Transcript (McCullough et al. 2014).

WFC3 Spatial Scanning (Exoplanets)
STScI Calibration Workshop
Aug 13, 2014
Peter R. McCullough
H2O Spatially Scanned (vertically)
H2O
Staring Mode (nominal)
th
0 order
st
1 order
nd
2 order
Build it and they will come…
Bean et al. 2013-14, 100+ orbits
Benneke & Crossfield et al. 2014-15, 100+ orbits
Scan ?
Who ?
McCullough & MacKenty (ISR WFC3-2012-08)
Spatially Scanned (vertically)
0th
1st
Staring Mode (nominal)
What ?
Spatial scanning means moving the telescope (HST) during an exposure, in this case
perpendicular to the dispersion of the spectrum.
-1
HST can scan at any rate from 0 to 7.8 arcsec s . The scan rate determines the effective exposure
time.
Where ?
Hubble/WFC3 (also now on Spitzer but not spectroscopically)
When ?
Since 2011 (and S.O.P. in the old days of photographic plates)
Why ?
Spatial scanning increases the number of photons recorded by a large factor because spatial
scanning reduces overheads considerably. For each detector preparation-readout cycle, we
obtain ~100 spectra (well-exposed rows on the detector) instead of just 1.
Spatial scanning is most useful for bright stars (H < 10 mag).
For very bright stars (H ~ 6 mag) such as HD 189733 and HD 209458, spatial scanning enables
observations that otherwise would saturate the detector even before the exposure begins.
Spatial scanning has enable the “greatest hits” science by increasing the S/N of exoplanet
spectroscopy by factors of a few.
How (to analyze the data) ?
This Presentation …
2nd
WFC3: Precision Infrared Spectrophotometry
with Spatial Scans of HD 189733b and Vega
P. R. McCullough (STScI & JHU), Nicolas Crouzet (STScI & Dunlap Inst.), Drake Deming (U. MD), Nikku Madhusudhan (Yale & Cambridge), & Susana E. Deustua (STScI)
Abstract
WFC3 Spatial Scans “Greatest Hits”
HD 189733b
The Wide Field Camera 3 (WFC3) on the Hubble Space
Telescope (HST) now routinely provides near-infrared
spectroscopy of transiting extrasolar planet atmospheres with
better than ~50 ppm precision per 0.05-micron resolution bin
per transit, for sufficiently bright host stars. Two
improvements of WFC3 (the detector) and HST (the spatial
scanning technique) have made transiting planet spectra
more sensitive and more repeatable than was feasible with
NICMOS. In addition, the data analysis is much simpler with
WFC3 than with NICMOS. We present time-series spectra of
HD 189733b from 1.1 to 1.7 microns in transit and eclipse
with fidelity similar to that of the WFC3 transit spectrum of
HD 209458b (Deming et al. 2013). In a separate program, we
obtained scanned infrared spectra of the bright star, Vega,
thereby extending the dynamic range of WFC3 to ~26
magnitudes! Analysis of these data will affect the absolute
spectrophotometric calibration of the WFC3, placing it on an
SI traceable scale
We scanned the gas giant exoplanet HD 189733b with WFC3 in June, 2013.
Its transit spectrum (blue data points, black model) shows water vapor
features at 1.2 microns and 1.4 microns similar to those of HD 209458b as
reported by Deming et al. (2013) and commensurate with a solar-abundance
clear atmosphere (McCullough et al. 2014). Its eclipse spectrum (blue data
points, black model) is flat and its molecular features are lower contrast
than similar models predict (Crouzet et al. 2014).
From top to bottom, transit spectra of HD 209458b
and XO-1b (Deming et al. 2013), GJ436b (Knutson
et al. 2013), and GJ 1214b (Kriedberg et al. 2014).
The number of transits and estimated uncertainty in
each R~70 spectral channel are listed.
HD 209458b, H=6.4, 1 transit, 36 ppm
HD 189733b, H=5.6, 1 transit, 69 ppm
XO-1b, H=9.6, 1 transit, 96 ppm
Spatial Scans: How and Why
0th
1st
Staring Mode (nominal)
The spatially scanned
spectrum of the star
GJ1214 is labeled with
its 0th, 1st, and 2nd order
light and compared to a
nominal staring-mode
slitless spectrum of the
nd
2 same field (blue outlined
inset). The images are
512 columns wide, of
the detector’s 1024. The
scan was 40 pixels high.
