(53c) Comparison of Ozone Analyses between Camx and Denfis for Selected Monitoring Sites in an Ozone Nonattainment Area

Ground-level ozone is a secondary air pollutant produced by photochemical processes of nitrogen oxides (NOx) and volatile organic compounds (VOC). The Houston-Galveston-Brazoria (HGB) is identified by the USEPA as an ozone nonattainment area, and it is important to identify and quantify essential factors affecting ozone formation to improve the State Implementation Plans (SIP) for effective ozone reduction. In this study, site-specific ozone analyses were performed using a CAMx (Comprehensive Air Quality Model with Extensions) model and a DENFIS (Dynamic Evolving Neural-Fuzzy Inference System) algorithm at several selected ozone monitoring stations in the HGB area. The CAMx is a multi-scale, three dimensional photochemical grid model which simulates dynamic ozone concentrations under specific meteorological and emission inventory conditions. The DENFIS algorithm can generate data clusters along with the corresponding Takagi-Sugeno fuzzy rule set to predict dynamic ozone concentrations. Raw datasets for the fuzzy system were downloaded from the Texas Air Monitoring Information System on the TCEQ website. They were processed through a series of filters to ensure the quality of data before modeling. Each fuzzy system takes one year of hourly measurements as inputs for ozone modeling, which includes 11 meteorological variables and 5 types of air pollutants. The DENFIS algorithm uses the ECM (Evolving Clustering Method) for data clustering. Cluster center generation is distance-based, which has two modes: offline constrained optimization mode, and online one-pass mode. In the offline mode, pre-processed data were sampled every 17 hours for model testing, and the rest were used for model training; whereas in an online mode, new fuzzy rules can be dynamically inserted to update the model in real-time. The adaptive training is controlled by: (a) threshold value that determines number of cluster centers; (b) maximum number of iterations; (c) step size for least square optimization; and (d) width of the triangular membership function. Defuzzification of the system is a standard first order Takagi-Sugeno type.

The objectives of the study were to compare the effectiveness of the two models for ozone prediction and investigate the corresponding advantages and disadvantages of the two models for sensitivity analysis for the studied sites in the HGB area based on the 2012 Ozone Episode recently made available by TCEQ (Texas Commissions on Environmental Quality). Additional fuzzy inference analyses based on the 2013 through 2016 ozone data were also performed to establish the year-by-year trends of essential factors affecting ozone concentrations at selected monitoring sites in the HGB area.