Optimizing Aircraft Trajectories with Multiple Cruise

Altitudes in the Presence of Winds

Hok K. Ng1
University of California, Santa Cruz, Moffett Field, CA 94035
Banavar Sridhar2 and Shon Grabbe3
NASA Ames Research Center, Moffett Field, CA 94035

This study develops a trajectory optimization algorithm for approximately minimizing

aircraft travel time and fuel burn by combining a method for computing minimum-time
routes in winds on multiple horizontal planes, and an aircraft fuel burn model for generating
fuel-optimal vertical profiles. It is applied to assess the potential benefits of flying userpreferred routes for commercial cargo flights operating between Anchorage, Alaska and
major airports in Asia and the contiguous United States. Flying wind optimal trajectories
with a fuel-optimal vertical profile reduces average fuel burn of international flights cruising
at a single altitude by 1-3%. The potential fuel savings of performing en-route step climbs
are not significant for many shorter domestic cargo flights that have only one step climb.
Wind-optimal trajectories reduce fuel burn and travel time relative to the flight plan route
by up to 3% for the domestic cargo flights. However, for trans-oceanic traffic, the fuel burn
savings could be as much as 10%. The actual savings in operations will vary from the
simulation results due to differences in the aircraft models and user defined cost indices. In
general, the savings are proportional to trip length, and depend on the en-route wind
conditions and aircraft types.

I.

Introduction

dvanced air traffic management systems are designed to optimize the throughput and efficiency of the national
airspace system in the United States to accommodate growth in air traffic. Developing optimal aircraft

trajectories not only enhances air traffic flow but also helps the aviation industry cope with increasing fuel costs and
1

Research Scientist, University Affiliated Research Center, Mail Stop 210-8, Member AIAA.
Senior Scientist for Air Transportation Systems, Aviation Systems Division, Fellow AIAA.
3
Research Scientist, Systems Modeling and Optimization Branch, Mail Stop 210-10, Associate Fellow AIAA.
2

helps to reduce aviation-induced climate change. Classical aircraft trajectory optimizations are solved by applying
calculus of variations to determine the optimality conditions, requiring the solution of non-linear Two-Point
Boundary Value Problems (TPBVPs) [1]. In order to reduce the modeling order and hence the complexity, the
energy state approximation has been successfully applied to generate an optimal vertical flight profile [2,3].
Alternatively, a more general solution to aircraft trajectory optimization can be obtained by singular perturbation
theory which approximates solutions of high order problems by the solution of a series of lower order systems with
the system dynamics separated into low and fast modes [4,5]. However, finding the global optimal solutions to the
TPBVPs remains a daunting task.
With the tremendous advancement in numerical computing power, TPBVPs can be converted to nonlinear
programming problems that are solvable even for problems with many variables and constraints using numerical
algorithms such as direct collocation methods. Neglecting aircraft dynamics and applying shortest path algorithms in
graph theory, an optimal trajectory can be approximated by the path that minimizes the total link cost connecting the
origin and destination in a pre-defined network. The graph methods often require large computation time and
memory space but guarantee global optimal solutions. These numerical methods are widely applied for problems in
constrained airspace such as minimum cost-to-climb and obstacle-avoidance.

However, the majority of the

aforementioned trajectory optimizations neglect the effect of wind in the problem formulation.
The following studies include wind effects. In 1931 Ernst Zermelo [6] first proposed the minimum-time path
problem for a boat moving through a region of strong currents. Bryson and Ho approached the problem by
determining the analytical equation governing the optimal heading dynamics that satisfies Pontryagins Minimum
Principle [1]. This results in a solution to the minimum-time path for aircraft travel at a constant altitude and speed
in strong winds given correct choice of the initial heading. Generally, the initial heading angle can be obtained by an
iterative process but convergence problems may occur due to the presence of nonlinearities such as jet streams. The
search could converge to a local minimum depending on the initial choice. Recent studies propose selecting the
initial heading by combining calculus of variations and graph theory, and claim that these methods always generate
global optimal solutions with moderate computational effort [7,8]. These studies are focused on computing the
minimum-time path at a single flight level and have not considered the minimum-fuel route that is divided into
segments at several flight levels, identified by incorporating aircraft weight and fuel burn modeling into the
optimization procedures.

This study develops a practical trajectory optimization algorithm for aircraft that approximately minimizes cost
of time and fuel burn by combining the method for computing minimum-time routes in winds on multiple horizontal
planes and the aircraft fuel burn model for generating fuel-optimal vertical profiles. It is applied to evaluate the
potential benefits of flying wind-optimal routes in a seamless airspace for commercial cargo flights operating
between Anchorage, AK and major airports in Asia and the contiguous United States.
The section on Practical Aircraft Trajectory Optimization describes total operating cost, and develops the vertical
profile and horizontal maneuver that minimize fuel burn and travel time in winds for commercial aircraft. The
section on Experimental Setup applies the optimization algorithm to calculate wind-optimal trajectories for
commercial cargo flights between Anchorage, AK to major cities in the U.S. and Asia. The Results section presents
potential benefits of flying minimum-time trajectories in winds with a fuel optimal vertical profile. The Conclusions
section summarizes the results.

II.