(4.8 arcsec).
AAS Conference, Jan 8, 2014, Washington D.C. USA, Poster 347.21
Flux (electrons/sec)
8•10
8
Vega: coadded spectra G141
6•108
Vega
4•108
Spectra of Vega for the -1st orders of the two WFC3 IR grisms (Deustua et al.
8
2•10
2014). Solid lines represent “slower” scan rates (< 5 ”s-1); dashed lines
represent faster scans (> 7 ”s-1). Left: at 24 Å/pix, Pa  is just visible at 1.09
0 and Pa  at 1.05 microns. Right: at 46 Å/pix, Paschen  is the dip at
microns,
1.0 and
1.2the series
1.4
1.8 between 1.6 and 1.7 microns are Br
1.28 microns,
of1.6features
Wavelength (microns)
13,12, 11.
6•108
4•108
2•10
8
Vega: coadded spectra G102
8•108
Flux (electrons/sec)
Spatially Scanned (vertically)
GJ 436b, H=6.3, 4 transits, 40 ppm
Flux (electrons/sec)
 Spatial scanning means moving the telescope (HST) during
an exposure, in this case perpendicular to the dispersion of
the spectrum.
 HST can scan at any rate from 0 to 7.8 arcsec s-1. The scan
rate determines the effective exposure time.
 Spatial scanning increases the number of photons recorded
by a large factor because spatial scanning reduces
overheads considerably. For each detector preparationreadout cycle, we obtain ~100 spectra (well-exposed rows
on the detector) instead of just 1.
 Spatial scanning is most useful for bright stars (H < 10
mag).
 For very bright stars (H ~ 6 mag) such as HD 189733 and
HD 209458, spatial scanning enables observations that
otherwise would saturate the detector even before the
exposure begins.
 Spatial scanning has enable the “greatest hits” science in
the sidebar (at far right) by increasing the S/N of exoplanet
spectroscopy by factors of a few.
HD 189733b, H=5.6, 1 eclipse, 57 ppm
0
0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2
Wavelength (microns)
6•10
GJ 1214b, H=9.1, 12 transits, 28 ppm
Vega: coadded spectra G141
8
4•108
2•108
References
0
1.0
1.2
1.4
1.6
Wavelength (microns)
1.8
Crouzet et al. 2014, ApJ, in prep.
Deustua et al. 2014, in prep.
Deming et al. 2013, ApJ, 774, 95
Knutson et al. 2014, Nature, 505, 66
Kreidberg et al. 2014, Nature. 505, 69
McCullough et al. 2014, ApJ, in prep.
Spatial Scanning with HST for Exoplanet Transit Spectroscopy
and other High Dynamic Range Observations
P. R. McCullough & J. MacKenty (STScI)
Why would we do this?
Scanned Imaging
Scanned Spectroscopy
Spatially Scanned (vertically)
ABSTRACT
We are reviving an old technique of spatially scanning the
telescope to improve observations, in this case with the
Hubble Space Telescope's Wide Field Camera 3 (HST WFC3).
Spatial scanning will turn stars into well-defined streaks on
the detector, or, for example, spread a stellar spectrum
perpendicular to its dispersion. There are at least two
motivations for implementing such a capability: 1) reducing
the fraction of overhead in observations of very bright stars
such as those suitable for spectral characterization of
transiting planets, and 2) enabling observations of very bright
primary calibrators that otherwise would saturate the IR
detector. We report results (McCullough & MacKenty 2011)
from two engineering tests in which a star was imaged with
WFC3 IR under various parameterizations of HST's scanning
speed and orientation, and with or without a grism in place.
CONCLUSION: It works!
The re-introduction of spatial scanning to HST has been
demonstrated for WFC3 IR and UVIS. We expect that the two
primary motivations (see Abstract) are achievable.
TWO TESTS
We executed two on-orbit tests under HST program 12325,
one on 2011-03-24 (IR imaging only) and the other 2011-04-18
(UVIS imaging & IR spectroscopy). The first test proved that
we can scan a star across the IR detector in a straight line at a
constant rate with well-defined starting and finishing
positions. The second test demonstrated 1) the same capability
for the UVIS detector (perhaps indicating an adjustment is
needed in the synchronization with the UVIS channel), 2) that
scanning works also for spectroscopy using the G141 grism
with the IR detector, and 3) measured the operational
overheads associated with a time series of scans.
