Probabilistic Predictions of PM2.5 with an Analog Ensemble

Accurate air quality (AQ) predictions can provide individuals and communities with timely information to help them limit exposure and reduce health problems caused by poor air quality. The decision-making process effectiveness in the areas of public health and AQ can be improved by developing accurate deterministic predictions and a reliable quantification of their uncertainty provided by the ensemble forecasts. The analog ensemble (AnEn) technique has been extensively tested for the probabilistic prediction of both meteorological variables and renewable energy. In this study we apply the AnEn technique to probabilistic predictions of O3 and PM2.5 surface concentrations. The deterministic predictions of O3 and PM2.5 concentrations by the Community Multi-scale Air Quality (CMAQ) Chemical Transport Model (CTM) over 564 sites of AIRNow Environmental Protection Agency (EPA) network across the US are used as input to the AnEn. The AnEn is built from a historical set of deterministic predictions and observations of the quantity to be predicted. For each forecast lead time and location, the ensemble prediction of a given variable is constituted by a set of measurements of the past (i.e., 1-hour averages of PM2.5 or O3 concentrations). These measurements are those concurrent to past deterministic predictions for the same lead time and location, chosen based on their similarity to the current forecast. The forecasted variables used to identify the past forecast similar to the current one are called analog predictors. In this application we use as predictors, in addition to a few meteorological variables, the O3 and PM2.5 concentrations forecasts over the continental US generated by the U.S. EPA CMAQ CTM model. The forecasts are issued at 12 UTC with a 24 hours frequency for lead times between 0-48 hours ahead over the period 01 July 2014 - 31 July 2015. The first 6 months of this period are used for training purposes while the remaining part for the verification. Attributes of probabilistic predictions such as statistical consistency, reliability, resolution, sharpness, and the spread-skill relationship are verified. The AnEn provides reliable, sharp, and statistical consistent probabilistic AQ predictions, at a fraction of the real-time computational cost of traditional ensemble methods.