15 sources of wildfire smoke forecasts for North America » Yale Climate Connections

A case in point: You would think that a heavy rainstorm would clean the air of smoke – and it often does – but at my house in southeast Michigan in June, pollution from the main health-damaging component of wildfire smoke, PM 2.5, increased to the red “Unhealthy” level during a six-hour rain storm that dumped over a half-inch of rain.

Read: Silent calamity: The health impacts of wildfire smoke

This counterintuitive event occurred when the circulation associated with the rain storm pulled down particles from an elevated plume of smoke overhead. Worse, the models didn’t know that the elevated plume was there and did not properly forecast the increase in my Air Quality Index, or AQI. When the models finally did show an extensive band of “Unhealthy” air, the forecast for the placement of the band was displaced by several hundred miles.

Satellites are good at telling us how much total smoke is present in a column from the surface to the top of the atmosphere, but not at the concentrations of smoke at various elevations, which you typically need to make a good smoke forecast.

What’s more, the satellites used by some models pass over a specific area only once a day. As a result, if a fire starts after the satellites pass over or if the fire significantly changes size after the satellite passes, that information will not be in the forecast until the following day.

As a result, smoke forecasts are often way off, particularly beyond one day. This is particularly so if new fires start or if established fires diminish because of firefighting efforts or precipitation. Smoke models typically do not allow for fires to spread horizontally, and they have to make assumptions about how high the smoke plume will rise, what type of vegetation is burning, what chemical reactions will affect the smoke, and other hard-to-quantify factors. A 2024 paper documenting NOAA’s AQMv7 model said it this way: “the AQMv7 demonstrates large bias/error for PM2.5 near and downstream of wildfire sources in the western U.S., indicating uncertainties in fire emissions, transport, and plume chemistry.”

How to use smoke model forecasts

Use multiple models to understand your wildfire smoke risk, and check the initial smoke concentrations the models are using at the beginning of their runs to see if they match the actual concentrations. Put more trust in models that get the initial conditions right! (But even if a smoke forecast does not get the absolute levels of PM 2.5 right, it can be useful to see trends.)

Look particularly closely where major fires are burning. Sometimes the models do not know where fires are burning because of cloud cover, or they assume that there is a major source of smoke where fires may have suddenly died down. I’ve seen errors in the assumed starting concentrations of smoke of over a factor of 10, for both underestimates and overestimates. Errors are more likely to occur when dense cloud cover is over major fires, preventing accurate satellite estimates of fire intensity and initial smoke concentrations.

Keep in mind that PM 2.5 particles are also emitted by many other human sources, such as cars, power plants, and industrial activity. It is good to know the total PM 2.5 levels from all sources, not just wildfire smoke. During a wildfire smoke episode on a hot summer day with light winds, PM 2.5 from smoke often will not be the dominant source of total PM 2.5, particularly in cities.

How to check your current air quality

I recommend these sources of current PM 2.5 data:

• EPA’s Fire and Smoke Map, which includes data from official EPA monitors and four kinds of low-cost backyard sensors, most notably from purpleair.com
• Purpleair.com, a company that has a huge global network of low-cost backyard PM2.5 sensors. I’ve had one in my backyard for over 10 years and have been very pleased with it (the cost is $239-$299). Note that these inexpensive devices are known to read about 10-40% too high during episodes of intense wildfire smoke, depending upon the levels of smoke, according to a 2024 study. The purpleair.com site allows you to apply the appropriate correction factor: At the upper left of the map, click on the text, “US EPA PM2.5 AQI,” then choose the “Apply Conversion” option, then select “US EPA.” The purpleair.com site also allows you to display current PM 2.5 levels as AQI or concentrations in micrograms per cubic meter, with the latter option being handy to compare to what smoke models are predicting.
• The NOAA Hazard Mapping System Fire and Smoke Product has an interactive map of smoke plumes and fire locations. There is also a twice-per-day text analysis of the smoke situation, plus areas where blowing dust is occurring, based on satellite analyses.
• IQAir has a nice global map of current PM 2.5, including a fancy 3D map on a rotatable sphere.
• For Canada and Alaska, aqmap.ca has a map with PM 2.5 observations from official monitors and four types of backyard sensors.