SPECIFYING A SCAN
Referring to the figure below, the observer must define the
scan rate, scan orient angle α, and scan direction (forward,
meaning from a to b, or reverse, from b to a). Currently the
scan rate must be less than 1 arcsec s-1 (limited by flight
software parameters), but we are investigating if it can be
increased to 7.84 arcsec s-1 (limited by gyros). For a given scan
rate, the scan length on the detector is determined by the
effective exposure time, which depends on both the
commanded exposure time and whether the star is scanning
“upstream” or “downstream” of the IR detector’s rasterscanned readout.
The overheads per WFC3 IR
exposure in a series in our
tests were 19 s, 40 s, and 67 s,
for nominal staring-mode,
scanning mode (forward then
reverse), and scanning mode
(forward every time). The
scans were at 1 arcsec s-1.
0th
1st
Staring Mode (nominal)
INFRARED: In one test, we trailed the star P330E across the
WFC3 IR detector repeatedly, with rates of 1, 1, and 0.5 arcsec s-1
respectively. Photometry (above), integrated transverse to each of
the three trails, show intermittent oscillations with ~3%
amplitude at ~1.3 Hz, due to image motion or “jitter” during the
scans. The 2nd and 3rd trails have been offset vertically by 0.1 and
0.2 for clarity.
According to Bradley (2011), at these scanning rates, the jitter
measured by the Fine Guidance Sensor (FGS) in closed-loop
guiding was not significantly different than for nominal staringmode operation of HST, ~0.0035 arcsec rms in each of two
orthogonal axes. The power spectrum of the jitter contains peaks
at 6.2, 3.3, 2.9, 1.3, and some less than 1 Hz, caused by the
NICMOS cooling system, the high gain antenna, the aperture
door, and the solar arrays.
The photometric oscillations are commensurate with the jitter.
For example, consider the time-dependent position of the star
D(t) = V0 t + A sin ω t,
where the first term is the uniform scan rate and the second term
models the oscillatory jitter. The velocity is
V(t) = V0 + A ω cos ω t,
and hence the variability of the time for a star to cross a pixel,
which determines the photometric variability along a scan, is
sinusoidal with 3% amplitude, i.e. A ω / V0 = 0.03 for A = 0.0035
arcsec, ω = 2π(1.3 Hz), and V0 = 1 arcsec s-1 .
For a detector well-calibrated to linearity, oscillations of the sort
described above should not affect the photometric integral over
the entire scan, in principle, except for the jitter’s effect on the
precise location (and hence time) of the start and finish of each
scan. For IR imaging, the latter exception can be made negligible
by scanning parallel to the fast-read direction of the IR readout.,
i.e. “horizontally.”
Ultraviolet-visible (UVIS): In another test, we trailed the star GJ
1214 repeatedly across the same pixels of the UVIS detector at a
rate of 1 arcsec s-1 for eight 5-s F814W exposures. Integrated over
the entire 125-pixel long scan, the photometric repeatability was
4.0 mmag rms, or 40% greater than the quadrature sum of
Poisson noise (1.1 mmag) and shutter noise (2.7 mmag). For the
latter we adopted 12 milliseconds rms of shutter jitter per 5-s
exposure (Hilbert 2009). The overheads per WFC3 UVISM512C-SUB exposure in our tests ranged from 58 s to 92 s; the
58 s value was the most common, and it was the same for staring
mode or scans.
2nd
The spatially scanned
spectrum of the star
GJ1214 is labeled with
its 0th, 1st, and 2nd order
light and compared to a
nominal staring-mode
slitless spectrum of the
same field (blue outlined
inset). The images are
512 columns wide, of
the detector’s 1024. The
scan was 40 pixels high.
(4.8 arcsec).
In this test, we trailed the star GJ 1214 across the WFC3 IR
detector repeatedly at a rate of 0.1 arcsec s-1 perpendicular to
the dispersion of the G141 grism. By coincidence we observed
for 15 minutes during the egress of the transiting extrasolar
planet GJ 1214b.