Practical Aircraft Trajectory Optimization

This section presents the practical trajectory optimization algorithm that approximates the minimization of the
total cost of travel time and fuel consumption for aircraft on a spherical surface. Optimizing performance of a single
aircraft from takeoff to landing in the presence of air traffic constraints and winds is computationally intensive and
time consuming. Future air traffic simulation systems will be required to optimize and simulate tens of thousands of
aircraft trajectories in seconds or minutes. However, it is impractical and may not be necessary for these systems to
generate thousands of optimal and detailed aircraft trajectories. The details of the trajectory optimization required
depend on the performance index, the approach and the application.
This study adapts a practical optimization approach by assuming a typical structure for an aircraft trajectory and
focuses on optimizing direct operating cost during cruise when the time and fuel savings have the most impact. A
typical aircraft trajectory consists of an initial climb, a steady-state cruise, and a final descent. Here, aircraft
performance is optimized for the cruise phase only. Typical aircraft profiles are applied for generating trajectories
during initial climb and final descent since additional air traffic constraints are involved in these stages and travel
time and fuel savings are generally small compared to those during cruise. The cruise trajectory is divided into
segments on several altitudes as the optimal cruise altitude increases due to the reduction in aircraft weight as fuel is
used. The legal cruising altitudes and the en-route step climb times are optimized based on Eurocontrols Base of
Aircraft Data, Revision 3.6 model [9]. The aircraft optimal heading during cruise is the solution of the Zermelo

problem derived on a spherical Earth surface in the absence of constraints. The horizontal trajectory segments are
optimized based on the cost-to-go associated with extremals (trajectories) generated by forward or backward
integrating the dynamical equations for optimal heading and aircraft motion from various points in the airspace. This
computationally efficient algorithm searches for optimal solutions by combining calculus of variations and dynamic
programming.
The next subsection defines aircraft operating cost and outlines the practical optimization procedures. The
optimization procedures include two stages. The subsection on Optimal Vertical Profile describes solving the
optimal aircraft cruise altitudes and en-route step climb times. The subsection on Horizontal Trajectory Generation
presents the aircraft model and the procedures for calculating optimal aircraft headings in winds on multiple flight
levels.
A. Aircraft Operating Cost
The direct operating cost for a cruising aircraft can be written as:

J=

tf
t0

[Ct + C f f (m,h,V )]dt,

(1)

where Ct and C f are the cost coefficients of time and fuel. The fuel flow rate, f, can be approximated by a function

of aircraft mass, m, altitude, h, and airspeed, V. Aggarwal [10] minimizes direct operating costs for long-range

transport
type aircraft in the absence of winds, and shows that the aircraft Mach number remains almost constant
during cruise. This result is adopted here to simplify the search for the optimal solution.
For a chosen constant airspeed indexed in i, Vi , the cost function is rewritten as:

J =
i

tf
t0

[Ct + C f f (m,h,Vi )]dt.

(2)

The optimal Vi can be determined through an iterative process such that the arrival time constraint is met and

operating cost is minimized. This operating cost depends on the total travel time and fuel flow rate during cruise.
Total
travel time is minimized by determining the optimal aircraft heading that yields the minimum-time trajectory
in winds and the fuel flow rate is minimized by controlling cruise altitudes given m and Vi . The computational
effort is high, especially for long haul flights, when both winds and aircraft fuel burn are considered in the
optimization since en-route meteorological conditions vary with location, and
aircraft fuel burn depends on altitude.

In order to improve the computational efficiency of the approach, a two-stage optimization strategy is proposed
for narrowing down the search scope to find approximate solution to the optimization problem. The first stage
optimizes an aircraft vertical profile along the track to determine the optimal cruise altitudes and en-route step climb
times given the flight levels constraints. The second stage develops wind-optimal horizontal trajectory segments by
determining optimal aircraft heading in winds based on the cruise altitudes and step climb times that are solved in
the first stage.
B. Optimal Vertical Profile
This subsection develops an aircraft vertical profile based on the fuel consumption model in Eurocontrols Base
of Aircraft Data Revision 3.6 (BADA) [9]. First, formulas for calculating the cruise altitude that has the minimum
fuel burn rate are derived by assuming a constant cruise speed. Then, they are applied for optimizing the aircraft
vertical profile along track based on the aircraft weight and true airspeed.
The fuel burn for aircraft during cruise, F, is calculated as:

F = t f,

(3)

where t is elapsed time. The fuel burn rate, f , for jets and turboprops is determined by the specific fuel

consumption, SFC , and thrust, Th, and can be expressed in the following form:

f=

C fcr
1000

SFC Th,

SFC = C f 1 (1+

V
).
Cf 2

(4)

C fcr ,C f 1 , C f 2 are the thrust specific fuel consumption coefficients and VTAS is the true air speed. During cruise,

thrust equals aerodynamic drag force, D, that is a function of aerodynamic drag coefficients, CD , air density, ,

as:
aircraft wing reference area, S , and true air speed, and is expressed

1
D = CD SV 2 .
2

(5)

The drag coefficient, under normal conditions and except during approach and landing, is specified as a function
of the drag coefficient parameters, CD 0, CD 2 , and the lift coefficient, CL , where:

CD = CD 0 + CD 2CL2 ,
CL =

2mg
.
SV 2

(6)

Note that the aerodynamic drag is a function of air density that depends on altitude.
The formulas for calculating the optimal cruise altitude that has the minimum fuel flow rate are derived by
taking the first order variation of the fuel flow rate with respect to the cruise altitude to zero. Under the International
Standard Atmosphere (ISA) conditions, the tropopause is at 11,000 meters altitude; and the optimal aircraft cruise
altitude, hopt , at or below the tropopause can be calculated by the following formula:

f (m,V )K T R gas

hopt = [1 e

2(g +K T R gas ) 0_ ISA

](

1000T0 _ ISA
).
6.5

(7)

