Prediction of surface ozone exceedance days using PCA with a non-parametric T2 control limit

•A PCA model with non-parametric control limit is derived in this paper.•Results show that this method can increase true prediction rate significantly.

It is noted that the data for ozone precursors and meteorological variables exhibit non-normal distributions. Thus, to detect surface ozone exceedance days, we propose a principal component analysis (PCA) model with a non-parametric T2 control chart. The input variables include concentrations of ozone (O3), nitrogen oxide (NO), and nitrogen dioxide (NO2), wind speed (WS), relative humidity (RH), solar radiation (SR), surface and aloft temperatures (T). In addition, process variation indicators of meteorological factors are proposed to capture dynamic weather patterns. Ozone precursors and meteorological measurements of non-exceedance days during extended summers (May 16th–Oct. 15th) from 2000 to 2007, which include 1153 days, are used to train a PCA model for the Livermore Valley, CA. Data of ozone exceedance days for the same period are used for validation of the model. Summer data from 2008 to 2009, which include 11 exceedance days, are used to test this prediction model. An ozone exceedance day is triggered when any T2 value of this day exceeds the non-parametric T2 control limit. Compared to a conventional PCA with a Hotelling T2 control chart, the true prediction rate (TPR) of ozone exceedance days using this PCA model with a non-parametric T2 control chart is increased from 45.45% to 72.73%. When process variation indicators (PVIs) are introduced into the PCA model with a non-parametric T2 control chart, the TPR of ozone exceedance days is shown to increase to 100%.