Ponente
Descripción
The use of probabilistic forecasting has been growing in a variety of disciplines because of its potential to emphasize the degree of uncertainty inherent in a prediction. Interpretation of probabilistic forecasts, however, is oftentimes a difficult endeavor and can deter users who may benefit from such forecasts. To encourage broader use of probabilistic forecasts and to show their practicality in the field of air quality, we explore possible ways of interpreting forecasts from a statistical probabilistic air quality surface ozone model known as Regression in Self-Organizing Map (or REGiS) in detail. Four procedures to convert probabilistic to deterministic forecasts are studied using the data from the Philadelphia metropolitan area in southeastern Pennsylvania. The procedures are based on calibrating threshold probability with 1) climatological relative frequency, 2) the more likely event, 3) the threat score, and 4) the bias ratio. REGiS is trained using 2000-2011 ozone season (May 1 to September 30) data, calibrated using 2012-2014 data, and evaluated using 2015-2018 data. Assessment of skill scores indicate that the way in which a probabilistic forecast is converted to a deterministic forecast matters. Using REGiS for the Philadelphia study area and time period, the greatest skill is derived from its probabilistic forecasts through a calibration process based on climatological relative frequency. For other probabilistic models and situations, different calibration procedures may be more beneficial.
