Estimating Emissions from Air Concentration Measurements

Abstract

A one-year-long experiment in which two different tracers were simultaneously released from two different locations was used to test various hybrid receptor modeling techniques to estimate the tracer emissions using the measured air concentrations and a meteorological model. Air concentrations were measured over an 8-hour averaging time at three sites 14 to 40 km downwind. When the model was used to estimate emissions at only one tracer source, 6 percent of the short-term (8-h) emission estimates were within a factor of 2 of the actual emissions. Temporal averaging of the 8-h data enhanced the precision of the estimate such that after 10 days 42 percent of the estimates were within a factor of 2 and after six months all of them (each source-receptor pair) were within a factor of 2. To test the ability of the model to separate two sources, both tracer sources were combined, and a multiple linear regression technique was used to determine the emissions from each source from a time series of air concentration measurements representing the sum of both tracers. In general, 50 percent of the short-term estimates were within a factor of 10, 25 percent were biased low, and in another 25 percent the regression technique failed. The bias and failures are attributed to low or no correlation between measured air concentrations and model calculated dispersion factors. In the regression method increased temporal averaging did not consistently improve the emission estimate since the ability of the model to distinguish emissions between sources was diminished with increased averaging time. However, including progressively longer time periods (more data) into the regression or spatially averaging the data over all the receptors was found to be the most effective method to improve the estimated emissions. At best about 75 percent of the estimated monthly emission data were within a factor of 10 of the measured values. This suggests that the usefulness of meteorological models and statistical methods to address questions of source attribution requires many data points to reduce the uncertainty in the emission estimates.