Document v6x23yev3qNnDJ7kz78EV6Lm6

AR226-2319 Charles J Zarzecki 04/02/2003 03:14 PM To: Catherine A Barton/AE/DuPont co: Subject: Beychok Article C a th ie , I finally was able to review the Beychok article on error propagation In dispersion m odeling. This could add fuel to Dave Rurak's fire. It's common knowledge that Gaussian models overpredict by a factor o f 2, however, due to error propagation, he says as high as 80X. Also, as far as 1-hour concentrations are concerned, the actual averaging tim e for Pasquill's dispersion coefficients range from 3 minutes to 30 minutes, depending who you talk to. I always thought they w ere 15-m inute averages. This short-term to 1-hour assumption can be shown (by Beychok) to result in a 2.5X overprediction. However, in my opinion, since the 1-hour period is the basic tim e-step in the ISC (and other) .model, an annual average concentration at any given receptor is 8,760 1-hour concentrations divided by 8,760 hours. If it predicts poorly fo r each 1-hour period, then it predicts poorly for the whole year. Dave has to understand that the model is a screening tool. The more accurate inform ation you put into a model, the greater th e accuracy o f the prediction. Instead of lambasting (is that a real word?) the model's 1-hour prediction, we should investigate how to fine-tune the input data and model options (e.g. particle settling, hours of operation, hourly emission rates, more accurate model (A ER M O D ), e tc...). O r, go out and do some sam pling to get the "real thing". W hat do you think? Beychok references about a half-dozen other publications that discuss the shortcomings of Gaussian m odels. Some o f them go way back. As you can see, nobody really took them seriously. W hat f am getting at: the regulatory agencies (O EPA, W V D EQ ) are not going to want to hear about how bad IS C is for particular tim e periods. They may be open to other "accepted" models and to sharpening our pencils. ' Regards, Charlie Z. EID747732