AN ALTERNATIVE LUNAR EPHEMERIS MODEL FOR ON-BOARD

FLIGHT SOFTWARE USE

David G. Simpson
Flight Software Branch
NASA Goddard Space Flight Center, Greenbelt, Maryland 20771

ABSTRACT

In calculating the position vector of the Moon in on-board flight software, one often
begins by using a series expansion to calculate the ecliptic latitude and longitude of the
Moon, referred to the mean ecliptic and equinox of date. One then performs a reduction
for precession, followed by a rotation of the position vector from the ecliptic plane
to the equator, and a transformation from spherical to Cartesian coordinates before
finally arriving at the desired result: equatorial J2000 Cartesian components of the
lunar position vector. An alternative method is developed here in which the equatorial
J2000 Cartesian components of the lunar position vector are calculated directly by a
series expansion, saving valuable on-board computer resources.

INTRODUCTION

The calculation of the orbit of the Moon is one of the oldest problems in celes-
tial mechanics. Its solution has had great historical significance as a test of Newton’s
theory of gravity, with much of the early work on the problem having been done by
Newton himself in his discussion of the two- and three-body problems in Book I of
the Principia. In past centuries, accurate predictions of the position of the Moon have
also been of great practical interest as a navigational aid for seafaring vessels, prompt-
ing the English government and scientific societies to offer rewards for accurate lunar
prediction tables. 1 The resulting body of work developed during the eighteenth and
nineteenth centuries forms the basis of the lunar theory still in use today.
Modern lunar theory was first developed by G.W. Hill 2 5 in 1878, and later ex-
panded and improved by E.W. Brown 6 in 1896. The problem of lunar motion ad-
dressed by Hill and Brown is a surprisingly difficult one; while the underlying physical
laws are very simple, the motion itself is quite complex. 7 11 The basic motion of the
Moon around Earth is affected by many strong perturbations such as those due to the
Sun, the other planets, and Earth’s equatorial bulge. These perturbations result in an
advancement of the line of apsides of the lunar orbit, a regression of the line of nodes,
and other periodic perturbations superimposed on these motions. For high accuracy, it
is necessary to compute hundreds of periodic variations in the motion, although com-
puting only the most important few terms results in a level of accuracy that is adequate
for flight software use.
There have been two major reasons for calculating the position of the Moon in
spacecraft on-board computer flight software. First, one often wishes to write flight
software to prevent the spacecraft from pointing sensitive instruments at the Moon,
which can have an apparent magnitude as bright as 12 at full Moon. 12 Second, one

175
1999 Flight Mechanics Symposium, NASA GSFC

may require the flight software to calculate stellar aberration corrections. 13 For high
accuracy, this requires calculating the velocity vector of Earth with respect to the Earth-
Moon barycenter, which in turn requires a calculation of the lunar velocity vector. If
the flight software can calculate a lunar position vector, then this velocity vector may
be found by differentiating the lunar position vector with respect to time.

REVIEW OF CURRENT MODELS

A number of approaches for calculating a lunar position vector are currently used
by spacecraft flight software. In the flight software for the Hubble Space Telescope’s
DF-224 flight computer, for example, one finds the position of the Moon using a simple
two-body model. The standard two-body calculations 14 are modified somewhat to
allow for the motion of the nodes and apsides of the lunar orbit. A new set of orbital
elements is uplinked from the ground every few days to keep the error in the model to
within acceptable limits, on the order of 1ı . While this model is not highly accurate, it
has the virtue of being very fast—a necessity for the 1970s-vintage flight computer.
An approach commonly used with more modern flight computers is based on the
low-precision formulae given in the Astronomical Almanac.15;16 This model is based
on earlier work done by the Almanac Offices of the United States and United Kingdom 17
and by Eckert, Walker, and Eckert,18 all of which are based on Brown’s lunar theory. 6
In this model, one begins by using series expansions to calculate the ecliptic longi-
tude , ecliptic latitude ˇ, and horizontal parallax  of the Moon, referred to the mean
ecliptic and equinox of date:

