task 3

Task 3 consists in the testing of the optimization model and solution methods developed in Tasks 1 and 2 for real cases studies. The tests will be conducted using high-quality data for airports of different sizes. We will start by implementing the model to small and midsize airports, for which we expect to solve the model using exact methods (and commercial software packages). Then, we will implement the model to increasingly larger airports. At least for some of them, it will in principle be necessary to resort to the approximate methods developed in Task 2.

The small and midsize airports we will use as case studies are Madeira (FNC) and Porto (OPO). In 2016, FNC operated (approximately) 21,000 aircraft movements, while OPO operated 93,000 movements. The latter airport is among the fastest growing airports in Europe. Moreover, it provides a setting that is significantly larger than the airports solved to date using exact solution methods (which all operate less than 50,000 aircraft movements per annum). The large airports we will study are the airports of Lisbon (LIS) and Sao Paulo (GRU). LIS handled 179,000 aircraft movements in 2016, and is currently one of the top 10 European airports with higher average delay per flight. GRU operated 295,000 aircraft movements in that same year. It is the largest and most congested airport in South America (and the 30^th-largest in the world).

The information on slot requests and slot allocation necessary to carry out the case studies will be provided by the slot coordination entities of Portugal (ANA), and Brazil (ANAC), as stated in the support letters attached to this application. Note that, in addition to a wide range of airport sizes, these airports also provide different application settings, given their differences in terms of the number of airlines that operate there, the proportions of historic slots among airline requests, the types of origin-destinations markets served, etc. Therefore, our results can provide insights on the applicability of our model to a broader set of airports worldwide.

The expected results of our computational experiments fall into two categories. First, we will prove the realism of our model by showing that it captures the main trade-offs currently made on an ad hoc basis by the slot coordination entities. Second, we aim to characterize the benefits that the model can provide, by comparing our results to historical slot allocation decisions made in practice, and quantifying the corresponding improvements (e.g., in terms of total number of slots accepted, number of slots displaced from their preferred times, and additional objectives).