About the Data and the MAPS Forecast Model
What the map shows
The map shows how many new COVID-19 infections are expected each day in every part of the contiguous United States, down to neighbourhood scale. The colour of each spot is the number of new infections per square kilometre per day.
- Density, not risk. Cities look darker because more people live close together. A rural county can have a high infection rate per person and still look pale. Hover over a location to see its own numbers over time.
- Logarithmic colour scale. Each colour step is roughly three to four times the one below it, so small colour changes mean large differences.
- Estimates, then forecasts. Days up to the most recent data are estimates built from wastewater surveillance. Days after that are model forecasts. Press play to step through time.
- New infections, not cases. The numbers include people who never test or feel ill, so they are much larger than reported case counts.
Where the numbers come from
The forecast starts from what people shed into sewers, not from test results. Testing collapsed after 2022; wastewater still sees everyone.
| Input | Source | What it gives the model |
|---|---|---|
| Daily new infections by state | WHN estimates from CDC National Wastewater Surveillance System data, converted to infections with factors calibrated against earlier wastewater (Biobot) and IHME estimates | How many people are being infected now, and over the past 500 days |
| Variant mix | Genomic sequencing (GISAID; CDC variant proportions by HHS region) | Which variant families are circulating and how fast each is growing |
| Weather | gridMET daily temperature, humidity and heat index (Climatology Lab, University of Idaho), about 4 km, interpolated to the model grid | Indoor crowding in cold weather and in air-conditioned heat |
| School and holiday calendars | State school calendars; dated holiday and event schedule | Extra mixing when schools are in session and around holidays |
| Population | Gridded population for the contiguous US | Where people live, at roughly 1 km |
The WHN wastewater method is described on this page. Its own limitations carry into the forecast; see Known limitations below.
How MAPS works
MAPS simulates the epidemic the way a weather model simulates the atmosphere: the country is divided into a fine grid, and the model steps forward day by day, computing how infection spreads within and between grid cells.
- A fine grid. Each cell is about 1 km across (30 arc-seconds), more than 20 million cells over the contiguous US and its surroundings. Every populated cell keeps its own count of people who are susceptible, infected (with or without symptoms), recovering, recently immune and immune from longer ago, plus hospital admissions and deaths.
- Variants compete. About ten variant families are tracked separately, plus a pool for all others. Each has its own transmissibility and ability to escape existing immunity. Protection carries over between related variants more than between distant ones.
- Immunity fades with time. Protection is strongest just after infection and declines with time since infection, tracked in age bands. It never falls to zero, and susceptibility differs from person to person, so the most susceptible tend to be infected first.
- Conditions change transmission. Weather, school terms, holidays and behaviour adjust how easily the virus spreads. When local infections are high, people cut back contacts somewhat.
- Infection travels. People carry infection between neighbouring cells, so waves move across regions rather than rising everywhere at once.
- Starting point. Before each forecast, MAPS replays the previous 500 days of estimated infections by state and variant. That sets how much immunity each place has, and against which variants, on the first forecast day.
The forecast is an ensemble
Each forecast is built from nine model runs, not one. The runs differ in how transmissible the variants are (lower, central, higher) and how well they escape immunity (lower, central, higher), three by three. This is the same idea as an ensemble weather forecast.
- Blended, then shown. The nine runs are combined with weights that favour the runs matching recent observations best. The map shows that blended forecast.
- Spread is uncertainty. When the nine runs agree, the forecast is more trustworthy. When they fan out, the future depends on things we cannot yet measure well, such as how a new variant will behave.
- Skill falls with lead time. The first one to two weeks are the most reliable. Later weeks show where the epidemic is heading under current conditions, not a precise count.
- Checked against the past. MAPS is tested by starting forecasts on past dates and comparing them with what wastewater later showed. In a test started July 12, 2026, the central run stayed within about 20% of the observed weekly level for three months and placed the September peak within two days.
Hospital forecasts (coming)
MAPS also forecasts COVID-19 hospital admissions, and a map for public health officials will follow in the same format.
- Where. Admissions are assigned to about 5,300 individual hospitals, based on where each hospital's patients live.
- How. Infections in each grid cell are converted to admissions using admission rates that change month by month, calibrated against reported hospital data.
- Why it helps. Infections lead admissions by about one to two weeks, so an infection forecast gives hospitals advance notice of rising demand.
Known limitations
The forecast is only as good as the surveillance behind it, and several gaps are known.
- Recent wastewater is incomplete. The last one to two weeks rest on fewer treatment plants and are revised as late reports arrive, so the starting point of each forecast can shift.
- Not every state has its own data. States without enough reliable wastewater coverage are filled from the national per-person rate, which hides their local timing.
- Sequencing is sparse. Far fewer samples are sequenced than in 2021–2022, so the variant mix, especially by state, is uncertain and can lag reality by one to two weeks.
- The absolute level is calibrated, not measured. Converting virus in sewage to numbers of infections relies on factors set in earlier years. If shedding per infection has changed, all levels shift together; the timing and direction of trends are more reliable than the exact counts.
- Contiguous US only. Alaska, Hawaii and the territories are not modelled, because the weather data covers only the lower 48 states.
- Beyond the last observed day, weather is neutral. Forecasts do not yet use weather forecasts.
- A model, not a prophecy. New variants, policy changes and shifts in behaviour can change the course of a wave faster than any model anticipates.
Using the forecast
Use the map to judge whether infections near you are rising, steady or falling, and to plan ahead.
- Look at the trend. Hover over your location and read the curve. The direction over the next one to two weeks matters more than any single day's number.
- Compare places. The map shows where infection is concentrated now and where the forecast expects it to move next.
- Act on rising curves. A rising forecast is a good time to test before visiting vulnerable people, improve ventilation, wear a well-fitting mask in crowded indoor spaces, and stay current on vaccination.
- Check back. The forecast is updated as new wastewater and sequencing data arrive; each update starts from the latest observations.
MAPS is developed by Dr. Joseph Eastman (World Health Network), with wastewater estimates from the WHN team.