Above the tropopause, it is:

hopt =

where f (m,V ) = ln(

f (m,V ) RgasTtrop _ ISA

2g trop _ ISA

+11000,
(8)

4m 2 g 2CD 2
). Rgas is the real gas constant for air. The temperature gradient, KT , the sea level
S 2V 4
CD 0

air density, 0 _ ISA and the sea level temperature, T0 _ ISA , are all constant under ISA. The air density, trop _ ISA , and

temperature, Ttrop _ ISA , are also constants at the tropopause. The appendix shows the derivation of these formulas.

and aerodynamic drag

Optimal cruise altitudes are computed from Eqs. (7,8) based on the atmospheric constants

coefficients
that are aircraft type dependent. They also vary with aircraft mass and air speed. The thrust margin is

also checked to ensure that it is positive when determining the optimal altitudes for each selection of mass and
speed. The thrust margin is defined as the maximum cruise thrust minus the thrust required to maintain 100
ft/minute rate of climb at the selected air speed, at the optimal cruise altitude [11].
Figure 1 shows the optimal cruise altitudes for various combinations of mass and speed for the McDonnell
Douglas MD-11 aircraft. In general, optimal cruise altitudes are lower for higher weights and increase as air speed
increases. Note that optimal cruise altitude decreases at very high air speeds when the thrust margins become
negative. In these cases, optimal altitudes are recomputed and lowered to meet the thrust margin requirement.

0
00

260

00

33

250

00
50

0
00

220

210
200
450

37000

00
37

00

35

35000

37

240
230

35

0
70

39

0
00

00

39

00
00
410
390
460
470
480
490
True Airspeed, knots

41000

Aircraft Mass, 1000kg

270

500

Fig. 1 Optimal cruise altitudes for MD11

Altitude, 1000feet

39
Optimal
Operational

38
37
36
35
34
0

100
200
Time, minutes

300

Fig. 2 Vertical profile during cruise for MD11.

During a flight, the optimal cruise altitude increases due to the continuous reduction in aircraft weight as fuel is
used. An aircraft weight profile during cruise, for a constant airspeed, is generated using the initial aircraft weight at
the start of cruise and the time history of the fuel burn rate. Figure 2 plots the vertical profile for optimal cruise
altitudes in blue based on the weight profile in the absence of winds. Due to current air traffic regulation, however,
aircraft are not permitted to follow this optimal climb profile, and can instead only cruise on certain flight levels
depending on the direction of flights. This study adapts the standard in Reduced Vertical Separation Minima
(RVSM) that eastbound aircraft fly odd thousands of feet while westbound aircraft fly even thousands of feet. The
Legal
optimal legal cruise altitudes, hopt
, are determined given the set of flyable altitudes, h Legal , using the following

equation:

Legal
hopt
(t) = min
hopt (t) h Legal .
Legal
h

(9)

The vertical profile for an eastbound flight is plotted in magenta in Fig. 2. This vertical profile is generated
without winds and assumes an instantaneous step climb. It is an approximation for the vertical aircraft profile for
determining the en-route legal cruise altitudes and the total travel time at each altitude.
C. Horizontal Trajectory Generation
The second stage of the proposed optimization strategy optimizes aircraft horizontal trajectory segments in the
presence of winds based on the vertical profile developed in the first stage. Minimizing total aircraft travel time also
minimizes total fuel burn for a given vertical profile since the fuel burn rates are specified along the track. The
minimum-time trajectory consists of the horizontal trajectory segments on the pre-determined legal cruising
altitudes. They are optimized based on the cost-to-go associated with each extremal generated by forward or
backward integrating the dynamical equations for optimal heading and aircraft motion from various points in the
airspace. This subsection presents the dynamical equation for optimal aircraft heading and the horizontal trajectory
optimization algorithm.
The aircraft equations of motion at a constant altitude above the spherical Earths surface are:

V cos + u( , ,h)
=
,
Rcos

(10)

V sin + v( , ,h)
=
,
R

(11)

m = f ,

(12)

subject to the conditions that thrust equals drag, flight path angle is zero, and the boundary constraints are met. is

longitude and is latitude, is heading angle, and R is the Earths radius. The east-component of the wind

velocity is u( , ,h) , and the north-component of the wind velocity is v( , ,h) . It is assumed that the Earth is a

sphere and R >> h.

The previous study [12] applies Pontryagins Minimum Principle

to determine the heading angle that is assumed

tobe the control available for aircraft during cruise for minimizing the cost of travel time and climate impact. When
neglecting the climate impact, the horizontal trajectory is optimized by determining the heading angle that
minimizes travel time in the presence of winds. The dynamical equation for the optimal aircraft heading [12] is:

F ( , , ,u,v,V )
= wind
,
Rcos

(13)

where Fwind ( , , ,u,v,V ) is aircraft heading dynamics in response to winds and is expressed as:

Fwind ( , , ,u,v,V )

u( , ,h)
u( , ,h) v( , ,h)
+ cos 2 sin u( , ,h) + cos 2 cos

(14)

v( , ,h)
v( , ,h)
+ sin cos sin v( , ,h) + cos sin cos
+V cos sin + cos 2
].

