doi: 10.3934/ipi.2019035 [20] Robert G. McLeod, John F. Brewster, Abba B. Gumel, Dean A. Slonowsky. Sensitivity and uncertainty analyses for a SARS model with time-varying inputs and outputs. Mathematical Biosciences & Engineering, 2006, 3 (3) : 527-544. doi: 10.3934/mbe.2006.3.527

Impact Factor: