Documente Academic
Documente Profesional
Documente Cultură
Coordinated 1996 HST and IRTF Imaging of Neptune and Triton. I. Observations,
Navigation, and De onvolution Te hniques. by L.A. Sromovsky, P.M. Fry, K.H. Baines,
S.S. Limaye, G.S. Orton, and T.E. Dowling, submitted to I arus on 14 July 1999.
Abstra
t
To obtain sub-pixel a
ura
y in navigation of IRTF images of Neptune, we needed to
determine the NSFCAM pixel s
ale for ea
h lter used and as a fun
tion of lo
ation on the
dete
tor array to better than 0.1%. This was a
omplished using measurements of Uranian
satellites and stellar angular separations at dierent lo
ations on the dete
tor array. When
set to the nominal value of 0.148 /pixel we found that the NSFCAM image s
ale varied
00
with lter from 0.1470 /pixel to 0.1479 /pixel at the
enter of the image and in
reased
00 00
near the edges of the image as a result of barrel distortion. These results stri
tly apply only
for the opti
al
onguration employed in our 13-14 August 1996 observing run. They do
not ne
essarily apply for
ongurations that in
lude the UV-visible di
hroi
or the tip-tilt
se
ondary mirror.
1
Sromovsky et al., NSFCAM Image Geometry 2
0.01500 (Ja
obson, personal
ommuni
ation; Ja
obson 1991). On 13 August 1996, the Umbriel-
p
Ariel separation was about 2200 , so that ephemeris un
ertainty
ontributes about 0.07% 2
to the s
ale un
ertainty. It is not
lear to what degree this un
ertainty
an be redu
ed by
averaging, and over what time interval the errors a
tually vary. However, given that the pixel
s
ale is
onsistent within better than 0.1%, it appears that for a series of images taken within
an hour or so, the ephemeris error must be
onsidered a bias error, and applied after random
errors are averaged.
A sample image of Uranus is provided in Fig. 1. Three satellites
an be seen within the ( Fig. 1
amera frame, Ariel (the brightest), Umbriel, and Miranda (the dimmest). The pre
ision with
whi
h the satellites
an be lo
ated is a fun
tion of the image brightness, as determined from
Monte Carlo simulations of the tting pro
ess. For the spe
i
onditions of this image we
estimate un
ertainties of 0.014 pixels for Ariel, 0.019 pixels for Umbriel, and 0.088 pixels for
Miranda. By far the most a
urate image s
ale is obtained from the Ariel-Umbriel dieren
e
be
ause the high brightness of these satellites yields the best positional a
ura
y. All s
ale
determinations were thus obtained from Ariel and Umbriel, using 15 images in 5 dierent
lters. In ea
h of these images we rst measured the
oordinates of the satellites using two
dimensional gaussian ts, then
orre
ted the
oordinates for distortion, as des
ribed in Se
tion
2, and determined the distan
e between them in
orre
ted image pixels. We then determined
the Ariel-Umbriel position angle relative to the image y axis, from whi
h we
omputed the
ounter
lo
kwise angle from Celestial North to the
amera Y axis. The results are shown in
Fig. 2, where we nd a somewhat surprising dependen
e of image s
ale on imaging lter. Note ( Fig. 2
that the average J-lter (+ blo
ker) s
ale of 0.14700.0001 is nearly 0.6% smaller than the
average K-lter s
ale of 0.14790.0001, whi
h is a 9- dieren
e. While there is somewhat of
a trend towards in
reasing s
ale with in
reasing wavelength, that trend is interrupted by the
Spen
er 1.73-m lter, for whi
h we nd a lower value than at surrounding wavelengths.
The average s
ales derived from the satellite measurements on 13 August 1996 are shown
in Fig. 3, in
omparison with those derived using paired star observations. The latter results (
Fig. 3
estimating s
ales for CVF wavelengths of 1.59 and 2.1 m, for whi
h no satellite observations
are available. We used the s
aled relative information to adjust the pixel s
ales as follows. For
J, H, Spen
er 1.7, and K lters we adopted s
ales that were weighted averages of the absolute
and s
aled relative values, with the latter's weighting redu
ed be
ause of the in
reased varian
e
asso
iated with the s
ale fa
tor un
ertainty of 0.0001 ar
se
ond/pixel. For all the CVF lters
we
ould dis
ern no
lear s
ale dependen
e on lter, and thus adopted a single s
ale value equal
to that of the K lter, whi
h is
onsistent also with the relatively a
urate determination for
the 2.00-m lter. We dis
overed a signi
ant s
ale dieren
e for the J lter: with the blo
ker
in pla
e, the J lter s
ale in
reases by 0.33%. The observations of the Uranian satellites on 13
August and the star observations on 14 August both used a blo
ker, while the 14 August 96
Uranian satellite observations did not use a blo
ker with that lter. The adopted values are
summarized in Table I and shown as
ir
les in Fig. 3. The variability of these results from one ( .
