Zhengyuan Zhu

Department of Statistics and Operations and Research

University of North Carolina at Chapel Hill


Spatial Network Design to Detect Regional Trends in ground level Ozone

One goal of the Clean Air Act Amendments of 1990 (CAAA) is to reduce ambient concentrations of atmospherically-transported pollutants. Monitoring data from networks such as the Clean Air Act Status and Trends Monitoring Network (CASTNet) can be used to estimate regional trends of these pollutants to evaluate the effectiveness of the CAAA. This paper presents spatial network design methodology to optimize the network's ability to detect and quantify future regional trends in air pollution by adding or relocating monitoring sites. The 1997-2003 CASTNet ozone data is analyzed for trends using a two-stage approach similar to Holland et. al. (2000) to illustrate the design methodology. In the first stage, a site-specific trend is estimated after adjusting for the influence of meteorology and season. In the second stage, trend is assumed to vary over the eastern U.S. as a realization from a Gaussian random field, and Bayesian kriging methodology is used to estimate regional trends and uncertainties. A simulated annealing algorithm is then used to select the optimal locations for future monitoring stations under different practical consideration. We use the expected length of a Bayesian prediction interval as the design criterion to account for the uncertainty of estimating spatial covariance parameters underlying the estimates of regional trend. This design criterion is directly related to the network's capability to detect and quantify trends. Designs that minimize the Bayesian prediction interval are compared with the original design and designs that minimize the kriging variance. At the end we discuss how to determine the sample size of a network for a specific objective.


Back to Colloquium Series