Monte Carlo Studies of Uncertainties in UAM-V Predictions for the July 1995 OTAG Period

Photochemical grid models are being used to make decisions concerning emissions controls. An example is the application of UAM-V by the Ozone Transport Assessment Group (OTAG) to the July 1995 OTAG ozone episode in the eastern U.S. It has been shown by observations and by modeling that ozone and its precursors may be transported 1000 or more kilometers, and therefore emissions controls in one region may have effects in far distant regions. However, models such as UAM-V make use of many input files and many model algorithms, some of which have more uncertainty than others. The question addressed by the research is the effect of the uncertainties in input variables on the uncertainties in ozone predictions. The uncertainties (medians, variances, and distribution functions) of 128 UAM-V input variables (emissions, initial and boundary conditions, meteorological variables, and chemical reactions) were assessed by expert elicitations.

Monte Carlo methods were used with simple independent random sampling from the distributions of the 128 input variables. 100 runs were made for each of four different sets of median emissions assumptions: 1995 actual observed emissions, 2007 emissions with EPA-assumed controls, 2007 emissions with 50 % reduction in median NOx emissions, and 2007 emissions with 50 % reduction in median VOC emissions. The location of the predicted domain-wide hourly-averaged ozone maximum occurred in the southeast for most Monte Carlo runs. The distribution functions of 100 predicted ozone maxima were close to log-normal in all cases with a standard deviation approximately equal to one-quarter of the median. This same result occurred for one and eight-hour averages, for OTAG-wide maxima, for sub-domain maxima, for individual sites, and for the various median emission reduction scenarios. It can be concluded that the uncertainties in the prediction of ozone are sufficiently large, with the 95 % confidence range approximately equal to the median, that it is difficult to determine whether the planned emissions reductions will be effective.

The NO2 photolysis rate was the variable whose uncertainties were most strongly correlated to the uncertainties in ozone predictions. Also important are wind speed and direction, relative humidity, cloud cover, and biogenic VOC emissions. These results should allow future research to be focussed on better specifying the input variables that have the strongest influence on the ozone predictions.