 D 218ı:32 C 481 267ı:883 t

C6ı:29 sin.477 198ı:85 t C 134ı:9/
1ı:27 sin. 413 335ı:38 t C 259ı:2/
C0ı:66 sin.890 534ı:23 t C 235ı:7/
C0ı:21 sin.954 397ı:70 t C 269ı:9/
0ı:19 sin.35 999ı:05 t C 357ı:5/
0ı:11 sin.966 404ı:05 t C 186ı:6/ ; (1)

ˇ D C5ı:13 sin.483 202ı:03 t C 93ı:3/

C0ı:28 sin.960 400ı:87 t C 228ı:2/
0ı:28 sin.6 003ı:18 t C 318ı:3/
0ı:17 sin. 407 332ı:20 t C 217ı:6/ ; (2)

176
1999 Flight Mechanics Symposium, NASA GSFC

 D 0ı:9508
C0ı:0518 cos.477 198ı:85 t C 134ı:9/
C0ı:0095 cos. 413 335ı:38 t C 259ı:2/
C0ı:0078 cos.890 534ı:23 t C 235ı:7/
C0ı:0028 cos.954 397ı:70 t C 269ı:9/ : (3)

The horizontal parallax  gives the Earth-Moon distance r :

R˚
r D ; (4)
sin 
where R˚ D 6378:140 km is the equatorial radius of Earth (IAU 1976 value). 19
Having found the lunar ecliptic mean-of-date coordinates, one must then perform
a reduction for precession to epoch J2000 (2000 January 01 12:00:00 Barycentric Dy-
namical Time) to find the ecliptic J2000 coordinates ( 0, ˇ0 ). To sufficient precision,
this may be found using the formulae 20

ˇ0 D ˇ b sin. C c/ ; (5)
0 D  a C b cos. C c/ tan ˇ0 ; (6)

where the precession constants a, b, and c are given by

a D 1ı:396 971 t C 0ı:000 3086 t 2 ; (7)

b D 0ı:013 056 t 0ı:000 0092 t 2 ; (8)
c D 5ı:123 62 1ı:155 358 t 0ı:000 1964 t 2 ; (9)

and where t is the time in Julian centuries (cy) of 36 525 days from J2000:

t D .JDE 245 1545:0/=36 525 ; (10)

and JDE is the ephemeris Julian day.

The remaining step is to rotate the coordinates from the plane of the mean ecliptic
of J2000 to the mean equator of J2000, and to convert from spherical polar to Cartesian
coordinates:

X D r cos ˇ0 cos 0 ; (11)

Y D r .cos ˇ0 sin 0 cos "0 sin ˇ0 sin "0 / ; (12)
Z D r .cos ˇ0 sin 0 sin "0 C sin ˇ0 cos "0/ ; (13)

where r is given by Eq. (4) and "0 D 23ı 260 2100:448 is the obliquity of the ecliptic at
J2000 (IAU 1976 value). 21
This model has very good precision for on-board flight software use: the rms error
in the lunar position is about 0ı:11, with a maximum error of about 0ı:35.

177
1999 Flight Mechanics Symposium, NASA GSFC

A NEW MODEL

Many of the equations involved in computing the position of the Moon using the
method just described involve what is essentially a coordinate transformation, from
ecliptic mean-of-date coordinates to equatorial J2000 Cartesian coordinates. In this
paper, I investigate the possibility of calculating the equatorial J2000 Cartesian coordi-
nates directly by series expansions similar to Eqs. (1–3), thus eliminating the need for
performing the coordinate transformations in on-board flight software.
We begin by assuming that each of the J2000 equatorial Cartesian coordinates Xn
may be represented by Fourier sine series:
Nn
X
Xn D anm sin.!nm t C ınm / ; (14)
mD1

where X1  X , X2  Y , and X3  Z; Nn is the order of the series for Xn . We

now need to find the amplitudes anm , frequencies !nm , and phase constants ınm . This
may be done by fitting these parameters to the DE200 ephemeris model 22;23 using
an exhaustive search. DE200 is an ephemeris model developed at the Jet Propulsion
Laboratory, and has been used to produce tables in the Astronomical Almanac since
1984. It calculates Cartesian coordinates of Solar System objects, referred directly to
the mean equator and equinox of J2000.
For each coordinate, the terms of the series in Eq. (14) may be found one at a time
by simultaneously fitting the parameters anm , !nm , and ınm over a grid of possible
values to the DE200 model. An algorithm for accomplishing this involves calculating
the error a!ı between the DE200 model and a “test model” a sin.!t C ı/ using each
combination of parameters a, !, and ı:

