Skip to main content
Open Access Publications from the University of California

UC Berkeley

UC Berkeley Electronic Theses and Dissertations bannerUC Berkeley

Quantifying the Impact of Flight Predictability on Strategic and Operational Airline Decisions


In this thesis, we examine how the predictability of travel time affects both the transportation service providers’ strategic and operational decisions, in the context of air transportation. Towards this end, we make three main contributions. The first is the development of accurately measuring predictability of travel time in air transportation to best model airline decision behavior. The measure is sensitive to the different nature that’s driving the decision. The second is an empirical investigation of the relationship between the best-measured travel time predictability and the transportation service providers’ strategic and operational decisions to gain insights into the significance of the impact of predictability. The third contribution is proposing an algorithm to improve predictability in order to save cost in the strategic decision process through re-sequencing the departure queue at the airport.

We consider the strategic decision as the setting of the scheduled travel time for each trip that typically happened six months before the travel date. On the operational side, we investigate into the decision of the amount of fuel loaded to each flight in the daily operation. We assume that the decisions are based on the predictability of historical travel time performance. When quantifying predictability, it is important to realize that the service providers have different priority of considerations when making the strategic (scheduling) and the operational (fuel loading) decisions. Therefore, we apply different metrics for predictability to modeling the different decision behaviors and prove that the best-fitting measure of predictability is not uniform across different type of decisions. Regarding the strategic decision making, the profit-driven nature of the service provider encourages discounting the effects of extremely long historical travel times. Therefore, segmenting the historical travel time distribution is crucial in our effort of measuring predictability. On the other hand, when making day-to-day operational decisions, specifically fuel loading decisions in this study, the safety-driven nature of the service provider prevails over others and it pays more attention to extreme events. Therefore a metric capturing the tail effects such as the variance and standard deviation is a more appropriate measure of predictability in this context.

In modeling the relationship between predictability and scheduled travel time setting, we seek both analytical insights and empirical evidences. Firstly this relationship is studied with empirical data and multiple regression models. We develop the “percentile model” where the distribution of the historical travel time for an air trip is depicted by the difference between every 10th percentiles. We find that gate delay plays a minor role in setting scheduled travel time and that scheduled travel times have decreasing sensitivity to historical travel times toward the right tail of the distribution. To specifically link schedule setting with the trip’s on-time performance, a scheduled travel time adjustment model is further developed. Poor on-time performance leads to increased scheduled travel time in the next planning period. With the behavior model results showing that both the median travel time and the “inner right tail” of the distribution affect schedule setting, an impact study is conducted to validate these impacts with evidence in the historical data. This impact from behavioral modeling is validated with real data in year 2006-2008 and 2009-2011, and their corresponding scheduled travel times in the later period. Furthermore, by studying the travel performance difference based on different changes in scheduled travel time, we conclude that ignoring the impact on schedule changes when considering potential benefits of improved travel time distribution could lead to inaccurate results.

We complement the strategic behavioral modeling findings with proposing a practical algorithm that optimizes the sequence of departure queue on the airport to improve travel time predictability. The end objective is to reduce scheduled travel time through improved predictability and thus save cost for travel service providers. We present algorithms to sequence departures on a daily basis. For the objective function, scheduled travel time is viewed as a cost for airlines to be minimized. For each flight, the assigned slot generates a new travel time and this time contributes proportionally to the future scheduled travel time, as revealed in estimating the “percentile model”. Assuring that the on-time performance is not greatly sacrificed is also important. Therefore the objective function also includes delaying the flight’s arrival performance as part of the “cost of assignment”. In this way, we develop a multi-objective algorithm to sequence departure flights to improve predictability, reduce airline scheduled travel time, and increase on-time performance.

To investigate the relationship between predictability and fuel loading decisions, we develop a set of multiple regression models considering clusters of standard deviation of the estimates. The unpredictability under performance may cause decision makers (airline dispatchers) to load more fuel onto aircraft, and thus causing extra fuel consumed to carry this excessive fuel. We acquired a large and recent dataset with flight-level fuel loading and consumption information from a major US airline. With this data, firstly the relationship between the amount of loaded fuel and travel time predictability performance is estimated using statistical model. Predictability is measured with metrics such as standard deviation of travel time so that the tail effect of the distribution is properly captured. We find that one minute of standard deviation in airborne time within a month for the same OD pair and shift of day would lead to 0.95 minute increase in loaded contingency fuel and 1.85 minute loaded contingency and alternate fuel. Then, the impact of predictability on loaded fuel is translated into fuel consumption and ultimately, fuel cost for US domestic operations. If there is no unpredictability in the aviation system, the reduction in the loaded fuel would be 6.4 and 12.5 minute per trip, respectively. This ultimately translates into a cost to US domestic air carriers on the order of $88 – $345 million per year.

Main Content
For improved accessibility of PDF content, download the file to your device.
Current View