metadata-USACE-ERDC_SEIR.txt

team_name: US Army Engineer Research and Development Center
model_name: ERDC_SEIR
model_abbr: USACE-ERDC_SEIR
model_contributors: Michael L. Mayo <Michael.L.Mayo@erdc.dren.mil>, Michael A. Rowland
  <Michael.A.Rowland@erdc.dren.mil>,  Matthew D. Parno <Matthew.D.Parno@erdc.dren.mil>,
  Ian D. Detwiller <Ian.D.Dettwiller@erdc.dren.mil>, Matthew W. Farthing <Matthew.W.Farthing@erdc.dren.mil>,
  William P. England <William.P.England@erdc.dren.mil>, Glover E. George <glover.e.george@erdc.dren.mil>
website_url: https://github.com/erdc-cv19/seir-model
license: other
team_model_designation: primary
ensemble_of_hub_models: false
methods: The ERDC SEIR model makes predictions of several variables (e.g., reported
  new/cumulative cases  per day, etc.). Model parameters are estimated using historical
  data using Bayesian inference.
team_funding: US Army Geospatial Task Force
data_inputs: USA FACTS, JHU COVID-19 Git Repo
methods_long: 'The ERDC SEIR model is a process-based model that mathematically describes
  the virus  dynamics in a population center (e.g., state, CBSA) using assumptions
  that are common  in compartmental models: (i)   modeled populations are large enough
  that fluctuations in the disease  states grow slower than averages (i.e., coefficient
  of variation < 1) (ii)  recovered individuals are neither infectious nor become
  susceptible to  further infection.

  The model is similar to classic SEIR models but with additional compartments for  unreported
  infections and isolated individuals.  Model parameters describing reporting  rates
  and transmission rates are allowed to vary in time over over the historical  record
  as social and policy factors change.  Forecasts into the future, however,  are made
  under the assumption that the parameters are held constant in time.

  The model parameters are estimated from observations of the cumulative number of  cases
  using a Bayesian approach. Information from subject matter experts (SMEs) is  used
  to develop a prior probability distribution over the model parameters.  This  then
  combined with a statistical model of the model-data mismatch to produce a posterior  probability
  distribution over the model parameters that incorporates both the cumulative  observations
  and subject matter expertise.  The parameters that maximize the posterior  probability
  density are used to make forecasts into the future.  The effective reproduction  number
  is computed using the next-generation matrix method.

  Deaths are estimated from this model through a procedure which leverages state-level  data
  on recoveries and fatalities.'