Trend prognosis of regional ozone maxima in 1994 using various meteorologic data: appendix

The purpose of this study was to develop and test a statistical method for the short-term forecast of ozone concentrations. Austrian ozone monitoring data from April to September 1994 are used to develop the forecast model. It builds upon a multiple linear regression model developed earlier which uses the temperature of the forecast day, and the ozone maxima of the previous day as variables. In this study temperature difference between previous and forecast day, and wind velocity of the forecast day were additionally taken into account. Furthermore wind direction dependent regression models were developed using subsamples of the data set devided into 8 wind direction classes. Different regression function parameters have to be applied for each of the 40 selected ozone monitoring sites to allow forecasting of regional ozone maxima throughout Austria. It was found that regression models with temperature difference and wind velocity as additional variables did not improve the results. Wind direction dependent regression models only slightly improved the results for some wind directions at several monitoring sites. Best forecast results in general were achieved by using the base regression model with the temperature of the forecast day and the ozone maxima of the previous day as variables. Ozone forecast maps were calculated by spatial interpolation of the forecasted ozone maxima of the monitoring sites. Forecast accuracy is within ± 10 ppb on 70-80 % of the observed days. Errors higher than ± 10 ppb occur mainly on days with ozone maxima of 80 ppb and more. (author)