Predicting Pollutant Concentration Using Multiple Air Dispersion Models in a 2-D Superensemble.

摘要

Determining the level of air pollution is a modern day necessity for government regulators and industrialized sources. Air dispersion models are often used to determine the concentration of a pollutant. However, changing conditions and several assumptions made by the models limit their accuracy at various times. This research proposed combining several different air dispersion models (Gaussian Plume, Variable K Theory, Box, AFTOX, Hanna, SCREEN3) into a superensemble that has the ability to predict the concentration of a pollutant in both the downwind and crosswind directions. The superensemble member model pool was altered to determine the effect of adding high performing and poor performing models on its accuracy. In the interest of evaluating performance, the change in accuracy was measured through several statistical methods of evaluation (RMSE, MAE, NSE, R Squared, NMSE, Brier Score). It was found that in the prediction of the superensemble produced total average reduction of 27.9% from the other models over all of the statistical methods. Adding a less accurate model to the model pool decreased the superensemble by over 10%. In SO2 prediction, the superensemble produced an average reduction of 9.2% from the next lowest model. Multi-Criteria Decision Analysis tools were applied to determine which statistical method of evaluation would be optimal and it was determined that there is no one method that was best for every model.