for a = ami n to amax

for ! = !mi n to !max
for ı = ıPmi n to ımax
2100
a!ı = t D2000 ŒXDE200 .t/ a sin.!t C ı/2 ,

where the summation is over 216 points covering the interval A.D. 2000–2100. The
smallest error a!ı found gives the best fit parameters a, !, and ı. This process may be
repeated several times over successively smaller search ranges and finer grid spacings
in order to find more significant digits for the parameters. Once a term has been found,
it is subtracted from the DE200 data, and the whole process repeated on the remaining
data to find the next term in the series.
In the model given by Eq. (14), we assume that the amplitudes anm are all positive,
so that amplitudes may be searched over a grid of values between 0 and the maximum
in the data set. The amplitudes may be assumed to be positive without loss of generality
by allowing the phase constants ınm to be searched over the entire range 0 to 2: since
sin   sin. C /, any potential minus sign in the amplitude is simply absorbed as
an extra  radians added to the phase constant.

178
1999 Flight Mechanics Symposium, NASA GSFC

Fourier Transform of Lunar X Coordinate

11500

11000
Magnitude of Fourier Transform

1500

1000

500

0 5000 10000 15000 20000

Frequency, ω (rad cy-1)

Figure 1. Fourier spectrum of lunar X coordinate (A.D. 2000–2100).

Determining a search range for the frequencies !nm is somewhat more complicated
than it is for the amplitudes and phase constants. A search range for !nm may be
determined by examining the peaks in the Fourier transform XOn .!/ of the DE200 data:
Z 1
XOn .!/ D Xn .t/ e i!t dt ; (15)
1

where Xn .t/ is the position coordinate at time t, and ! is the angular frequency. This
Fourier transform may be calculated by using the DE200 model to compute the lunar
position vector at N discrete time points tk , then finding the discrete Fourier transform
XOn .!p /:
NX1
XOn .!p / D Xn .tk / e i!p tk ; (16)
kD0
where Xn .tk / is the position vector at time point tk , !p D 2p=tN is the angular
frequency, and p D 0; 1; 2; : : : ; N 1. For this study, N D 214 time points were

179
1999 Flight Mechanics Symposium, NASA GSFC

chosen over the time interval A.D. 2000–2100; the magnitude of the resulting Fourier
transform jXO1 .!p /j for X is shown in Figure 1. For each term in the series expansion
(Eq. 14), a search range is taken around one of the peaks in the Fourier spectrum.
This exhaustive search process, which is essentially a curve fit to the DE200 model,
required about one week of computer time to find each term in a series, and some five
months of computer time to find the complete solution to seven terms per series. The
final results are:

X D Π383:0 sin .8399:685 t C 5:381/

C 31:5 sin .70:990 t C 6:169/
C 10:6 sin .16 728:377 t C 1:453/
C 6:2 sin .1185:622 t C 0:481/
C 3:2 sin .7143:070 t C 5:017/
C 2:3 sin .15 613:745 t C 0:857/
C 0:8 sin .8467:263 t C 1:010/   106 m ; (17)

Y D Π351:0 sin .8399:687 t C 3:811/

C 28:9 sin .70:997 t C 4:596/
C 13:7 sin .8433:466 t C 4:766/
C 9:7 sin .16 728:380 t C 6:165/
C 5:7 sin .1185:667 t C 5:164/
C 2:9 sin .7143:058 t C 0:300/
C 2:1 sin .15 613:755 t C 5:565/   106 m ; (18)

Z D Π153:2 sin .8399:672 t C 3:807/

C 31:5 sin .8433:464 t C 1:629/
C 12:5 sin .70:996 t C 4:595/
C 4:2 sin .16 728:364 t C 6:162/
C 2:5 sin .1185:645 t C 5:167/
C 3:0 sin .104:881 t C 2:555/
C 1:8 sin .8399:116 t C 6:248/   106 m ; (19)

where all angles are given in radians for convenience of use in software, t is the time
in Julian centuries from J2000 given by Eq. (10), and X , Y , and Z are the Cartesian
components of the lunar position vector, referred to the mean equator and equinox of
J2000. The terms are arranged in order of decreasing contribution to the reduction in
the error of the model.
One of the primary advantages of this model is that it allows a lunar ephemeris to
be programmed in flight software using very little code. Using Eqs. (17–19), an entire
lunar ephemeris model may be programmed in just a few lines of C code:

