Introduction

What is the Flu MetroCast Hub?

The Flu MetroCast Hub brings together modeling teams from across the country to share weekly forecasts of flu activity in U.S. cities and metro areas. It is led by the epiENGAGE team at the University of Texas at Austin and the University of Massachusetts Amherst.

Why forecast flu at the metro level?

Predicting how many people will visit the emergency department for flu helps us see how the season is unfolding, how severe it may be, and how much pressure hospitals could face. While most forecasting efforts focus on states or the whole country, the Flu MetroCast Hub collects forecasts at the metro-level.

Looking at metro areas helps us:

• Spot local trends that state-level forecasts can miss, so decisions can be more accurate and timely.
• Get ready for the future by building the tools, data, and expertise needed to respond to future outbreaks.
• Share useful insights that public health officials, healthcare systems, and community leaders can put to work.

The 2026–27 forecasting season

From October 21, 2026 through May 19, 2027, participating teams will submit weekly forecasts of the percentage of emergency department visits due to influenza from the current week up to three weeks ahead. Teams can also share forecasts that project how the rest of the flu season may unfold.

View The Latest Forecasts

Current Metro Areas

Target: Percentage of emergency department visits due to influenza

• Health Service Areas (HSAs) across the U.S.
  • Data Source : CDC’s Emergency Department Visits for Viral Respiratory Illness
  • Target areas: 70 metropolitan HSAs across Colorado, Georgia, Indiana, Oregon, Maine, Maryland, Massachusetts, Minnesota, South Carolina, Texas, and Virginia.
    A full list of included HSAs is available here.
  • Note: These HSAs include a central city and its surrounding counties, and we include all HSAs with populations of at least 250,000 (with the exception of Massachusetts).

• New York City
  • Data source : New York City Department of Health and Mental Hygiene
  • Target areas: NYC (citywide)

• North Carolina
  • Data Source: North Carolina Division of Public Health’s (NC DPH) statewide syndromic surveillance system
  • Target areas: 7 geographic regions comprised of adjacent counties

• Utah
  • Data Source: Utah Department of Health and Human Service’s statewide syndromic surveillance system
  • Target areas: 13 geographic regions comprised of adjacent counties

Models Included

• epiENGAGE-GBQR : A machine learning method that combines gradient boosting and quantile regression to predict conditional quantiles
• epiENGAGE-INFLAenza : A spatial time-series model that uses the R-INLA package for estimating forecast posterior distributions.
• epiENGAGE-Copycat : A pattern matching model that matches growth rate trends to historic growth rate curves.
• Epiforecasts-dyngam : A Bayesian hierarchical GAM with a univariate AR(1) process
• epiENGAGE-baseline : simple time series model as a reference model
• epiENGAGE-ensemble_mean : An equally weighted mean ensemble that takes the mean at each quantile level of all eligible forecasts
• epiENGAGE-log_norm : Linear pool (a.k.a. distributional mixture) ensemble of quantile forecast submissions.

Information for Modelers

Anyone interested in submitting a model should review the guidelines for the 2026/2027 season on the Flu MetroCast Hub repository.

The latest-data.csv file in the Flu MetroCast Hub target data directory provides data that modelers can use to fit a model. Latest-data is specially formatted target data that includes the most up-to-date values for all dates for which target data is available, not just dates corresponding to the subset of target_end_dates for which the Hub provides forecasts.

Latest-data is the best source for modelers to get a single version of the target data with only the most recent observations, and including all historical data available from past seasons.