The utility of scanning for the UVIS grism will be limited to
200 nm to 400 nm, due to order overlap longer than 400 nm.
POTENTIAL APPLICATIONS
We anticipate a number of potential applications for spatial
scanning with HST. You are encouraged to discuss possibilities
with us. Here are a few. For transit spectroscopy of exoplanets
with WFC3, spatial scanning could increase the number of IR
photons recorded by a factor of 10 for a star like GJ 1214
(H=9.09 mag), because spatial scanning reduces overheads
considerably. For even brighter stars such as HD 189733,
spatial scanning will enable observations that otherwise would
be compromised by pre-exposure saturation of the IR detector
that lacks a physical shutter. For a slitless stellar spectrum, one
requires a wavelength-dependent flat field, but now we can get
what we want directly, in orbit, using a spatially-scanned
spectrum of a reference star for calibration. If we can increase
the maximum scan rate to 2 arcsec s-1, the primary calibrator
star Vega will not saturate the IR detector pixels with the G141
grism in its -1st order, which is ~4.6 mag less efficient than the
+1st order (Kuntschner et al. 2011). Similarly, in imaging mode,
spatial scanning should permit unsaturated IR photometry of
calibrators without concern that the star “burns in,” i.e. gradually
appears to brighten (by ~0.5%) with time due to its own electronic
“after images.” For example, a G2 V star, with H = 9.9 mag,
scanned at 1 arcsec s-1 and imaged with the F125W filter will
be marginally below saturation of the IR detector.
Acknowledgments
Zach Berta (Harvard), the P.I. of HST program 12251, “The first characterization of a
Super-Earth Atmosphere,” gracious agreed to our request to use GJ 1214 also for our
2nd test. Many persons, especially Merle Reinhart, Alan Welty, and William “Bill”
Januszewski (STScI) helped plan the observations. Art Bradley (SSES) and Mike
Wenz (GSFC) provided FGS data and analysis. Many persons, especially Adam Reiss
and Susana Deustua, suggested applications for spatial scanning.
References
Bradley, A. 2011, Spacecraft System Engineering Services, private communication of
unpublished report “FGS Vehicle Scan (#52 Command) On-Orbit Test of 2011:083”
Hilbert, B. 2009, WFC3 ISR 2009-25, “WFC3 SMOV Program 11427: UVIS Channel
Shutter Shading”
Kuntschner et al. 2011, WFC3 ISR 2011-5, “Revised Flux Calibration of the WFC3
G102 and G141 grisms”
McCullough & MacKenty, J. 2011, WFC3 ISR 2011-NN, in preparation
Exploring Strange New Worlds, May 1-6, 2011, Flagstaff AZ USA
Calibration of Scanned Spectra
incomplete bibliography
McCullough et al. (2012)
McCullough & MacKenty (2012)
[ Berta et al. (2012), GJ 1214 transits (staring) ]
Deming et al. (2013), XO-1 & HD 209458 transits
Swain et al. (2013)*
Kreidberg et al. (2014a), GJ 1214 transit
Knutson et al. (2014), GJ 436 transit
McCullough et al. (2014), HD 189733 transit
Crouzet et al. (2014) in press, HD 189733 eclipse
Stevenson et al. (2014) in press, WASP-43 phase curves
Kreidberg et al. (2014b) submitted, WASP-43 transits
Unrefereed
* Alternative view
Caveat Emptor
A contradictory view is available, http://arxiv.org/abs/1311.3934 :
Comparing the WFC3 IR Grism Stare and Spatial-Scan Observations for Exoplanet
Characterization, Mark R. Swain, Pieter Deroo, Kiri L. Wagstaff (15 Nov 2013)*
We report on a detailed study of the measurement stability for WFC3 IR grism stare and spatial scan observations.
The excess measurement noise for both modes is established by comparing the observed and theoretical measurement
uncertainties. We find that the stare-mode observations produce differential measurements that are consistent and achieve
∼1.3 times photon-limited measurement precision. In contrast, the spatial-scan mode observations produce
measurements which are inconsistent, non-Gaussian, and have higher excess noise corresponding to ∼2 times the
photon-limited precision. The inferior quality of the spatial scan observations is due to spatial-temporal variability in the
detector performance which we measure and map. The non-Gaussian nature of spatial scan measurements makes the use of
χ2 and the determination of formal confidence intervals problematic and thus renders the comparison of spatial scan data
with theoretical models or other measurements difficult. With better measurement stability and no evidence for nonGaussianity, stare mode observations offer a significant advantage for characterizing transiting exoplanet systems.
