In order to protect sensitive individuals and support possible episodic emissions control efforts, a pilot program was undertaken to forecast episodes of high ozone concentrations for 24-72 h in the Baltimore metropolitan area. This program utilized Classification and Regression Tree (CART) algorithms as well as standard regression analysis and expert forecasts. All approaches tend to underpredict peak O3 concentrations at 24 h. The low bias varied from 5 ppbv for the expert forecast to 7 ppbv for the CART forecast. The standard error for forecast O3 varied from 17 ppbv for the regression forecast to 23 for the CART forecast. For high O3 (> 120 ppbv) events, the expert forecast has the best success with a detection rate of 50%, and skill scores varying from 0.40 to 0.50. For expert forecasts of greater than 115 ppbv, verifying against observations of greater than 120 ppbv, the detection rate rises to 75% with skill scores of 0.60-0.66. One of the principal sources of forecast error was underprediction of mid-day surface temperature by the standard meteorological models during extremely warm episodes. When perfect forecasts of meteorological parameters are utilized, the forecast skill of the objective measures increases to approximately the skill of the expert forecasts. Improvements can be made by utilizing more recent data for initialization of the ozone regression equations and the use of local forecasts to supplement temperature forecasts during warm periods.
All Science Journal Classification (ASJC) codes
- Environmental Science(all)
- Atmospheric Science