EPA human-generated forecasts

In the U.S., the gold standard for knowing how smoke will impact air quality is information at EPA’s AirNow.gov. It provides a graphical air quality forecast for today and tomorrow and offers links for text-based air quality forecasts from the state air quality department, where available. These will specifically discuss wildfire smoke when it is expected to reduce air quality.

PROS: The highest-quality smoke forecasts available, combining model data with expert human interpretation. The forecasts are also available on the excellent (and free) AirNow app.
CONS: Forecasts only go out one day.

Environment Canada human-generated forecasts

In Canada, Environment Canada has three-day text-based forecasts of its Air Quality Health Index (AQHI), which includes the impact of wildfire smoke. Environment Canada’s Interactive Weather Alerts map shows where Air Quality Warnings and Air Quality Statements are in effect. Clicking on the appropriate location will bring up text with the smoke forecast.

PROS: The highest-quality Canadian smoke forecasts available, combining model data with expert human interpretation.

Computer model wildfire smoke forecasts

Computer models to predict surface wildfire smoke concentrations are run by multiple agencies in the U.S. and Canada. These models are relatively new — the first operational smoke model, HRRR-Smoke, came online in 2020. Improvements are coming at a rapid pace. The output from the models for the main health-damaging component of wildfire smoke, PM 2.5, is expressed in micrograms per cubic meter, so you will have to know the conversion factor to get the Air Quality Index (AQI). In the U.S., the EPA standard for 24-hour PM 2.5 levels is 35 micrograms per cubic meter, which corresponds to an AQI of 100 (orange, “Unhealthy for Sensitive Groups”). The threshold for the red “Unhealthy” range is 55.4 micrograms per cubic meter, which corresponds to an AQI of 150. You can use EPA’s AQI Calculator to do these conversions.

Keep in mind that the U.S. 24-hour PM 2.5 standard of 35 micrograms per cubic meter is set higher than it should be, according to many recent scientific studies: Breathing PM 2.5 has increasingly been found to cause more severe health problems than previously realized. The Canadian 24-hour PM 2.5 standard is 27 micrograms per cubic meter, and the international World Meteorological Organization standard is just 15 micrograms per cubic meter, equivalent to an AQI of 62 on the U.S. scale. I develop a cough if I breathe outside air for more than a few minutes when the PM 2.5 AQI from wildfire smoke reaches about 60 – close to the WMO standard, and I believe that the EPA boundary for an AQI of 100 – the beginning of the orange “Unhealthy for Sensitive Groups” range – should be at the WMO standard of 15 micrograms per cubic meter.

Below, I give an overview of seven computer model forecasts of wildfire smoke and/or total PM 2.5. It is important to realize that your health is affected by total PM 2.5 – the combination of small particles from wildfire smoke plus other sources, such as car exhaust, power plants, and factory emissions. Many smoke models (like NOAA’s flagship HRRR-Smoke model) only give PM 2.5 levels from wildfire smoke, which will underestimate the total health burden of all of the PM 2.5 in the air.

U.S. wildfire smoke models

The U.S. air quality model I like the most for predicting wildfire smoke is NOAA’s Air Quality Model (AQM v7), which became operational in 2024. It provides hourly pollution forecasts for North America going out three days. From the “Air Quality Elements” drop-down menu, select “Bias-Corrected PM2.5,” which will display PM 2.5 forecasts from all sources, including wildfire smoke. There are options to plot hourly, 24-hour average, or maximum daily PM 2.5. The numbers shown on the images are in micrograms per cubic meter, but the colors are contoured with colors corresponding to the AQI scale (green, yellow, orange, red, purple, and brown) so that you don’t have to mentally convert from micrograms per cubic meter to AQI.

PROS: Shows levels of total surface PM2.5, not just from smoke (emissions from cars, factories, etc. are often the dominant source of PM2.5); output shows both micrograms per cubic meter and AQI.
CONS: Lower resolution (13 km); runs only twice per day (midnight and noon UTC, which is 8 p.m. and 8 a.m. EDT); data doesn’t arrive until about seven hours after initialization time.

The other U.S. model I like to use is HRRR-Smoke, the top operational U.S. model for forecasting wildfire smoke. The meteorology is done by the High-Resolution Rapid Refresh (HRRR) model, which also provides 48-hour weather forecasts nationwide with a fine grid resolution of three kilometers. HRRR-Smoke makes 18-hour smoke forecasts every hour and 48-hour smoke forecasts every six hours (at 0, 6, 12, and 18 GMT) for the U.S. (there is a separate version of the model for Alaska, HRRR-AK Smoke). The most useful option to predict air quality is “near-surface smoke.” There are also options to look at the smoke 1,000 feet high, 6,000 feet high, and through the whole depth of the atmosphere.

PROS: The highest-resolution computer smoke model.
CONS: Includes PM 2.5 only from smoke, not from human emissions or blowing dust.

RAP-Smoke is a coarser resolution (13 km) U.S. model useful for predicting smoke transport on a regional level. Forecasts go out 21 hours and are generated every hour. The model runs considerably faster than HRRR-Smoke, so data is available several hours sooner. The RAP- and HRRR-Smoke User’s Guide has details.

Global wildfire smoke models

The model I like the most for long-range smoke forecasts for North America is NASA’s GEOS-FP model (Goddard Earth Observing System Forward Processing). The early run (0 UTC or 8 p.m. EDT) goes out a full 10 days, the longest of any smoke/PM 2.5 model. The late run (12 UTC or 8 a.m. EDT) makes five-day forecasts. Keep in mind that forecasts more than a few days in advance will be highly inaccurate! This is particularly so if new fires start or if established fires diminish because of firefighting efforts or precipitation.

PROS: Global; forecasts all PM 2.5 (from smoke plus human-emitted air pollution); only model to make ultra long-range PM 2.5 forecasts (10 days)
CONS: Coarse resolution (about 25 km); data not available until about 11 hours after the initialization time; grids are continental-scale, making it hard to see local detail; inaccuracy: I’ve seen the model incorrectly initialize smoke levels, resulting in a PM 2.5 forecast five times too high.

CAMS (the European Copernicus Atmosphere Monitoring System) provides five-day global forecasts of PM 2.5 that include wildfire smoke. There is a 10-km higher-resolution version of the model for Europe (and in France, the PREVAIR model offers five-day PM 2.5 forecasts for that nation). Windy.com has a nice interface on its website (and on the free version of its app) to display the CAMS data on its animated map with wind streamlines.

PROS: Global; forecasts all PM 2.5 (from smoke plus human-emitted air pollution); one of only two long-range PM 2.5 forecast models (five days).
CONS: Coarse resolution (about 40 km), and the grids it runs on are large, so it’s hard to see local detail; runs only once per day (at 0Z); the minimum concentration of the contours is set too high (20 micrograms per cubic meter).

The SILAM model from the Finnish Meteorological Institute has a seven-day global PM 2.5 forecast with a grid available for the CONUS.

PROS: Long-range, seven-day forecasts available.
CONS: Coarse resolution; can’t save the URL for the CONUS grid in a bookmark, so you must navigate from the default European map to get the CONUS map.

Canadian wildfire smoke models

Another of the top-five smoke models I like is Environment Canada’s three-day air quality forecast for North America from its Air Quality Model of surface PM 2.5, updated twice daily at 0 UTC (which updates around 1 a.m. EDT) and 12 UTC (which updates around 1 p.m. EDT). Imagery for total PM 2.5 from all sources, and just from smoke, is available.

