Report no

Acknowledgments The authors would like to thank Midori Tanino and Ved Sud of the FAA for their support and encouragement. Within the Volpe Center, we would like to thank Dick Bair, Rick Oiesen and Ken Howard for reviewing the report and their valuable comments and suggestions that improved its content. We also thank Mary Costello for editing this report. The Enhanced Traffic Management System (ETMS) predicts traffic demand in the National Airspace System (NAS) up to 24 hours in the future to determine potential congestion in airspace or airports. To identify congestion, it finds time intervals and NAS elements (i.e., sectors, airports, and fixes) where predicted demand exceeds the capacity that has been input into ETMS. Based on the duration and magnitude of congestion, traffic flow management (TFM) specialists decide whether to take action to bring traffic demand down to capacity through various traffic management initiatives (TMIs), such as Ground Delay Programs (GDPs), Airspace Flow Programs (AFPs), or Miles-in-Trail (MIT). ETMS produces deterministic predictions of traffic demand and does not take into account the random errors in these predictions. This uncertainty in predictions creates uncertainty in the information that TFM specialists use in their decision-making process. A recent direction in TFM research is concerned with acknowledging the uncertainty in predictions and creating probabilistic TFM that considers the uncertainty in the decision-making process. The premise behind probabilistic TFM is that traffic managers will make better decisions if they use data and tools that reflect the uncertainty in the system. Probabilistic TFM is based on probabilistic representation of traffic demand and capacity of NAS elements through the respective probability distributions that allow for determining the probabilities of congestion. In order to obtain the probability distributions for probabilistic TFM, a thorough statistical analysis of prediction errors is needed to characterize the prediction uncertainties. This report presents the results of analysis and characterization of uncertainty in traffic demand predictions using ETMS data and probabilistic representation of the predictions. Our previous research, described in two prior reports, was focused on analysis of aggregate 15-minute traffic demand predictions in ETMS, on improving the accuracy of these predictions and increasing the stability of the ETMS monitor/alert function, while not explicitly considering the uncertainty in predictions of flight events for individual flights. This study continues the previous one. It also focuses on uncertainty in traffic demand predictions, but, unlike the previous one, it explicitly considers uncertainty in individual …