Forecasting Models for Predicting Ozone Levels Near Cincinnati, Ohio

This paper presents the development of two empirical ozone-forecasting models for predicting hourly ozone concentrations near Cincinnati, Ohio. This study examined the hourly ozone data collected between 1995 and 1999 at an ozone monitoring station in the Cincinnati area. Two different models, the KZ model and the NN model, were developed using data from 1995 to 1998. The KZ model was developed by time series/ statistical analysis of existing data whereas the NN model was developed using artificial neural network. It was found that the NN model performed better as compared to the KZ model for the entire ozone season of 1999. However, a detailed analysis of the forecasting ability of the models indicated that the performance of the KZ model was better for high ozone concentration periods as compared to the NN model. On the other hand, the predictions by the NN model were better for time periods when the ozone concentrations were less than 60 ppb. The findings of the study indicate that a single robust model using any one of the methods for predicting hourly ozone concentration may not be a great performer. Therefore, a composite forecasting system combining the NN and KZ models is proposed to overcome the limitation of these models.