Measuring Uncertainty in Airspace Demand Predictions for Traffic Flow Management Applications

Abstract

Traffic flow management (TFM) in the U.S. is the process by which the Federal Aviation Administration (FAA), with the participation of airspace users, seeks to balance the capacity of airspace and airport resources with the demand for these resources. This is a difficult process, complicated by the presence of severe weather or unusually high demand. TFM in en-route airspace is concerned with managing airspace demand, specifically the number of flights handled by air traffic control (ATC) sectors; a sector is the volume of airspace managed by an air traffic controller or controller team. Therefore, effective decision-making requires accurate sector demand predictions. While it is commonly accepted that the sector demand predictions used by current and proposed TFM decision support systems contain significant uncertainty, this uncertainty is typically not quantified or taken into account in any meaningful way. The work described here is focused on measuring the uncertainty in sector demand predictions under current operational conditions, and on applying those measurements towards improving the performance and human factors of TFM decision support systems.

Preliminary investigation of trajectory prediction impact on decision support automation

“…Not only may multiple DSTs affect each other, [21] but they may be using different trajectory predictors, or a trajectory predictor in different ways (e.g., open loop to obtain an estimation and close loop to match a trajectory constraint). Some DSTs may operate at higher, more strategic levels, and there can be a cascading effect as the TP errors propagate from the tactical to the strategic levels, [22][23] and finally to the overall system performance. [24] Like TP performance, DST performance can be measured using a variety of metrics.…”

Performance Impact of Improved Departure Time Prediction Relative to Sector Demand & Arrival Time Predictability

“…These two error factors are ranked highest in terms of their influence on sector demand prediction based on interviews with air traffic controllers and researches. [12] identified and analyzed three variables that have the strongest effects on uncertainty distributions; (1) look-ahead time, (2) prediction peak count and (3) sector traffic type. [11] does not rank uncertainty sources but investigates prediction performance by weather type, flight plan submission, regulation and flight type.…”

“…[11] does not rank uncertainty sources but investigates prediction performance by weather type, flight plan submission, regulation and flight type. Reference Sample time Sample airspace Analyzed prediction performance metrics [12] 286 days 754 U.S. sectors peak count [9] 30 days Spanish FIR sector entry time [10] 5 days 2 U.S. ACC's sector occupancy count, sector entry time [11] 4 days 2 U.S. ACC's sector entry time, hit rate Table 2.1 shows sample time, sample airspace, and studied prediction performance metrics used in the articles that are discussed in this section. Take note that prediction error for the reviewed papers is calculated by the "actual time" minus "predicted time".…”

“…A distinction is made between four traffic categories based on the predominant type of traffic passing through the sector; (A) arrival, (D) departure, (E) en-route and (M) mixed [12]. The considered traffic consists of proposed and active flights.…”

Probabilistic Demand Prediction Model for En-Route Sector

“…According to the uncertain influence of the traditional trajectory prediction to the flow prediction, Sandip [10] points out that the uncertainty and complexity is inherent in the traffic flow prediction, and established the aggregated dynamic stochastic model based on the Poisson distribution. Wanke [11], [12] researches the traffic demand uncertainty in sectors, and analyzes the main factors affecting the demand prediction. Chatterji [13] measures the uncertainty of the sector traffic demand prediction based on the stochastic departure time.…”