= [sin cos

A recent study [8] developed the same dynamical equation for computing minimum-time path. These studies [8,

12] reduced the trajectory optimization problem from a two-point boundary value problem to an initial value
problem. Numerical algorithms such as collocation methods, used in [12], or interpolation techniques, as used in [8],
can be applied to determine the optimal initial aircraft heading. In the case that an aircraft cruises at a single altitude,
the minimum-time trajectory is completely specified by integrating Eqs.(10, 11, 13) simultaneously from the origin
to the destination using the optimal initial aircraft heading.
In general, an aircraft cruises at multiple altitudes to accommodate weight reduction, thereby minimizing fuel
burn, as described in the previous subsection. The minimum-time trajectory is the combination of wind-optimal
extremals on several different altitudes, each solved using wind conditions on that altitude. This study employs the
concept of dynamic programming for constructing the minimum-time trajectory with wind-optimal extremals at
different cruise altitudes. The optimal vertical profile provides the initial and subsequent optimal cruise altitudes as
well as the transition times between the altitudes. The following five steps compute the horizontal minimum-time
trajectory.
1.

Using a range of different initial heading angles at the start of the initial cruise segment, a collection of
wind-optimal extremals is generated by forward integrating Eqs. (10, 11, and 13) until the first step
climb time. This step is illustrated by plotting the wind-optimal extremals in blue in Fig. 3 for a flight
from Anchorage to Hong Kong.

Fig. 3 Wind-optimal extremals from Anchorage, AK and Hong Kong

2. Using a range of different final heading angles at the destination, another collection of wind-optimal
extremals is generated by backward integrating Eqs. (10, 11, and 13) at the second cruise altitude. This
is done for a fixed period of time, estimated based on the difference between the total travel time and
the first step climb time. The Magenta curves in Fig. 3 shows these extremals at the second cruise
altitude. These extremals provide the minimum time-to-go to the destination from any points along
them. Delaunay triangulation and interpolation techniques can be applied to estimate the time-to-go
from any point in the covered airspace region using the points on the extremals.
3.

Identify the extremal from the collection generated in step 1 above that has the minimum time-to-go at
the first step climb time (i.e. the end of the extremals). This step determines the wind-optimal trajectory
segment at the first cruise altitude and the aircraft position at the first step climb time. The trajectory
segment and the step climb position are shown by the blue dashed line and the blue cross in Fig. 3. Note
that the minimum time-to-go calculated based on the extremals in step 2 is only an approximation when
the aircraft cruises at more than two altitudes.

4.

After climbing to the new cruise altitude, repeat step 1 and step 2 at the new starting position and
altitude. This is demonstrated in Fig. 4, which plots the new starting position as a blue cross and the
next step climb position as a magenta cross. The extremals and the minimum-time trajectory segment

at the new altitude are plotted as solid and dashed magenta lines. This step is repeated until the last
cruise altitude is reached.

Fig. 4 Wind-optimal extremals on 2 altitudes

5.

When the last cruise altitude is reached, the minimum-time trajectory at the last altitude is calculated by
forward integrating Eqs. (10, 11, and 13) using the correct heading angle at the start of the last cruise
altitude. The heading angle is calculated by interpolating the aircraft headings from the backward
extremals at the final cruise altitude.

The horizontal wind-optimal trajectory for a flight from Anchorage, AK to Hong Kong on August 1, 2010 is
shown in Fig. 5. The wind vectors at flight levels 360, 380 and 400 are plotted in blue, magenta and green,
respectively. A flight level is a standard altitude of an aircraft in hundreds of feet. Each trajectory segment is
optimized with respect to the winds at the associated flight level.
The wind magnitudes and directions are taken from the Global Forecast System (GFS). GFS is a global
numerical weather prediction computer model run by the National Oceanic & Atmospheric Administration (NOAA)
four times a day. It produces forecasts up to 16 days, and produces a forecast for every 3rd hour for the first 180
hours, and after that, every 12 hours. The horizontal resolution is roughly equivalent to 0.50.5 degree
latitude/longitude. GFS data has 64 unequally-spaced vertical isobaric pressure levels ranging between 0.25-1000
mb, with enhanced resolution at low and high altitude.

Fig. 5 The wind-optimal trajectory from Anchorage, AK to Hong Kong

The computation time for solving the wind-optimal trajectory on 3 flight levels is about 8 seconds when running
a Matlab R2012a program on a Mac Pro with dual 2.66 GHz 6-core processors and 64 GB memory.

The

computation time for solving a wind-optimal trajectory at a single flight level is about 6 seconds on the same
platform. This algorithm can be implemented conveniently in other computer languages since it does not require
complex numerical solvers. Note that generating the wind-optimal extremals in parallel can further reduces the
computation time.
The performances of the two wind-optimal trajectories are evaluated by simulating the complete aircraft
trajectories from takeoff to landing for a Boeing 777-200 with a takeoff weight of 258,300kg and a cruise speed of
486knots using the wind-optimal flight paths. The typical aircraft profiles during takeoff, step climbs, and landing
are based on BADA [9]. The horizontal trajectory during cruise is simulated based on the wind-optimal flight path
at a single flight level and the wind-optimal flight path on multiple flight levels with en-route step climbs,
respectively. The total fuel consumption and travel time for the wind-optimal flight on a single flight level is 64.4
tonnes and 569 minutes, respectively. The wind-optimal flight that cruises on 3 flight levels burns 62.4 tonnes of

fuel and travels for 562 minutes. Flying wind-optimal with en-route step climbs saves about 3.2% of fuel and 1.2%
of travel time when compared to that on a single flight level.
This practical approach integrates the technique of aircraft wind-optimal heading and the concept of Dynamic
Programming (DP) to approximately optimize the performance of aircraft trajectories in the presence of winds. The
details of deriving the dynamical equation for aircraft wind-optimal heading can be found in the past studies [8, 12].
The application of DP for determining the subsequent step climb locations is discussed in the Appendix.

III.

