Ricardo helps Canadian National railroad assess viability of natural gas traction; locomotive testing underway
An open letter to Ford from a customer

Rice/EPA study finds ground-level ozone falling faster than CMAQ model predicted

Researchers from Rice University and the US Environmental Protection Agency (EPA) report in a paper in the journal Atmospheric Chemistry that particularly in Northeastern cities, ozone levels dropped even beyond what was anticipated by cutting emissions of NOx from 2002 to 2006. The study suggests the Community Multiscale Air Quality (CMAQ) model significantly under-predicted the reduction in ozone.

Following major controls of NOx, ozone has come down more quickly than anticipated. This is good news. But it also poses a challenge because states rely upon models to predict whether they’ll attain ozone standards in the future. If the models have key uncertainties that affect their responsiveness, that can affect the states’ control strategies.

—Daniel Cohan, co-author

Ozone is not emitted directly but instead forms near the ground from precursor emissions of NOx and hydrocarbons. Modeling of this complex chemistry is important to help states comply with federal standards for ozone, which now stand at 75 parts per billion (ppb) and may be tightened by the Obama administration. A recent Rice study showed a positive correlation between high ozone levels and cardiac arrest. (Earlier post.)

In 2002, the EPA implemented a cap-and-trade program known as the NOx SIP Call to curtail emissions of ozone-forming NOx from industries in Eastern states. The significant reduction in emissions over the subsequent four years provided a real-world experiment for the researchers to test how well computer models predict improvements in air quality.

When a model fails to replicate an observed change, a key challenge is to discern whether the discrepancy is caused by errors in meteorological simulations, errors in emission magnitudes and changes, or inaccurate responses of simulated pollutant concentrations to emission changes. In this study, the Community Multiscale Air Quality (CMAQ) model is applied to simulate the ozone (O3) change after the NOx SIP Call and mobile emission controls substantially reduced nitrogen oxides (NOx) emissions in the eastern U.S. from 2002 to 2006.

For both modeled and observed O3, changes in episode average daily maximal 8-h O3 were highly correlated (R2 = 0.89) with changes in the 95th percentile, although the magnitudes of reductions increased nonlinearly at high percentile O3 concentrations.

Observed downward changes in mean NOx (−11.6 to −2.5 ppb) and 8-h O3 (−10.4 to −4.7 ppb) concentrations in metropolitan areas in the NOx SIP Call region were under-predicted by 31%–64% and 26%–66%, respectively.

The under-predicted O3 improvements in the NOx SIP Call region could not be explained by adjusting for temperature biases in the meteorological input, or by considering uncertainties in the chemical reaction rate constants. However, the under-prediction in O3 improvements could be alleviated by 5%–31% by constraining NOx emissions in each year based on observed NOx concentrations. This demonstrates the crucial need to accurately characterize changes in precursor emissions when dynamically evaluating a model’s ability to simulate O3 responses to those changes.

—Zhou et al.

Faster-than-expected reductions in NOx emissions may explain some but not all of the gap. The remaining gap may result from inaccuracies in how the model represents the chemistry and transport of air pollutants, Cohan said.

How ozone responds to changes in NOx and hydrocarbons is a nonlinear chemistry. So it’s certainly possible that even the best models could be slightly inaccurate in defining those relationships. It tells us that, as modelers, we need to revisit the formulations, especially the chemistry.

—Daniel Cohan

The observed (dash line) and modeled (solid line) daily maximal 8-h O3 as a function of percentile n, averaged over all monitors inside the NOx SIP Call region. Zhou et al. Click to enlarge.

While it may be preferable for models to be a bit conservative rather than too aggressive in predicting ozone improvements, Cohan said, the models are intended to represent air pollution as accurately as possible. A study by Cohan’s research group last year showed that regulatory modeling by states tended to slightly under-predict the ozone reductions that were actually achieved.

The goal of everyone in the process is to reach attainment in the most cost-effective manner possible, and we need accurate models to inform those decisions.

—Daniel Cohan


  • Wei Zhou, Daniel S. Cohan, Sergey L. Napelenok (2013) Reconciling NOx emissions reductions and ozone trends in the US, 2002–2006, Atmospheric Environment, Volume 70 Pages 236-244 doi: 10.1016/j.atmosenv.2012.12.038



This doesn't surprise me a lot. The entire study of environment, climate (change), appears to be an ongoing best guess , it is defineitly not an art. It is more a consensus of opinion .


An example of above, the Natioal Science Foundation just released a document (press release 13-097) that over the last 11,300 years , 20 to 30% of those years were warmer than present day. That is quite high in my opinion.


Gee Whiz,

Some of the tame puppy EPA scientific doggies usualy relied on to post the usual doom and gloom forcasts, betrayed the EPA eco-religionists. No matter how hard they tried to do otherwise, they HAD TO ADMIT the Air is clearing and Air Quality Compliance is being met.

Who would have thought such a day would ever occur?


I actually conducted a GLO validation study of CMAQ in 2005 for one forecast point (Knoxville, TN - http://www.srh.noaa.gov/ssd/techmemo/sr227.pdf).

However, it was our conclusion that inaccuracies of NWP (NAM) model projections generally correlated with poor GLO concentration projections of CMAQ.


I also conducted a study (no paper submitted) of the correlation of NOx concentrations to ambient GLO concentrations at Look Rock, TN (a very remote location near the Great Smoky Mountains National Park). GLO concentration correlated very weakly with NOx concentrations (R^2 = 0.007).

Unfortunately, the monitor at Look Rock did not have a VOC detector, so I was unable to see how GLO correlated with VOC concentrations.

Nevertheless, this is another data point which tends to point to NWP model prediction errors rather that a larger-than-anticipated response to NOx reductions (not that I'm criticizing NOx reductions) or errors in the CMAQ chemistry.


Some party pooper has to come along and turn a good news into a bad news.

The comments to this entry are closed.