Available Data Can “Teach” You How to Model Air Quality

Our team was the Principal Investigator for EPA’s first major urban-scale air toxics study to model air quality.  This was the Philadelphia Integrated Environmental Management (IEMP) study of Philadelphia in the early 1980s.  We inventoried emissions for a large number of chemicals.  

Furthermore, these data are modeled and compared to a 10-station air monitoring network we established.  The modeled and measured data additionally matched reasonably well.  Our team had a question: what statistical tests should we do to demonstrate model performance? 

Our team contacted John Irwin, a senior modeler at the EPA. He responded to us—before the internet—and encouraged us to conduct any tests we desired, emphasizing that those tests were just for reference. Moreover, he stressed that allowing the measured data to “teach” us how to model effectively is far more critical. His insights continue to guide our approach to air quality modeling even 40 years later.

Models are complex.  Furthermore, many equations are used, and modeling involves sophisticated methodology.  A modeler can have an excellent grasp of all these concepts and calculations.  So, the data knows more.  It was there then and can provide a roadmap for refining the analysis.