Experimental Setups

The practical trajectory optimization algorithm is applied to assess the potential benefits of flying wind-optimal
trajectories with an optimal vertical profile for commercial cargo flights operating at Anchorage, AK. The trajectory
computations use air traffic and global wind data from October 2010, obtained from the GFS. The performance of
the wind-optimal trajectories is evaluated by simulating the complete aircraft trajectories from takeoff to landing
using the wind-optimal flight paths. The typical aircraft profiles during takeoff, step climbs, and landing are based
on BADA.
Ted Stevens Anchorage International Airport (ANC) is a major hub airport for cargo flights between the U.S.
and Asia, being less than 9.5 hours from 90% of the industrialized world. For this paper we analyze nearly 12,500
Federal Express (FedEx) and United Parcel Service (UPS) cargo flights from 2010. Based on these data, Fig. 6
shows the 10 most popular origin airports for flights to ANC. Almost 80% of inbound FedEx and UPS flights depart
from these 10 origin airports. Over 90% of outbound FedEx and UPS flights from ANC fly to the 10 destination
airports shown in Fig. 7. Table 1 lists the airport names and codes for the top origins and destinations. RJAA and
KIND is the top origin and destination, respectively, in additional to the nine common airports. About 99% of the
aircraft in the data set belong to the five aircraft types shown in Fig. 8. These are the McDonnell-Douglas MD-11
(MD11), Boeing 767-300 (B763), Boeing 747-400 (B744), Boeing 777-200LR (B77L), and McDonnell-Douglas
DC-10 (DC10).
Table 1 Top airports

Louisville International (KSDF)

Memphis International (KMEM)
Hong Kong International (VHHH)
Shanghai Pudong International (ZSPD)
Seoul Incheon International (RKSI)
Newark Liberty International (KEWR)

Taiwan Taoyuan International (RCTP)

Osaka Kansai International (RJBB)
Oakland International (KOAK)
Narita International (RJAA)
Indianapolis International (KIND)
Aircraft trajectories are optimized based on the aerodynamic parameters and aircraft performance data provided
by BADA for the five aircraft types described above. BADA [9] does not include the B77L, so the aerodynamic
parameters and aircraft performance data for the Boeing 777-200 (B772) are used instead. High takeoff mass is
assumed for each aircraft type since commercial cargo flights are usually heavily loaded. Typical aircraft climb and
descent profiles during initial climb, en-route step climbs and final descent are obtained for each aircraft type from
BADA [9] performance files for generating the complete aircraft trajectories. These are defined based on altitude,
air speed and mass for each phase.
The optimal speed depends on the cost performance index that is generally specified by the airlines and chosen
according to their priorities. The current optimization procedure solves for the optimal speed by iterating a range of
cruise speeds that requires a more detailed aerodynamic model. In addition, current BADA parameters, such as the
drag coefficients, may only be valid at the specified nominal speeds. Due to these limitations, this study does not
optimize cruise speed for each flight. In this paper, the true air speed during cruise is calculated from the typical
Mach number suggested by BADA.

Fig. 6 Top 10 origins in 2010

Fig. 7 Top 10 destinations in 2010

Fig. 8 Top 5 aircraft types in 2010

IV.

Results

This section presents the potential benefits for commercial cargo flights of flying wind-optimal trajectories with
an optimal vertical profile. The results are based on air traffic and global wind data from October 2010. Past studies
focus on developing minimum-time trajectories in winds at a single altitude without considering potential fuel
savings from en-route step climbs. The subsection on Comparing Wind Optimal Trajectories, below, compares wind
optimal trajectories with an optimal vertical profile and on multiple flight levels (WOMFL) to wind optimal
trajectories at a single flight level (WO1FL). The trajectories are compared for cargo flights between ANC and the
top 10 origin and destination airports for flights to and from ANC. The subsection on Comparing Flight Plan and
Wind Optimal Trajectories assesses the performance of the wind optimal trajectories with an optimal vertical profile
(WOMFL) and the flight plan trajectories (FP) along the same vertical profile.

3.5

MD11
B763
B744
B772
DC10

Savings, %

2.5
2
1.5
1
0.5
0
0.5

KSDF

KMEM

VHHH

ZSPD

RKSI
KEWR
Origin Airports

RCTP

RJBB

KOAK

2.5

MD11
B763
B744
B772
DC10

2
1.5
Savings, %

RJAA

1
0.5
0
0.5

KSDF

KMEM

ZSPD

KOAK

RCTP
VHHH
KIND
Destination Airports

KEWR

RJBB

RKSI

Fig. 9 Potential fuel benefits for step climbs

A. Comparing Wind Optimal Trajectories
This subsection analyzes the WOMFL and WO1FL trajectories for cargo flights to and from ANC. The
performance of the WOMFL trajectories is evaluated by assessing and comparing average fuel consumption and
travel time of these trajectories to that of the WO1FL trajectories.
Figure 9 shows the average fuel savings in bar graphs for each origin, destination and aircraft type, for all cargo
flights in October, 2010 to and from ANC. Flying the WO trajectories has a positive fuel saving for all of the origins
and destinations except KOAK and KIND. Fuel savings are between 0.1% and 3% depending on the airport pairs.
They are directly proportional to the length of the routes, which determines the number of en-route step climbs. The
average travel time, fuel burn, and number of step climbs for the top 10 origins are shown in Table 2. The average
time and fuel savings are shown in parentheses.