180
1999 Flight Mechanics Symposium, NASA GSFC

for (n=0; n<3; n++)

{
x[n] = 0.0;
for (m=0; m<7; m++)
x[n] += a[n][m]*sin(w[n][m]*t+delta[n][m]);
}
Calculations for the reduction for precession, rotation from the ecliptic to the equator,
and transformation from spherical polar to Cartesian coordinates have essentially been
“absorbed” into the series coefficients, and so do not need to be performed explicitly.

DISCUSSION OF THE NEW MODEL

An examination of the frequencies in the terms of the Astronomical Almanac model

of Eqs. (1–3) and of the new model of Eqs. (17–19) gives some interesting insights into
the lunar motion. The frequencies in the Astronomical Almanac model are all computed
as functions of the mean anomalies and mean longitudes of the Sun and Moon, 16 while
the frequencies in the model given by Eqs. (17–19) are determined entirely by a curve
fit. We examine the origins of some of the more prominent frequencies in both models
below.

Anomalistic Month

The dominant term in the expressions for the ecliptic longitude  (Eq. 1) and hor-
izontal parallax  (Eq. 3) have a frequency of 477 198.85 deg cy 1 . In deriving the
Astronomical Almanac series, this frequency was computed as the rate of change of
the Moon’s mean anomaly. Since the mean anomaly is measured in the plane of the
orbit from the perigee point, one complete cycle of the mean anomaly requires the
same amount of time as the Moon’s motion from its perigee point to its next perigee. It
comes as no surprise, then, that this frequency of 477 198.85 deg cy 1 is equal to one
revolution per anomalistic month of 27.554 550 days, where an anomalistic month is
the time required for the Moon to move from perigee to perigee.

Draconic Month

For the ecliptic latitude ˇ (Eq. 2), the dominant term has a frequency of 483 202.03
deg cy 1 . This was computed as the rate of change of the Moon’s mean longitude,
which is measured from the vernal equinox to the ascending node along the ecliptic
plane, then from the node to the Moon along the orbit plane. The Moon will have
ˇ D 0 only when it is at one of the nodes of the orbit, and it will next have ˇ D 0 again
(crossing the node in the same direction) when it returns to the same node again. We
might therefore expect that the dominant term in the expression for the ecliptic latitude
will be the time required for the Moon to move from an orbital node back to the same
node. Indeed, the frequency of 483 202.03 deg cy 1 is equal to one revolution per
draconic month of 27.212 221 days, where a draconic month is the time required for
the Moon to move from an orbital node back to the same node.

181
1999 Flight Mechanics Symposium, NASA GSFC

Sidereal Month

In the series for X , Y , and Z in the new model (Eqs. 17–19), on the other hand, the
dominant terms all have a frequency of about 8399.685 rad cy 1 , which is equal to 1
revolution per sidereal month of 27.321 662 days, where a sidereal month is measured
with respect to the fixed stars. This is a reflection of the model having its coordinate
system fixed in space (mean of J2000 equatorial coordinates).

Motion of the Apsides

A comparison of the model of Eqs. (1–3) with the new model of Eqs. (17–19) shows
that the new model includes an important term that does not appear in the conventional
model, having a frequency of about 70.99 rad cy 1 . This frequency reflects the motion
of the line of apsides of the lunar orbit. The expected frequency of this motion may be
computed from the periods of the anomalistic and sidereal months:
2 2
! D
sidereal mo. anomalistic mo.
!
2 2 days
D  36 525
27:321 662d 27:554 550d cy
1
D 70:9932 rad cy (20)

in close agreement with the frequencies found using the curve fit.

