When Air Quality Models and Measurements Diverge: What Houston VOC Data Reveal

Why the Air Quality Models We Rely On May Be Missing Real-World Pollution

Every day, decisions that affect the air you breathe are guided by computer models that predict pollution levels. But what happens when those models don’t match reality? That’s exactly what our team set out to study in Houston, Texas, where pollution comes from a complex mix of traffic, shipping, and industrial sources.

Our research, published in ACS ES&T Air in June 2026, examined how predictions from a regulatory air quality model compared to real-world, hour-by-hour measurements near Houston’s Ship Channel.

How this applies to everyday life

Regulators use air quality models to help evaluate whether an area will meet federal air quality standards and to design plans for reducing pollution. If a model underestimates real-world pollution levels, control strategies to reduce pollutant emissions and improve air quality could be ineffective or even make the problem worse.

What we found

We compared predictions from a regulatory air quality model against real-world, hour-by-hour measurements near Houston’s Ship Channel. We focused on volatile organic compounds, or VOCs. This includes chemicals like benzene, ethane, propane, and ethene that come largely from industrial sources and react in sunlight to form ground-level ozone, the main ingredient in smog.

The results were striking: the model consistently underestimated how much of these chemicals were actually in the air, with average underprediction ranging from approximately 33% for benzene to 72% for ethane across the monitoring network. The errors were largest overnight and in the early morning, exactly when the monitors picked up sudden VOC pollution spikes that weren’t captured by the model. That means the model also underestimated how much ozone these emissions could form. Thus, in real life, smog-forming conditions may be worse than the model suggests.

We wanted to know why this was happening, so we tested it directly: when we doubled the VOC emissions data fed into the model, its predictions lined up much more closely with what was measured at the monitors. This told us the problem isn’t the science behind the model; it’s the underlying emissions data. For many types of industrial sources, current emissions inventories are built from annual averages and assumptions about routine operations. This approach can smooth over short-term pollution spikes and may miss the specific mix of chemicals actually being released.

Why it matters

Air quality models shape real-world decisions. If the data fed to those models causes the model to underestimate actual pollution, then decision-makers may not have the full picture of what’s really happening in the air. Better emissions data leads to better models, better policies, and ultimately cleaner air for everyone.

About the Authors

Alejandro Valencia is an Air Quality and Emissions Modeler and Project Manager specializing in photochemical, dispersion, and emissions modeling to address complex environmental challenges. With expertise in air quality modeling, environmental data analysis, and software development, he supports projects that inform sound scientific and regulatory decision-making. Alejandro holds a BS in Chemical Engineering from Florida State University and both an MS in Environmental Engineering and a PhD in Environmental Science and Engineering from The University of North Carolina at Chapel Hill. Outside of work, he enjoys rock climbing, mountain biking, trail running, hiking, camping, and photography.