The potential fuel savings are very small for all domestic flights since most of them have only one step climb.
Flights between ANC and KOAK or KIND have the shortest travel time. For these routes, performing an en-route
step climb actually increases the total fuel burn. Note that the additional fuel burn required for each step climb is not
considered when developing the aircraft vertical profile, which assumes instantaneous climb for simplicity in the
optimization. Most international flights save more than 1% fuel depending on the aircraft types. The B744 and B772
have the largest amount of fuel savings while the MD11 and B763 have smaller and similar amount of fuel savings.
The WOMFL trajectories for all the domestic flights have equal or longer travel times than the corresponding
WO1FL trajectories. The additional time involved in climbing and descending is not negligible for these relatively
short trips. The WOMFL trajectories for all international flights except RJAA, which is the shortest of the
international routes, have positive time savings. The time savings not only depend on the trip length but also on the
flight direction. The travel time for each flight is affected by en-route wind magnitude and direction. Therefore,
long-distance wind-optimal trajectories not only gain fuel savings from en-route step climbs but may also benefit
from high-speed winds at higher altitudes.
Table 2 Average wind-optimal performance

Origin

AC
Type

Time,
[minute]

Fuel,
[tonne]

Steps
[no.]

KSDF

MD11

368 (0)

52.5 (0.03)

B763

384 (-1)

33.0 (0.01)

B744

363 (-1)

66.1 (0.15)

MD11

369 (0)

52.6 (0.03)

B772

368 (1)

47.7 (0.20)

DC10

374 (1)

51.9 (0.21)

MD11

545 (4)

74.0 (1.32)

B744

534 (5)

93.1 (2.25)

B772

542 (4)

67.4 (2.06)

MD11

462 (2)

64.1 (0.76)

B763

478 (3)

40.0 (0.75)

B744

453 (3)

80.6 (1.41)

B772

459 (3)

58.3 (1.40)

MD11

411 (2)

57.9 (0.63)

B763

426 (3)

36.2 (0.60)

KMEM

VHHH

ZSPD

RKSI

Origin

KEWR

RCTP

RJBB

KOAK
RJAA

AC
Type

Time,
[minute]

Fuel,
[tonne]

Steps
[no.]

B744

404 (3)

72.9 (1.12)

MD11

386 (0)

54.8 (0.06)

B763

402 (-1)

34.3 (0.04)

B772

385 (-1)

49.7 (1.13)

MD11

498 (1)

68.5 (0.75)

B763

516 (1)

42.8 (0.71)

B744

489 (2)

86.2 (1.48)

MD11

389 (1)

55.1 (0.45)

B763

402 (2)

34.4 (0.46)

B772

386 (4)

49.9 (1.15)

MD11

225 (0)

33.3 (0)

DC10

228 (-1)

33.3(-0.07)

MD11

362 (-1)

51.6 (0.13)

B763

374 (-1)

32.3 (0.14)

DC10

365 (1)

50.7 (0.29)

B. Flight Plan and Wind Optimal Trajectories

In the future, with improvements in air traffic management, route structures are likely to become more flexible.
With these more flexible route structures, future air traffic will be able to choose to fly their preferred routes to a
greater extent than is the case today. This subsection assesses the potential benefits for the FedEx and UPS cargo
flights when they fly WOMFL trajectories instead of FP routes. Both the WOMFL and FP trajectories follow the
optimal vertical profiles. Table 3 presents the average savings per flight, in descending order, for the top origins and
destinations to and from ANC. The potential savings for the inbound flights to ANC are shown in the blue cells and
the data for the outbound flights from ANC are in the green cells. The abbreviations n/a indicate where data are
unavailable for some flight directions.

Figure 10 shows the average percentage savings for fuel in solid bar graphs

for each origin, destination and aircraft type.

Table 3 Potential benefits for wind-optimal

Airport

AC
Type

Time
Savings
[minute]

Fuel
Savings
[tonne]

VHHH

MD11

13

1.49

54

6.10

Airport

RKSI

RCTP

ZSPD

RJBB

KSDF

KEWR

KMEM

RJAA

KIND

AC
Type

Time
Savings
[minute]

Fuel
Savings
[tonne]

B763

n/a

45

n/a

3.14

B744

17

52

2.63

7.60

B772

50

0.69

5.35

MD11

15

47

1.81

5.71

B763

36

38

2.67

2.83

B744

46

0.56

7.26

B772

n/a

23

n/a

2.61

MD11

14

35

1.60

4.10

B763

25

33

1.76

2.32

B744

16

n/a

2.40

n/a

MD11

20

26

2.46

3.13

B763

25

10

1.84

0.68

B744

26

4.09

0.48

B772

n/a

0.55

n/a

MD11

16

0.98

2.01

B763

10

17

0.77

1.27

B744

n/a

15

n/a

2.40

B772

20

11

2.28

1.28

MD11

0.95

0.94

B763

10

0.33

0.75

B744

0.69

0.96

MD11

0.82

0.41

B763

n/a

0.38

n/a

B772

0.78

0.60

MD11

0.59

0.84

B772

0.63

0.70

DC10

0.58

0.70

MD11

n/a

0.04

n/a

B763

n/a

0.65

n/a

DC10

n/a

0.12

n/a

MD11

n/a

n/a

0.34

Airport

KOAK

AC
Type

Time
Savings
[minute]

Fuel
Savings
[tonne]

B772

n/a

n/a

0.54

MD11

0.27

DC10

0.14

The average fuel savings are highly correlated to the time savings for each origin, destination and aircraft type.
The savings for each flight are proportional to the total travel time difference between the WOMFL and FP
trajectories since both follow the same vertical profile. The magnitudes of time savings are affected by en-route
wind conditions and the structure of the original flight plans.
8

