Experiment
Where does it go once it’s in the air?
Wildfires, burn piles, refineries, power plants, and the new wave of data centers with their own gas generation. Emissions start in one place and end up somewhere else. We wanted to see how they move around the state, so we built a simulator of Utah’s wind on Utah’s real terrain, and put it online for anyone to use.
Launch the simulatorFree, no account. Works on desktop and phone.
The question
Emissions released anywhere in Utah are steered by ranges, canyons and valley floors. Smoke from one county can drain into the next. On winter nights a lid of warm air can sit over the valleys and hold everything beneath it in place for days, far from where it started.
A wind rose from one weather station cannot show that. We needed the actual terrain, the actual weather, and the physics that ties them together. The project started with a local question, where the exhaust from a large data center’s on-site gas plant in Iron County would end up, and grew into a tool for any source: put smoke anywhere in the state, choose a forecast day or a past date, and watch which towns it reaches and how many hours it takes to get there.
What it does
Thousands of flow lines trace the wind over every ridge and valley in the state. Drop smoke anywhere and the model carries it with real weather over real ground.
01
Live mode blends the National Weather Service forecast with dozens of airport wind observations near the ground and NOAA HRRR winds aloft. Step forward through the forecast, about 7 days from NWS and out to 16 days with NOAA GFS, or replay any date since 2022 from archived model and airport winds. A manual mode lets you set the wind yourself.
02
Drop up to 16 colour-coded sources anywhere on the map, as a continuous plume or a single puff, and tune emission, source size, plume rise up to 4 km, and lifetime up to 72 hours. A live list shows which towns the smoke reaches at ground level and how many hours after release.
03
Wind source, mixing height, day-time heating or night-time cooling, turbulence, and smoke settings sit beside the map. Hover anywhere for elevation, wind speed and direction, mixed-layer depth, and smoke at the ground and aloft. Switch between a flat 2D map and tilted 3D terrain, search for a city, and toggle roads, stations and forecast arrows.
The model
Elevation for all of Utah and its edges comes from USGS 3DEP data at about 157 m resolution. The atmosphere is a grid of 16 terrain-following layers, about 40 m thick near the ground, so air rises over ridges, sinks in their lee, and squeezes faster through gaps and canyons.
Every step, the wind field is corrected so air is never created or destroyed. That one rule is what makes flow speed up through a narrow pass, pile up against a range, and pour around a mountain instead of through it.
The mixing height, the lid of warm air that caps the valley, comes from the forecast or is set by hand. Smoke released under the lid tends to stay under it, and ridges that poke through the lid act as walls.
Daytime heating drives air up the slopes. At night the ground cools, and cold air drains downhill and pools in the valleys, carrying whatever it holds with it.
Proving the physics
Before trusting it with real questions, we checked the model against something every Wasatch Front resident knows: a winter inversion. In this constructed example, a warm lid sits about 450 m above the valley floor and a light breeze drifts up from the south. Smoke released at Tooele, Provo and Nephi cannot rise through the lid, so it follows the valleys, and by the end of the first day the Salt Lake Valley is filled wall to wall.
Settings: Manual wind from 170° at 2 m/s, mixing height 1,750 m, night-time cooling 0.70, continuous plumes, 72 h lifetime. An illustration of a typical inversion, not a replay of a specific day.
How it was built
The whole thing is a static site. The heavy lifting happens on the visitor’s own graphics card, and the weather comes straight from public government feeds.
01
The simulation runs as WebGL2 shaders on the visitor’s GPU. Wind and smoke are moved with a second-order semi-Lagrangian scheme, a pressure-style projection keeps the flow mass-conserving, vertical mixing is solved implicitly so it stays stable at large time steps, and vorticity confinement restores the small eddies that coarse grids smear away. One second on screen is roughly one to two hours of simulated time.
02
The browser calls the National Weather Service API, Open-Meteo’s NOAA HRRR and GFS feeds, and the Iowa Environmental Mesonet directly. A one-time Python job maps the grid to NWS forecast points and airport stations, so there is no backend, no API key, and nothing to keep running.
03
A Python build step downloads elevation tiles, stitches them into one grid covering Utah plus a margin, so terrain upwind of the border still shapes the wind coming in, and packs it into a compact binary file. It joins US Census place data so the map knows every town and its population.
04
Grid resolution can be set from about 2.2 km to 1.25 km cells to trade detail for speed. Thousands of animated flow lines trace the wind, and the layout switches to a bottom sheet on phones. Keyboard shortcuts cover the desktop.
Search for a town, choose Place smoke, and click where the emissions start. Then step through the forecast and watch which towns show up in the arrivals list. Needs a browser with WebGL2, which every current desktop and phone browser has.
Utah Wind & Smoke is a simplified illustration of airflow and smoke transport, not an official smoke or air-quality forecast, and it is not endorsed by NWS, NOAA or USGS. It tracks where emissions travel, not their chemistry or health effects. For health decisions, check the Utah Division of Air Quality and AirNow.
Maps, models, live public data, and interactive tools that make a hard question easy to see. If you need one built, we would like to hear about it.