ERROR ANALYSIS

The results shown in Eqs. (17–19) have been checked against the DE200 ephemeris
model by using DE200 to generate lunar X , Y , and Z coordinates at 220 (over one mil-
lion) time points between A.D. 2000 January 1 and A.D. 2100 January 1, corresponding to
roughly one point every fifty minutes for 100 years. The model shown in Eqs. (17–19)
was run at the same time points, and the results compared with the DE200 results. This
error analysis shows an rms position error between DE200 and the new model of Eqs.
(17–19) of 0ı:341, and a maximum error of 1ı:033.

182
1999 Flight Mechanics Symposium, NASA GSFC

CONCLUSIONS

Three lunar ephemeris models for on-board flight software use have been discussed.
A modified two-body model is very fast, but is of low precision and requires constant
maintenance in the form of periodic updates of orbital elements from the ground. The
model currently in common use, which is based on the low-precision formulae in the
Ast r onomicalAlmanac, is of very good precision and will run indefinitely without
ground intervention, but requires code to convert the calculated ecliptic mean-of-date
coordinates to equatorial J2000 Cartesian coordinates. The method developed in this
paper is of intermediate precision, requires the least code of the three, and will also run
indefinitely without ground intervention. It may have applications for small missions
where computer resources are limited and its precision is acceptable.

REFERENCES

1. F.R. Moulton, An Introduction to Celestial Mechanics (Dover Publications, New

York, 1970), p. 363.

2. G.W. Hill, “Researches in the Lunar Theory,” Am. J. Math., 1, pp. 5–26 (1878).

3. Ibid., pp. 129–147.

4. Ibid., pp. 245–261.

5. G.W. Hill, “On the Part of the Motion of the Lunar Perigee Which Is a Function of
the Mean Motions of the Sun and Moon,” Acta Math., 8, pp. 1–36 (1886).

6. E.W. Brown, An Introductory Treatise on the Lunar Theory (Cambridge University

Press, Cambridge, 1896).

7. D. Boccaletti, G. Pucacco, Theory of Orbits, Volume 1: Integrable Systems and

Non-perturbative Methods (Springer, Berlin, 1996), pp. 293–299.

8. J.M.A. Danby, Fundamentals of Celestial Mechanics (2nd ed.) (Willmann-Bell,

Richmond, 1988), pp. 371–388.

9. J. Meeus, Astronomical Algorithms (Willmann-Bell, Richmond, 1991), pp. 307–

317.

10. M. Chapront-Touzé, J. Chapront, Lunar Tables and Programs from 4000 B.C. to A.D.
8000 (Willmann-Bell, Richmond, 1991).

11. F.R. Moulton, An Introduction to Celestial Mechanics (Dover, New York, 1970),
pp. 337–365.

12. N.E. Howard, The Telescope Handbook and Star Atlas (Crowell, New York, 1967),
p. 4.

183
1999 Flight Mechanics Symposium, NASA GSFC

13. P.K. Seidelmann (ed.), Explanatory Supplement to the Astronomical Almanac (Uni-
versity Science Books, Mill Valley, California, 1992), pp. 128–130.

14. S.W. McCuskey, Introduction to Celestial Mechanics (Addison-Wesley,

Reading, Massachusetts, 1963), pp. 32-69.

15. The Astronomical Almanac for the Year 1999 (U. S. Government Printing Office,
Washington, 1998), p. D46.

16. B. Emerson, Nautical Almanac Office Technical Note Number 48: Approximate
Lunar Coordinates (Science and Engineering Research Council, Royal Greenwich Ob-
servatory, June 1979).

17. Improved Lunar Ephemeris, 1952–1959 (U. S. Government Printing Office, Wash-
ington, 1954).

18. W.J. Eckert, M.J. Walker, D. Eckert, “Transformations of the Lunar Coordinates
and Orbital Parameters,” Astron. J., 71, pp. 314–332 (1966).

19. The Astronomical Almanac for the Year 1999 (U. S. Government Printing Office,
Washington, 1998), p. K6.

20. Op. cit., p. B19.

21. Op. cit., p. K6.

22. E.M. Standish, “Orientation of the JPL Ephemerides, DE200/LE200, to the Dy-
namical Equinox of J2000,” Astron. Astrophys., 114, pp. 297–302 (1982).

23. E.M. Standish, “The observational basis for JPL’s DE200, the planetary ephemerides
of the Astronomical Almanac,” Astron. Astrophys., 233, pp. 252–271 (1990).

184