MD11
B763
B744
B772
DC10

Savings, %

6
5
4
3
2
1
0

KSDF

KMEM

VHHH

ZSPD

RKSI
KEWR
Origin Airports

RCTP

RJBB

KOAK

RJAA

10
MD11
B763
B744
B772
DC10

Savings, %

KSDF

KMEM

ZSPD

KOAK

RCTP
VHHH
KIND
Destination Airports

KEWR

Fig. 10 Potential fuel benefits for flying wind-optimal

RJBB

RKSI

Flying the WOMFL trajectories has a positive saving in fuel and time for all the origins and destinations when
compared to the FP trajectories. The domestic flights have up to 3% savings that are equivalent to about 5-10
minutes in time and 0.4-1.0 tonnes of fuel for the top three domestic airports: KSDF, KEWR and KMEM. The
international flights can potentially save more than 10% by flying wind optimal trajectories. For the outbound flights
to VHHH, RKSI and ZSPD, travel time is reduced by 23-54 minutes and fuel burn by 2.3-7.6 tonnes. In general, the
savings are proportional to the trip lengths. They also depend on the direction of flight and aircraft types. Longdistance international flights gain large time and fuel savings for all aircraft types. Domestic flights with MD11,
B744, or B772 have larger fuel savings than with B763 and DC10.
The domestic flights from ANC to KOAK and the international flights from RJAA to ANC have the smallest
savings, respectively. Note that the accuracy of the assessed benefits is greatly affected by the quality of the flight
plans in the database. FP trajectories that are simulated using flight plans with an incomplete list of waypoints lead
to an underestimation of the required travel time and fuel burn. This is because flights are assumed to fly great circle
routes between waypoints. If a flight plan has few waypoints, the FP trajectory is similar to the great circle trajectory
for that flight. Subject to the wind conditions, great circle trajectories can be very similar to wind-optimal
trajectories, especially for the short trips. Hence, the benefit of flying a wind optimal trajectory is underestimated for
these flights.
Several other factors also affect the accuracy of the results. The actual vertical flight path of each FP trajectory is
approximated by those of WOMFL trajectory since the actual aircraft takeoff weight and airspeed profile for each
flight are not available in the current database. Each simulated FP trajectory follows the vertical path that assumes a
typical aircraft weight and an airspeed profile from BADA instead of the original vertical profile. This
approximation may results in the performance underestimation for FP trajectories. In addition, the potential benefits
for WOMFL trajectories need to be discounted since they neglect the additional costs involved in operating inside
foreign airspace.

V.

Conclusion

This study developed a trajectory optimization algorithm that minimizes the cost of time and fuel burn by
integrating a method for computing minimum-time routes in winds on multiple horizontal planes and an aircraft fuel
burn model for generating fuel-optimal vertical profiles. It is applied to evaluate the potential benefits of flying
wind-optimal routes in a seamless airspace for commercial cargo flights operating between Anchorage, Alaska and

major airports in Asia and the contiguous United States. Flying wind optimal trajectories with fuel-optimal vertical
profiles reduces average fuel burn of international flights flying at a single cruise altitude by 1-3%. Long-distance
flights not only gain fuel savings from en-route step climbs but may also benefit from high-speed winds at higher
altitudes. The potential fuel savings of performing en-route step climbs are very small for the domestic cargo flights
since most of them have only one step climb. The optimal trajectories reduce fuel burn and travel time relative to
their flight plan routes for all the airport pairs. Domestic flights can potentially save up to 3% on fuel burn and travel
time, which is equivalent to about 5-10 minutes in time and 0.4-1.0 tonnes of fuel for the top three domestic airports.
International flights can potentially save up to 10% on fuel and travel time by flying wind optimal trajectories,
reducing travel time by 23-54 minutes and fuel burn by 2.3-7.6 tonnes. The actual savings in operations will vary
from the simulation results due to differences in the aircraft models and user defined cost indices. In general, the
savings are proportional to the trip lengths, and depend on the en-route wind conditions and aircraft types.

Appendix
A. Optimal Cruise Altitudes
Taking the first derivative of the aircraft fuel flow rate with respect to the altitude to obtain the following
equations,

f C fcr
SFC
D
=
(D
+ SFC ),
h 1000
h
h

(A1)

where

SFC C f 1 VTAS
=
and
h
C f 2 h
D 1 &
C 4m 2 g 2 )
2
= (CD 0 S VTAS
D22
+ .
2
h 2 '
S VTAS
* h

SFC is a constant when VTAS remains unchanged for various cruise altitudes, i.e.,

general,

(A2)

VTAS
SFC
=0
= 0 . In
h
h

C 4m 2 g 2

D
2
= D22
0; and solving the equation:
= 0 yields CD 0 S VTAS
. This implies
2
h
h
S VTAS

2 =

CD 2 4m 2 g 2
.
4
CD 0 S 2 VTAS

(A3)

Below the tropopause (altitude h < 11000 m), the temperature is calculated as a function of altitude

TISA = T0 _ ISA 6.5 h /1000,

(A4)

and the air density is calculated as a function of temperature

ISA = 0 _ ISA (TISA /T0 _ ISA )

g
1
K T R gas

(A5)

Above the tropopause (h >= 11000 m), the temperature is a constant. The air density is

ISA = trop _ ISA e

g
)(h 11000)
R gas Ttrop _ ISA

(A6)

The formulas for the optimal cruise altitudes under ISA are obtained using Equations (A3-A6)

B. Application of Dynamic Programming

The concept of DP is applied to determine the locations for en-route step climbs at the step climb times. The
optimal step climbs locations are approximated based on the total cost-to-go associated with each wind-optimal
extremal. Assuming that there are N-1 possible step climbs and n possible locations for each climb, a forward
dynamic programming approach with N-1 stages and n possible states in the set of SN for each stage can be
formulated as:

J(SN ) = 0
J(SN1 ) = min{J(SN ) + CN1 (SN1 )}
SN1

J(SN2 ) = min{J(SN1 ) + CN2 (SN2 )}

SN2

(A7)

J(S1 ) = min{J(S2 ) + C1 (S1 )}

S1

J(S0 ) = min{J(S1 ) + C0 (S0 )}

S0

Note that, the starting cost at the origin J(SN) is zero with N stages to go. The optimal cost-to-go at the N-1 stage,
J(SN-1), is the link cost, CN-1(SN-1), from the origin to the current state. The link cost is the associated direct operating
cost specified by Eq. (2) along the wind-optimal flight path. In general, the optimal cost-to-go at any arbitrary stages
depends on the optimal cost-to-go in the previous stage and the minimum cost from previous stage to the current

stage. The optimal cost is repeatedly solved until the destination is reached at S0. The optimal solution can be found
in O(N x n x n) operations. Each operation involve computation of the link cost and determining the optimal cost-togo. The computation time for this approach is very high since the optimal solution at each state has to be computed
to provide the final solution.
The algorithm developed in this study approximates the aforementioned approach by solving N-1 individual DP
problems that each has one climb (N=2) in order to reduce the computational effort. Each of the 1-stage DP
problem can be written as the following:

J(S2 ) = 0
J(S1 ) = min{J(S2 ) + C1 (S1 )}
S1

(A8)

J(S0 ) = min{J(S1 ) + C0 (S0 )}

S0

J(S0 ) = min{C1 (S1 ) + C0 (S0 )}

S1

The solution for each problem is the optimal step climb location that becomes the origin for the subsequent
problems. The complexity of this algorithm is reduced to O(N x n). This algorithm when applied in the real time
also accommodates the frequently updated weather forecast and dynamic nature of atmospheric conditions by
optimizing the aircraft trajectory segments at each step climb using most updated weather forecast.
The DP approach in Eq. (A8) is equivalent to that of Eq. (A7) when aircraft performs only 1 step climb. This is
true for domestic cargo flights operating between the 2 most popular city pairs. The algorithm can be formulated by
adapting Eq. (A7) for flights with more than 1 step climb to gain better accuracy.

The current two-steps

optimization approach limits aircraft en-route step climbs at the time instants and flight levels that are determined by
the aircraft vertical profile to enhance computational efficiency. In general, a larger selection for possible aircraft
cruise altitudes and climbing times that results in larger N and n can be selected for the DP approach. Combining the
horizontal and vertical optimization procedures provides a better approximation of the global optimal solution but
significantly increases the computational efforts.

References
[1] Bryson, A. E., and Ho, Y. C., Applied Optimal Control, Taylor and Francis, Levittown, PA, 1975, Ch. 2, pp. 42-89.
[2] Bryson, A. E. and Desai, M. N., and Hoffman, W. C., Energy-State Approximation in Performance Optimization of
Supersonic Aircraft, Journal of Aircraft, Vol. 6, No. 6, 1969, pp. 481-488.

[3] Erzberger, H., and Lee, H. Q., "Constrained Optimum Trajectories with Specified Range," Journal of Guidance and Control,
Vol. 3, No. 1, 1980, pp. 78-85.
[4] Calise, A. J., Extended Energy Management for Flight Performance Optimization, AIAA Journal, Vol. 15, No. 3, March
1977, pp. 314-321.
[5] Calise, A. J. and Moerder, D. D., "Singular Perturbation Techniques for Real Time Aircraft Trajectory Optimization and
Control," NASA CR-3597, 1982.
[6] Zermelo, E., ber die Navigation in der Luft als Problem der Variationsrechnung On the Navigation in the Air as a
Problem of the Calculus of Variations, Jahresbericht der Deutschen Mathematiker-Vereinigung, Vol. 39, 1930, pp. 4448.
[7] Bijlsma, S. J., Optimal Aircraft Routing in General Wind Fields, AIAA Journal of Guidance, Control, and Dynamics, Vol.
32, No. 3, MayJune 2009, pp. 10251028.
[8] Jardin, M. R., and Bryson, A. E., Methods for Computing Minimum-Time Paths in Strong Winds, AIAA Journal of
Guidance, Control, and Dynamics, Vol. 35, No. 1, 2012, pp. 165-171.
[9] User Manual for the Base of Aircraft Data (BADA), Revision 3.6, Eurocontrol Experimental Center (EEC) Note No. 10/04,
Project ACE-C-E2, Sept. 2004.
[10] Aggarwal, R., Dushman, A., and Calise, A. J., An Analysis of Operational Procedures and Design Modifications for
Aircraft Fuel Conservation, In: Atmospheric Flight Mechanics Conference for Future Space Systems, Boulder, Colo.,
August 6-8, 1979, Collection of Technical Papers. (A79-45302 19-01) New York, AIAA Inc., 1979, p.232-242, AIAA-19791656.
[11] Liden, S., Optimum Cruise Profiles in the Presence of Winds, IEEE/AIAA 11st Digital Avionics Systems Conference, Oct.
1992, Doi:10.1109/DASC.1992.282147.
[12] Ng, H. K., Sridhar, B., Grabbe, S., and Chen, N., Cross-Polar Aircraft Trajectory Optimization and the Potential Climate
Impact, IEEE/AIAA 30th Digital Avionics Systems Conference, Oct. 2011, Doi:10.1109/DASC.2011.6096060.