* Reiterated March 2014 at Caltech JWST-exoplanets meeting
HD 209458b and XO-1b
Deming et al. 2013
HD 189733b transit & eclipse
Three permutations of
McCullough, Crouzet,
Madhusudhan, Deming 2014
HD 209458b, H=6.4
1 transit, 36 ppm
H2O
XO-1b, H=9.6
1 transit, 96 ppm
H2O
WASP-43b
Stevenson, Bean et al. 2014
(full phase, near Eclipse)
H2 O
H2 O
(new phase, near transit)
Scanned FLT images are
modern art … not science!
FLT
IMA
Steps of Calibration
Search ArXiv for “McCullough Deming”
Direct Image
We use the direct image mainly to verify the positions and fluxes
of neighboring stars in the field of view.
Steps of Calibration
1. Discard data from first orbit (optional) of four or five orbits per visit
(per transit or eclipse).
2. Start with IMA files (or make your own from RAW files (tedious) )
A. Dark-corrected
B. Nonlinearity corrected
C. Flagged-pixel masks (optional)
3. Sort into Forward & Reverse scanning directions
A. POSTARG2 keyword is helpful
4. Form forward differences of consecutive reads.
5. Divide by flat field
A. It’s useful to ID bad pixel & cosmic ray hits.
B. We use F139M but it hardly matters because …
C. We eventually divide by the best-possible self-calibrated “flat,” derived from outof-transit data before and after the transit event.
6. Interpolate values for bad pixels and cosmics.
7. Measure and subtract sky value from each image.
8. Sum over appropriate rows and columns to convert 2-D spectrum to 1-D
spectrum.
9. Shift/interpolate 1-D spectra to common grid (optional).
10. Calibrate column  wavelength.
11. Smooth 1-D spectra (optional; cf. Deming et al. 2013). Of course,
smoothing reduces noise and spectral resolution (see McCullough et al.
2014 for factor (0.817) relating box-car & Gaussian smoothing).
12. Form exoplanet spectrum = (OUT - IN)/OUT = 1 – IN/OUT
Wiggles, Drifts and Smoothing
Swain 2013
ArXiv preprint
Wavelength Calibration
In McCullough et al. 2014, we used a calibrated, ground-based,
higher-resolution spectrum of a similar star, and transformed it
iteratively (by smoothing, shifting, stretching, re-binning) until it
best matched our observed WFC3 stellar spectrum.
Others (e.g. Bean, Kriedberg & Stevenson) have used proper WFC3
distortion-parameterized wavelength calibration files, and the direct
image to set the zero point. This is tedious and prone to
bookkeeping blunders. It may be necessary for very long scans (>>
100 rows) due to distortion. On the other hand, because we
generally use 4-column-wide bins in final spectra, small shifts due
to distortion hardly matter.
Limb Darkening & Star Spots
The presumed limb-darkening model parameters do not affect the
differential transmission spectrum much at all: its shape and relative
heights of features remain the same. However, limb-darkening critically
affects the white-light transit depth, and hence comparisons with other
pan-chromatic observations. The white-light transit depth also is affected
by star spots away from the transit chord (especially for HD 189733b).
Caveat Emptor: rawima
Zero signal array (522x522)
Last ima.fits
Order 1
Star A
Star B
1800 e
38000 e
800 e
Scan direction
Ragged edge of “if” statement within signal box at ~260
e/pixel in 0th read.
Order 2
Summary
1. Transiting exoplanet spectra require very high S/N (10,000:1 or more).
2. Scanning records 10-100 times more photons than staring.
3. Scanning prevents saturation of very bright stars.
4. Analysis of scanned spectra is easier than that of staring-mode spectra.
A. Scanning homogenizes over pixels and pixel phase.
B. Relaxes calibration needs, especially for non-linearity of each pixel.
C. No need for MCMC modeling to bound uncertainties.
1. Scanning produces so-much-better results than staring, nearly all
exoplanet spectroscopy observations are now using scanning.
A. Possible exception = a very close double star (scanning will overlap spectra)
Questions?