هاب های حمل و نقل هوایی در سیستم فدرال اکسپرس: تجزیه و تحلیل استفاده از سوخت

This paper provides a data based analysis of FedEx air freighter activities from selected hub locations. The basic idea is that air freighters have a set of range and payload parameters and their corresponding fuel burn depends on weight and distance. Data from 2011 to 12 (FlightAware) are used for 180,000 + flights on origin, destination and aircraft type. The particular aircraft vary widely in payload, but additional parameters may be derived from industry web sites and BTS. The research uses flight activity at hubs such as Memphis and Indianapolis (among others) and computes the aggregate distance flown on specific aircraft. The linkage between the hub and aggregate fuel use (assuming that the out bound flights are allocated to the hub) will give some quantifiable measures of the costs allocated to the hub. The paper examines particular aspects of the air freight system that are especially vulnerable to a spike in the costs of aviation fuel. These observations suggest that traffic to regional air express and air freight hubs is likely to respond in complex ways to fuel costs.

Air freight networks of integrators such as FedEx and UPS represent a significant and well-studied component of US and global transport systems. An excellent overview of the challenges of moving freight by air is in Morrell (2011). All-cargo carriers and combination passenger/cargo carriers are also important for freight and mail, but these are not discussed here. The reader is referred to Morrell (2011) for a detailed comparison of these other forms of air freight. In this paper the term air freight is used in a broad sense to refer to the materials carried by the integrators. It combines packages, documents, and larger freight items. A detailed comparison of operational aspects of FedEx and UPS is in Cosmas and Martini (2007). More specifically, Bowen (2012) compares the structure of UPS and FedEx networks, emphasizing the significant role of network organization for these carriers. The macro design of such networks has also been given a lot of attention (O'Kelly and Miller, 1994 and Campbell and O'Kelly, 2012). At a more micro level, such networks solve a complex geographic distribution problem using feeders, spokes, and high volume inter-hub links (Kuby and Gray, 1993). Prior work has proposed models of certain aspects of these systems (Hall, 1989) but there is a need for further detailed examination of hubs in the freight sector, especially with respect to the network's usage of circuitous routes and their intensive use of fuel as an input.1 There is considerable interest from both applied policy and academic modeling perspectives in the efficient use of aircraft and their impact in terms of GHG and other emissions (see World Bank, 2012; Brian et al., 2009, GHG in airports).2 From an operational point of view, air carriers devote a very large fraction of their total costs to jet fuel, and they are vulnerable to uncontrollable variations in these costs. In an effort to minimize costs, freight carriers optimize fleets and plan their flights in the most effective way (see Armacost et al., 2002 and Armacost et al., 2004). They may also pursue other options, such as: equipment changes, modification of route structure, substitution of bio-fuels, and hedging (The World Bank, 2012). This paper considers the role of hubs in air express and air freight transportation, with particular attention to the estimated total fuel cost as well as the share attributable to individual hubs.

Data from FlightAware were used to tabulate the detailed operations and to examine the day-to-day variations of the carrier (FDX). For example, the raw data are essential for examining issues such as peak loads and latest arrivals. It is possible to use the FlightAware data to compute the actual flight time – except for 9547 records with missing data, the departure and arrival times are given, and with careful date and time processing, the elapsed minutes can be computed. Individual day-to-day variability in the elapsed flight time for a specific OD pair and piece of equipment confirms that the FDX operation works within very tight tolerances. The variability in time (on a specific mission) from day to day is quite small for a fixed piece of equipment. It is useful to have access to the underlying detailed daily descriptions because this helps us to know that a city pair has a particular mix of aircraft and that this is the resultant of a day-to-day packaging of flows into a set of suitable aircraft. For example a monthly summary from O to D might suggest 20 F1, and 40 F2 aircraft. Daily data would allow a confirmation of this as a week day daily package of 1 F1 and 2 F2 flights; to be clear other packages with the same aggregate outcome cannot be ruled out with aggregate data alone. T-100 segment data may be used to obtain very similar results; there are four main advantages of this added step. (1) It is possible using T-100 segment data to establish the same data from a widely available source. (2) This computation may be used to check the results from the disaggregate work. (3) Furthermore, this technique has the added advantage that the few instance where rather broad assumptions (load factor, path deviation etc.) in the original data analysis were made can be either refined using more complete empirical data, or can be experimentally adjusted in order to gauge the sensitivity of the results to these assumptions. (4) Finally, although this is not pursued here, the method is reproducible for other time frames and other carriers. As an alternative approach, the fuel calculation was re-done with the departures and the flight time, using fuel burn rate per hour (from P-52 quarterly data). The fuel prediction from the distance based model in this paper is strongly linearly related to the fuel prediction from a time accounting approach. This of course is because the fuel consumption per minute or per mile relate to the same physical problem of lift. The P-52 report gives a larger number: the discrepancy is due to the more complete account in P-52, the fact that the aircraft fly more than the minimal amount of fuel, and that the actual paths may deviate from the straight line or great circle distance by larger factors than assumed here, and of course the fact that T-100 only reports trips with one end in US. Equipment implications: continued reliance on equipment that is superseded by more efficient ones is suboptimal, but understandable in view of the fixed investment and difficulty in making instantaneous adjustments. Solutions may include equipment swap, opportunities to bundle two or three smaller flights into one; opportunities to optimize refueling stops or stage length, and so on. Against this cost scenario too, the carrier has to recognize that the demand for air freight is cost sensitive, and if fuel price surcharges make the service prohibitive, alternative modal arrangements are likely to become desirable (FedEx Annual Report 2013). The option to switch from air to truck or to combine stages with trucking to an air-hub is clearly also very attractive (O'Kelly and Lao, 1991), and related research is continuing to explore that option. This issue is not covered in the present paper but could be the topic of further research. Another interesting “next step” is the kind of focused detailed data presented in Heinitz and Meincke (2013). Their paper recognizes some aspects of a complex real system that are often simplified away in models, and encompass an unusually comprehensive set of all cargo carriers, belly freight, and intermodal connectivity. The wealth of operational detail, unfortunately limits the spatial scope: they represent the interacting entities at a more aggregate 90 zone level for the world (i.e. trade off more realism for lower level of spatial detail). In the other direction, even further disaggregation beyond what is used here is possible. The data could, potentially, provide the detailed flight trajectories of all the system data – allowing for example the detailed individual flight paths. This could be extremely beneficial for an analysis involving controlled descent or for the operational reconstruction of the flight paths (Cosmas and Martini, 2007). The concern is that with these detailed data, we might not be able to “see the woods for the trees” and there are many interesting questions at the slightly more macro level. That topic requires access to more detailed information, and in the future the scope of this work might be expanded to cover these concerns. The ideal would be a constrained model with a fixed fleet of available aircraft that could in turn be gauged as to the financial merit of replacement. This tactic would make sense because in the absence of a constraint the ideal combination of aircraft will simply be to select the ones with the highest efficiency. As shown in this paper, the reality in the results is that there is much more complex balance of desired and available aircraft.