William Christensen

Air monitors begin showing an unhealthy level of arsenic in the atmosphere. Can regulators quickly discover which of several industries in the vicinity may be to blame?

Brigham Young University statistician William F. Christensen and three colleagues apparently have come up with a better way to track down such polluters. Their article, "Clustering composition vectors using uncertainty information," published this week in the journal Environmetrics (London), demonstrates a way to find the source of pollution. Their statistical method is credited with cutting misidentification by a third.

The other authors are Ann M. Dillner of the University of California-Davis; James J. Schauer of the University of Wisconsin, Madison; and C. Shane Reese of BYU.

Christensen, an associate professor of statistics at the Provo university and the lead author, said the technique is one tool among several in analyzing and pinpointing sources of pollution. Wind direction and concentrations of pollutants also are important.

The new technique takes account of relationships between sizes of pollution particles. Some particulates may be only one micron in size, while others are larger. By comparison, a human hair is about 100 microns wide.

Larger particles may by spewed by mechanical means: road dust that's kicked up, grinding residue, or brake dust, he said. The smallest particles are "generally coming from combustion processes, high temperature processes" like car exhaust or incinerators.

"We can measure the size of particulates of different elements, such as lead or potassium or whatever," he said in a telephone interview. Experts "can look at the size distribution going on from the smallest size to the biggest size."

If vehicles are emitting two different elements, each element may have a range of sizes that is similar to the other. In element A, the proportion of larger particles to medium particles to small particles may be a sort of fingerprint, a pattern that also shows up in element B. This indicates both elements are coming from the same place.

"We use a statistical tool called cluster analysis that we adapted to meet this particular situation," he said. Should environmental regulators wonder where a particular toxic material was coming from, they could examine what sizes of particles of this material are in the sample. Its proportion of large to medium to small particles might meet the fingerprint of other elements emitted by the same smokestack.

The paper says some previous knowledge of the source's basic particle fingerprint is helpful. Otherwise, multiple observations are needed.

"It could allow you to identify potentially some additional sources that ... maybe were kind of flying under the radar," Christensen said.

Through this type of statistical analysis, producers of dangerous material could be detected.

The study examined particles sampled near Houston in 2000, according to the report. Through computer analysis, the technique was compared with other approaches, it adds. A BYU release says, "Christensen's method improves the ability to match emitted particles to their source by considering a wider variety of potential sources."

Previous methods "in some cases ruled out the true polluter."

Christensen said the team began working on the project about two years ago. They published one version of a paper with a more straightforward type of analysis, he said. But while he was working on that he believed there was a better way, using statistics theory.

"The statistical idea took about a year to germinate," Christensen added. The team did the analysis again using the new technique, resulting in sharp improvements.


E-mail: [email protected]