Decision Support Tool for Predicting Ground Delay Programs (gdp) and Airport Delays from Weather Forecast Data

During peak periods of operation, the National Airspace System (NAS) suffers from limited resources. Traffic Flow Management (TFM) is responsible for adjusting traffic flow demand to meet the available capacity. In the case of airports, TFM uses the Ground Delay Program (GDP) to allocate the constrained runway resources. This paper describes a tool for forecasting the likelihood of a GDP at an airport and the delays at an airport throughout the day and at specific time periods during the day. The tool, trained on historical data, predicted GDP’s at four OEP airports with accuracy at 73% and delays with accuracy of 76%. The implications of these results are discussed for airlines and TFM. David Smith and Dr. Lance Sherry 2 INTRODUCTION For efficient operation of the NAS, there is a need for the weather forecasting services and TFM products to estimate the reduction in capacity due to adverse weather. Weather forecast products are uncertain and the uncertainty increases with lead-time. Useful applications of weather forecasts requires either refinement, consultation, and application of the weather forecast to estimate air traffic capacity or decision support tools that take forecasts and make predictions based on past forecasts and those forecasts connections to NAS capacity. This paper describes a methodology used to create one such decision support tool known as the Weather Delay Prediction Tool. With this tool, the user enters the TAF for a given day and airport and the tool provides a GDP prediction, delay predictions, and estimated AARs. While weather forecasts are not 100% accurate, meteorologists have the ability to predict weather with increasing accuracy. Through a complex system of weather services, government agencies, pilots, and other aviation professionals receive the vast knowledge base in the form of up to date weather forecasts. These forecasts enable air traffic managers to make informed decisions regarding weather and flight safety. Most weather related research has focused on the improvement of the accuracy of the forecasts and for the pure weather professionals, this is worthwhile research. For the air traffic professional however, research should be focused on how the forecast affects the actual NAS performance. 3 Is there a potential for a Ground Delay Program (GDP)? What will be the average delay? What will be the aircraft arrival rates? It is not whether it rains which is important, but instead did the forecast make you bring your umbrella? These questions can be answered by developing a model using a pattern recognition tool known as a support vector machine (SVM). The SVM is trained with past data and then develops functions to enter present day data to predict future outcomes. The SVM is superior to other techniques such as regression because it does not require a linear relationship between the data. This way we can use individual data points with few possible choices and still come up with predictive functions. As seen in Figure 1, we collect TAF data, apply the SVM, produced GDP and delay predictions, and then apply queuing theory to predict AARs. David Smith and Dr. Lance Sherry 3 DATA AND METHODOLOGY The Terminal Aerodrome Forecast A variety of different forecasts products are produced and designed to be used in the preflight planning stage. One of these forecasts that traffic managers should be familiar with is the Terminal Aerodrome Forecast (TAF). A TAF is a report established for the 5 statute mile radius around an airport. Each TAF is valid for a 24-hour time period and is updated four times a day at 0000Z, 0600Z, 1200Z, and 1800Z. The TAF is reported in a standard international format code and must follow a set of rules that define what must be placed in each line and what criterion requires a new line. There are four different lines in a TAF, the first one gives location, valid time and given weather for that time until the next line of the forecast. The BECMG (becoming) line indicates between the two hours given, the weather will start to change from the previous line to the next line; while a FM (from) line indicates that after the given time, the weather will be what the line states; and the TEMPO line indicates that for time period given, the weather parameters in the line will happen for no more than 30 minutes of the hour, but will happen for at least 1 minute of the hour. This code may seem confusing to the casual viewer, however, this code allows a computer program to easily search for and identify certain weather conditions and the time that these conditions are forecasted. For this paper we collected all available archived TAF data for Newark, O'Hare, Atlanta, and Philadelphia airports. This data is only available from January 2002 to April 2007. Each set of TAF data was loaded into an Excel spreadsheet. Within the spreadsheet we programmed a macro to find, for four time periods during the day, the wind speed and direction, visibility, and ceiling conditions. For the same time periods, FIGURE 1 SVM Method Estimated AARs Queuing Theory