Tracking a Transport Smoke Event

On the morning of July 14, 2026, an explosive wildfire outbreak across northwestern Ontario — hot, windy conditions driving fire growth intense enough to generate its own pyrocumulonimbus cloud — sent a smoke plume drifting south into the United States. Over the course of the day, we tracked that plume from source to the Mid-Atlantic using more than half a dozen different satellite and model tools, and the exercise turned into a useful case study in exactly what each type of tool can and can’t tell you.

The Trigger: A PyroCb Over Northwestern Ontario

A pyrocumulonimbus cloud (pyroCb) forms when a fire’s heat output is intense enough to drive its own violent convective updraft — essentially a fire-generated thunderstorm. Unlike ordinary boundary-layer smoke, a pyroCb can inject material directly into the upper troposphere or lower stratosphere, letting it travel farther and persist longer than smoke that stays trapped near the surface.

Meteorologist Kyle Brittain (@BadWeatherKyle) flagged the event in near-real time, noting fire-darkened skies as far downwind as Thunder Bay. NOAA’s Hazard Mapping System (HMS) analyst narrative for the day corroborated the source: vigorous, heavy-density smoke plumes drifting eastward out of the central Northwest Territories, southern Ontario, and Saskatchewan/Manitoba, with a broad light-density smoke layer stretching from Missouri to Quebec.

Watching It Move: A Tool-by-Tool Comparison

We used several distinct products throughout the day, each answering a slightly different question.

GOES-19/GOES-16 TRUE COLOR
Gave us the clearest visual read on the smoke’s leading edge — a milky, desaturating haze that dulls green land surface without fully obscuring it. Comparing an 8:48 AM frame to a 10:01 AM frame showed the haze boundary advancing visibly south in just over an hour, from a line roughly along State College/Williamsport down toward Harrisburg.

GOES DAY CLOUD PHASE DISTINCTION RGB
Solved a problem true color can’t: distinguishing thin smoke from thin cirrus cloud, which can look nearly identical in plain true color. Smoke, lacking a strong ice- or water-phase signature, renders as a distinct teal/green band, while genuine cloud shows up white or pink — letting us confirm the haze crossing Pennsylvania was atmospheric smoke, not cirrus.

NOAA HRRR-SMOKE
The operational WRF-Chem-based model, updated hourly at 3 km resolution, gave us actual forecast concentration values in µg/m³ rather than just visual haze. The tradeoff: NOAA’s own interactive viewer for this product has had reliability issues over the years, and the static NCEP graphics, while fast-updating, lack city labels — great for grid-level analysis, unhelpful for placing a plume relative to a metro area at a glance.

NOAA/GSL DYNAMIC ENSEMBLE SCENARIOS FOR IDSS
The more capable current replacement, with a selectable HRRR ensemble member, adjustable smoothing and opacity, and a “difference from grand ensemble” mode — but its photoreal MapLibre terrain basemap competes visually with the smoke shading in a way flatter basemaps don’t.

AIRNOW FIRE AND SMOKE MAP
Built on WFIGS incident data plus GOES smoke polygons, uses a deliberately flat, low-saturation basemap that keeps city labels legible under a desaturated smoke fill — the clearest public-facing design of the tools we tried, though its automated plume-detection layer has a documented gap: it’s least reliable in the morning and doesn’t register diffuse, elevated smoke that’s clearly visible on true color imagery.

SURFACE PM2.5 MONITORS
IQAir and AirNow ground stations were the key to resolving what turned out to be the day’s central finding.

The Key Finding: Elevated, Not Surface-Mixed

Despite the smoke being visually obvious on satellite and confirmed by multiple independent sources, surface air quality readings across the corridor stayed in the Moderate range all day.

That combination — heavy visual haze aloft, unremarkable surface PM2.5 — points to the smoke riding in an elevated layer, decoupled from the surface mixed layer, likely reinforced by a building 1025 mb surface high pressure system centered over West Virginia that day.

Where the Deeper Tracking Tools Come In

Two more capabilities are worth knowing about for anyone doing this kind of tracking regularly, even though neither was fully applicable in real time today.

Aerosol Optical Depth (AOD) is a satellite-derived, quantified measurement of total column aerosol loading — smoke, dust, haze, pollution combined — that fills the gap between “I can see haze” and “how much is actually up there.” NOAA/NESDIS produces an operational GOES AOD product using the same public satellite data as everything else in this piece.

NASA’s GEOS-FP (Goddard Earth Observing System, Forward Processing) goes a step further, running a full global aerosol transport model that separates black carbon (wildfire/biomass smoke), mineral dust, sea salt, and sulfate pollution into distinct tracked species, updated in near-real-time twice daily. A black-carbon-isolated view is the closest thing that exists today to “a satellite that shows only the smoke” — definitively separating wildfire smoke from dust or regional haze rather than requiring visual inference from an RGB composite.

• GEOS-FP forecasts and near-real-time analyses — gmao.gsfc.nasa.gov/weather_prediction
• GEOS-FP daily-updated global smoke (brown carbon AOD) visualization — svs.gsfc.nasa.gov/5624
• GEOS-FP raw data portal — portal.nccs.nasa.gov/datashare/gmao/geos-fp
• NASA PACE mission (hyperspectral aerosol-type sensing from orbit) — pace.gsfc.nasa.gov
• NASA Worldview (browsable near-real-time satellite layers) — worldview.earthdata.nasa.gov
• NOAA GOES imagery (CIRA SLIDER) — rammb-slider.cira.colostate.edu
• NOAA Hazard Mapping System (fire/smoke satellite analyst product) — ospo.noaa.gov/products/land/hms.html

A Capability Under Strain

It’s worth noting that the science and public communication infrastructure behind tools like GEOS-FP sits at NASA Goddard, a center that has absorbed some of the deepest proposed cuts under NASA’s 2026 budget reductions, alongside broader agency-wide staff losses through layoffs and early-retirement programs. The underlying satellite data remains publicly available and the operational models are still running — but the layer of expertise that turns raw model output into clear, public-facing products is the kind of capacity that erodes quietly under sustained budget pressure, even while the hardware keeps flying.