3.2 Predicting Weather Impact on Airspace Flow Using Multiple Forecasts

Tuesday, 12 January 2016: 8:15 AM
Room 354 ( New Orleans Ernest N. Morial Convention Center)
Mark S. Veillette, MIT Lincoln Laboratory, Lexington, MA; and M. Matthews, H. Iskenderian, and R. DeLaura

A key decision that Traffic Flow Managers (TFMs) need to make when confronted with the potential loss of airspace due to convective weather is whether or not to throttle flow into and out of impacted airspace. This throttling of flow is accomplished by initiating Airspace Flow Programs (AFPs), which establish flow rates for aircraft passing through Flow Constrained Areas (FCAs). AFPs need to be planned several hours in advance of the impacting weather in order to be effective, and the planning requires the use of weather forecasts that are for 3 to 8 hours into the future. A variety of forecasts are available to TFMs at these lead times and they each contain a considerable amount of uncertainty. This makes it difficult for users to know how much weight to give to each forecast in their decision making process.

This talk will describe a methodology for combining multiple forecasts of different types (e.g. deterministic and probabilistic forecasts) into a single forecast of airspace capacity, as well as providing a dynamic measure of forecast confidence. To begin, relevant features are extracted from individual weather forecasts and are used to translate the forecasts into an ensemble of predictions for airspace impact. Machine learning techniques are then used to combine this ensemble of translated forecasts into a posterior distribution of impact for up to 12 hours into the future. An impact forecast and its corresponding uncertainty are then computed from these posterior distributions. Four forecast models are used in this work: the FAA's Corridor Integrated Weather System (CIWS) extrapolation forecast, NOAA's High Resolution Rapid Refresh (HRRR) model, NOAA's Short Range Ensemble Forecast (SREF) Calibrated Thunderstorm Probability, and NOAA's Localized Aviation MOS Product (LAMP). This presentation will discuss the machine learning techniques used combine output from these different types of forecast models to provide a measure of weather impact on airspace and its uncertainty, and the benefits to the aviation airspace decision process.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner