Quantile Regression-Based Estimation of Dynamic Statistical Contingency Fuel

Reducing fuel consumption is a unifying goal across the aviation industry. One fuel-saving opportunity for airlines is reducing contingency fuel loading by dispatchers. Many airlines' flight planning systems (FPSs) provide recommended contingency fuel for dispatchers in the form of statistical contingency fuel (SCF). However, because of limitations of the current SCF estimation procedure, the application of SCF is limited. In this study, we propose to use quantile regression-based machine learning methods to account for fuel burn uncertainties and estimate more reliable SCF values. Utilizing a large fuel burn data set from a major U.S.-based airline, we find that the proposed quantile regression method outperforms the airline's FPS. The benefit of applying the improved SCF models is estimated to be in the range $19 million-$65 million in fuel expense savings as well as 132 million-451 million kilograms of carbon dioxide emission reductions per year, with the lower savings being realized even while maintaining the current, extremely low risk of tapping the reserve fuel. The proposed models can also be used to predict benefits from reduced fuel loading enabled by increasing system predictability, for example, with improved air traffic management.