Table I
image to the next is smaller than might be expe
ted from the 1- ephemeris un
ertainty. In
Sromovsky et al., NSFCAM Image Geometry 4
fa
t, measurements of image s
ales on 14 August 96, not used in formulating the adopted values
(ex
ept for the J w/o blo
ker value), serve to
onrm those values with very high a
ura
y (see
Table I). Note that the K and Spen
er 1.7 values are indistinguishable from the adopted s
ales.
2. Un
ertainties listed here are in addition to the above s
ale fa
tor un
ertainty.
3. In
ludes distortion
orre
tions.
These results are
onsistent with the Baines et al. (1998) value of 0.14780.0010 ar
se
-
ond/pixel, though our values are at least ten times more pre
ise, and thus are able to
learly
resolve the lter dependen
ies.
From Ariel and Umbriel we also determined
amera orientation, spe
ied as the
ounter-
lo
kwise angle from
elestial north to the image y axis. As is apparent from Fig. 2, there is
relatively little variation in orientation angle that
an be asso
iated with lter
hanges. For 13
August 1996 we found an average orientation angle of 0.6860.008 , and for 14 August 1996
we obtained 0.6640.005 using only the more a
urate values. The dieren
e of 0.0240.009
is larger than suggested by our estimated un
ertainties, but is still well within our required
a
ura
y of 0.04 . We also used the beam swit
h dire
tion to estimate the image orientation,
and obtained a similar value. For the most part, deviations among various s
ale and angle
determinations, as evident in Fig. 2, seem to be
onsistent with our estimated un
ertainties.
It is important to note that all of the s
ale determinations a
ount for NSFCAM image
distortions that we determined from observations of star elds, as des
ribed in the following
se
tion. Without su
h
orre
tions apparent s
ale variations of several tenths of a per
ent are
possible.
We have ignored possible
hanges in s
ale asso
iated with fo
us variations that o
ur during
Sromovsky et al., NSFCAM Image Geometry 5
an observing run. This is justied by observations indi
ating no signi
ant dependen
e of s
ale
on fo
us position. During the 14 August 96 IRTF observing run we a
quired a set of images
ontaining 5 stars in addition to Neptune and Triton. This series in
luded J-ltered images
taken during fo
us adjustment. We measured distan
es between stars in ea
h image, and
analyzed how those distan
es varied with fo
us position. We found that the variations with
respe
t to fo
us were essentially random variations of about the magnitude expe
ted from our
un
ertainty estimates, and generally was well within 0.1% of the mean. There was no dis
ernible
trend indi
ating a dependen
e on fo
us, at least within the range that might be experien
ed in
a night of observing.
The ee
ts of atmospheri
refra
tion on apparent zenith, as given for example by Allen
(1964), s
aled by a fa
tor of 0.6 to a
ount for the redu
ed atmospheri
mass above 14,000
ft, yields a deviation of 60 ar
se
onds at 2 air masses. The dierential refra
tion, whi
h
has a dire
t ee
t on the apparent image s
ale in the altitude dire
tion, is 0.07% at visible
wavelengths and slightly smaller at near IR wavelengths, and in any
ase not very signi
ant.
As most quality observations are nearer to 1.4 air masses, where the ee
t is at 0.035%, and
ompletely ignorable.
Figure 2: NSFCAM image s
ale and orientation derived from Uranus satellite observa-
tions. The upper labels indi
ate image le number and satellite
ombination (U, A, and
M denote Umbriel, Ariel, and Miranda) The lower labels identify lters. For these data
the J lter was used with a blo
ker (see next gure for s
ale results when a blo
ker is
not used. The Baines et al. (1998) value was derived from 1995 images.
Sromovsky et al., NSFCAM Image Geometry 6
(x0 ; y0 ) = (x ; y ) + (x
o o x ;y
o y )(1 + k(r=r )2 )
o r (1)
where (x; y) are the raw
oordinates, (x ; y ) is the
enter of distortion (near the
enter of the
o o
image), r2 = (x x )2 +(y y )2 , and r =127 pixels. This provides a radially dire
ted
orre
tion
o o r
Sromovsky et al., NSFCAM Image Geometry 7
that in
reases quadrati
ally with distan
e from the
enter, with proportionality
onstant k.
This is
onsistent with the
lassi
al form of distortion derived in third-order aberration theory
(Born and Wolf 1980), for whi
h the aberration itself (dieren
e between ideal and a
tual image
oordinates) is proportional to the
ube of the o-axis distan
e. To determine the
oe
ient
k and the ee
tive
entral
oordinates (x ; y ), we sele
ted those values whi
h minimize 2 =
o o
X(dx (i; k)
2 + 2 , where
x y
N
X
= (dy (i; k)
2
y
N
beam positions a and b respe
tively, and x;a and x;b are the un
ertainties in those
ompo-
nents. Analogous denitions apply to the y-equation. Whether the solution obtained yields a
signi
ant and useful
orre
tion
an be assessed by the value of 2 at the minimum. A value
lose to N indi
ates that we have removed all the varian
e beyond what was expe
ted just
from the un
ertainty in tting gaussians to the star images. A larger value indi
ates that the
orre
tion fun
tion is not doing a good job of
hara
terizing the variations that are present.
