Prediction models for policy: Comparison of recent FACE observations of plant responses to ozone with predictions based on open-top chamber data

Background/Question/Methods

Periodic review of air quality standards entails comparisons of newly available data with the data and model results used in supporting past decisions. Past EPA ozone reviews have relied in part on simple empirical univariate exposure-response models for plants that were based on data from open-top chambers (OTC). In 1996, EPA developed and parameterized exposure-response models for the yield of selected crops and growth of selected tree species, using a three step process that relied on recursive model estimation and quantile statistics. Since that time, two long term research projects using Free Air (FACE) exposure methodology have produced observations for soybean and trembling aspen that can be compared to predictions from corresponding EPA models based on OTC experiments. Do the new observations, generated under widely different experimental conditions, validate model predictions based on previous data?

For aspen biomass data, the decrease in biomass observed between ambient and elevated ozone at AspenFACE was compared to the decrease predicted between those two values of exposure by the EPA model. The comparison was made for each of six consecutive years, with ozone exposure expressed as yearly cumulative average of the 90 day 12 hr W126. For soybean yield data, the decrease in yield observed between ambient and elevated ozone at SoyFACE was compared to the decrease predicted between those two values of exposure, in two separate years for which yield data were available for an assortment of genotypes. In addition, SoyFACE data for one genotype over 5 years, and 6 genotypes over two were aggregated to develop a model using the same three-step process as was used to create EPA’s predictive model, and the two models were compared.

Results/Conclusions

The agreement between prediction and observation was strikingly close in all cases. This demonstrates that a simple univariate model of exposure-response can have excellent predictive capabilities. The comparisons provide strong support for the process used to develop the models in 1996, and for the validity of both OTC and FACE experimental data. It should be noted that both the predictive model and the observations were aggregated over several levels of one or more variables, including study, genotype or years. It is to be expected that the agreement between predictions and observations for a single study comprising a single genotype in a single year would generally not be as close.

This abstract does not necessarily reflect EPA views or policies.