PROS: Forecasts all PM 2.5 (from smoke plus human-emitted air pollution); provides three-day forecasts; has zoom-in maps for Canadian provinces; decent grid resolution (10 km).
CONS: Does not initialize smoke concentrations well in some cases.

FireSmoke Canada (AKA The BlueSky Canada Smoke Forecasting System) produces a two-day surface smoke forecast that shows the locations of current major wildfires. There are options to see the hourly, daily average, and daily maximum smoke levels. It’s produced by the Weather Forecast Research Team at the University of British Columbia. It has an experimental forecast for Western Canada as well.

PROS: Shows boundaries of active fires; decent grid resolution of 12 kilometers.
CONS: Accuracy – I’ve seen the model improperly initialize smoke conditions, giving PM 2.5 levels over 10 times too high.

Experimental U.S. wildfire smoke models

RRFS-SmokeDust (the Rapid Refresh Forecasting System Smoke and Dust model) is NOAA’s next-generation, soon-to-be-operational weather-smoke forecasting model. RRFS-SmokeDust makes 18-hour smoke forecasts every hour and 3.5-day smoke forecasts every six hours (at 0, 6, 12, and 18 GMT) for the U.S. The three-kilometer high-resolution model has improved physics, data assimilation, and handling of meteorology compared to HRRR-Smoke. It’s slower to run than HRRR-Smoke and has only one large domain covering North America. It includes PM 2.5 from both smoke and dust.

RAP-Chem is an even fancier experimental model that does two-day forecasts of a number of air pollutants, including PM2.5, NOx, CO, and ozone. It forecasts not just smoke but all PM2.5, including the background levels of pollution from cars and power plants. The model runs once per day, at 6 UTC (2 a.m. EDT), and is rather slow to run: The 24-hour forecast point is typically available after 10 hours (around noon EDT), and the 48-hour forecast point comes in around 3 p.m. EDT. Data is not always available since it is experimental.

The U.S. Forest Service BlueSky modeling system is a suite of tools that allows you to combine a number of weather models, air pollution models, fire models, and smoke transport models. The tools are particularly valuable for showing high-resolution forecasts of wildfire smoke in the Western U.S. The EPA’s Community Multiscale Air Quality Modeling System (CMAQ) model is the main tool used to make the PM 2.5 forecasts. The U.S. Forest Service also has a portal, Wildland Fire/Air Quality Tools, which has BlueSky and other useful air quality info.

U.S. Pacific Northwest smoke model

AIRPACT is a high-resolution (4 km) air pollution model, which includes wildfire smoke, providing three-day PM 2.5 forecasts for the U.S. Pacific Northwest. It forecasts all PM 2.5 (from smoke plus human-emitted air pollution) and has the option to display PM 2.5 as AQI.

Apps for forecasting wildfire smoke

There are over 100 air quality apps for smartphones that provide PM 2.5 forecasts, and it is beyond the scope of this post to present a review of them. The most popular apps — like Plume Labs (integrated into AccuWeather) and AirVisual from IQAir — do not offer forecasts from any of the smoke models listed above, like HRRR-Smoke, CAMS, or FireWork. There are, however, at least three apps that may be more valuable, because they do offer wildfire smoke forecasts from some of the advanced smoke models:

The Google Weather App (built-in on many Android/Pixel phones) includes PM 2.5 forecasts with smoke model data sourced from BreezoMeter. (Note that the built-in weather app on iPhones offers only current AQI from BreezoMeter, not an AQI forecast.)

Gaia GPS (free 14-day trial) includes a “Smoke Forecast” overlay in the map layers that uses HRRR-Smoke model output. It offers real-time (current), 24-hour, and 48-hour smoke forecast maps at three-kilometer resolution across the continental U.S.

Windy.com has a dedicated surface PM 2.5 forecast layer powered by the European CAMS model. The free version of the app offers four-day PM 2.5 forecasts.

Read: How to protect yourself from wildfire smoke

Bob Henson contributed to this post.