The results of our t to 17 quartets of measurements on 17 star pairs in two dierent beam
positions are shown in Fig. 4. We found best t parameter values of k = 0:005500 0008
0006 , x = :
: o (
Fig. 4
137.533 39 , and y = 138.242 31 . The
orre
ted
oordinates yield a fa
tor of six redu
tion in 2
:
: o
:
:
and nearly fa
tor of three redu
tion in the standard deviation of dieren
es from the desired
zero value. The 2 minimum value is about 3-5 times N , indi
ating that the total noise in the
measurements is a fa
tor 1.7-2.2 greater than indi
ated by our un
ertainty estimates. Be
ause
our estimates are well founded and veried by observational
he
ks, this seems to imply that
the distortion
orre
tions themselves have a variability that is not fully
hara
terized by the
fun
tional form we have
hosen to t. The residual variation, though rather small, is of the
order of 0.1 pixels, and represents one of the larger
ontributors to the error budget in our oset
navigation method. The sign of k being positive means that the NSFCAM image has a barrel
distortion, so that features at the edge of the frame appear too
lose to the
enter of the frame.
In our appli
ationi this
orre
tion is used as follows. We rst determine Triton's
oordinates
within an image
ontaining Neptune and Triton. Next we
orre
t the Triton
oordinates to
values that would apply to an undistorted image. Using the image orientation and ephemeris
Sromovsky et al., NSFCAM Image Geometry 8
values for the angular osets between Triton and Neptune at the
entral time of the image,
we divide by the
entral image s
ale to determine the
orre
ted pixel oset, and add that to
the Triton
oordinates to obtain the
orre
ted position of Neptune's
enter in an undistorted
image. These
oordinates are then distorted by the inverse of the
orre
tion fun
tion, to dene
the position of Neptune's
enter in the distorted image.
The lo
al image s
ale in the distorted NSFCAM image varies with position. It is smallest
near the
enter, and in
reases towards the edge, i.e. there are more ar
se
onds per pixel at the
edges of the image. Unfortunately the lo
al s
ale is not isotropi
. The radial s
ale is given by
S (1 + 3k(r=r )2 ) and the tangential s
ale by S (1 + k(r=r )2 ), where S is the lter-dependent
o
o
o
S (r) = S (1 + 2k(r=r )2 )
o
(4)
At r = 127 pixels, this evaluates to S (r)=S = 1:01, while at a more typi
al distan
e of 60 pixels,
o
the ee
t is less than 0.25%. In any
ase, the lo
al s
ale is taken into a
ount in transforming
image
oordinates to planetary
oordinates.
Referen es
A
ton, C.H. 1996. An
illary Data Servi
es of NASA's Navigation and An
illary Information
Fa
ility. Planetary and Spa
e S
ien
e Vol. 44 , No. 1, 65-70.
Allen, C.W. 1964. Astrophysi al Quantities. 2nd Edition, Oxford Univ. Press In ., New York
Baines, K.H., P.A. Yanamandra-Fisher, L.A. Lebofsky, T.W. Momary, W. Golish, C. Kaminski,
and W.J. Wild 1998. Near-Infrared Absolute Photometri
Imaging of the Uranian System.
I
arus 132 , 266-284.
Born, M. and E. Wolf 1980. Prin iples of Opti s . 6th edition, Pergamon Press, Oxford.
Giorgini, J.D., D.K. Yeomans, A.B. Chamberlin, P.W. Chodas, R.A. Ja
obson, M.S. Keesey,
J.H. Lieske, S.J. Ostro, E.M. Standish, and R.N. Wimberly 1998. Horizons, JPL's On-
Line Solar System Data and Ephemeris Computation Servi
e , a user's guide available from
ftp://ssd.jpl.nasa.gov/pub/ssd/Horizons do
.ps.
Sromovsky et al., NSFCAM Image Geometry 9
Ja
obson, R.A., J.E. Reidl, and A.H. Taylor 1991. The orbits of Triton and Nereid from
spa
e
raft and Earthbased observations, Astron. , 565 -575.
and Astrophys. 247
Sromovsky, L.A., P.M. Fry, K. H. Baines, S.S. Limaye, G.S. Orton, and T. E. Dowling 1998.
Coordinated 1996 HST and IRTF Imaging of Neptune and Triton. I. Observations, Navigation,
and De
onvolution Te
hniques. 1999. Submitted to I
arus, 14 July 1999.