People throughout Pittsburgh were confused and concerned after some smartphones issued alerts about extremely hazardous air quality for the region earlier this month.
The Pennsylvania Department of Environmental Protection says the alert was the result of an equipment error in Beaver Falls, about 30 miles north of Pittsburgh. The state-operated sensor erroneously reported high levels of particulate matter 10, or PM10, that were similar to what occurs during wildfires.
That was then picked up by third-party apps that aggregate and predict daily air quality data from governmental and non-governmental sensors across the region.
Ana Hoffman with Carnegie Mellon University’s air quality engagement project—part of the university’s robotics and technology-focused CREATE Lab—said the error highlights why people should look at more than one air quality reading.
Single sensors can malfunction and often change throughout the day.
“A lot of things can set off a monitor, like a diesel truck going by or a train, and those are going to cause temporary spikes,” Hoffman explained.
Hoffman said it’s best to look at three different monitors. Doing so will indicate whether the spike is temporary and localized, or regional.
“If you see something that trends for a long time and seems to be corroborated by a few monitors, you can trust it,” Hoffman added.
At-home sensor company Purple Air maintains an online map that shows all of its sensors’ readings at once in real-time, which Hoffman said is one of the best tools for accurate monitoring.
Not all air pollution is the same
Before the state issued a statement about the sensor malfunction, Allegheny County officials initially blamed the false alerts on the prediction model used by several air quality data sources.
“Many apps try to predict what PM2.5 levels will be as the day goes on,” the Allegheny County Health Department said in a statement. “When the levels are increasing, the apps show a higher PM2.5 level for the rest of the day. When there is a weather inversion, the apps over predict PM2.5 levels until late morning.”
“The way the algorithm associated with most weather/air quality apps calculates AQI, places greater emphasis (i.e. statistical bias) on the more recent few hours of data, in order to develop their associated forecast. This can create a divergent result from the rolling 24-Hour average for PM2.5,” ACHD Bureau of Environmental Health Deputy Director Geoffrey Rabinowitz added in a statement on Nov. 1.
County officials said monitors measure air quality hourly and do not predict readings. At the time sent, county monitors measured “good” to “moderate” air quality levels for PM2.5, SO2 and ozone.
Levels of PM2.5, however, climbed later in the day, leading to a five-day air pollution episode warning. During that time, pollution-emitting companies were required to reduce their emissions.
Randy Sargent, director of visualizations at the CREATE Lab, said air-quality prediction technology is critical for that reason.
“We want to know ahead of time when pollution is going to be amplified because those would be really good times to cut back on the industrial processes that make so much of the really dangerous pollution in our area,” Sargent said.
While governments are required to report when measures particulate matter and other pollutants that exceed acceptable levels, Sargent said they don’t necessarily explain or analyze the different types of air pollution present and what those differences mean for residents.
"I think that sometimes we oversimplify when we measure PM 2.5 because a PM 2.5 from one source doesn't necessarily equal PM 2.5 from the other source," Sargent explained.
Sargent said the chemicals released in the coking process are especially hazardous.
To give residents a better idea of where daily air pollution comes from, the CREATE Lab developed Plume Pittsburgh, which visualizes air pollution patterns and industrial sources across Allegheny County.
