Instrument Science Report STIS 97-14
Absolute Flux Calibration for STIS
First-Order, Low-Resolution Modes
Ralph Bohlin, Space Telescope Science Institute
Nicholas Collins, Hughes STX/LASP/GSFC
Anne Gonnella, Space Telescope Science Institute
February 1998
ABSTRACT
Point source sensitivity curves are derived for the first order modes G140L, G230L,
G230LB, G430L, and G750L, which collectively span a wavelength range from 1140Å to
10320Å. The curves are determined by comparing wide slit (52" x 2") observations of the
spectrophotometric standard star GD153 to a pure hydrogen white dwarf model. The calibration is checked by comparing a calibrated STIS observation of the spectrophotometric
standard star G191B2B to its pure hydrogen model. Except for G140L, there is no good
evidence for changes in the sensitivity beyond ~1% through the end of 1997. Except for the
G140L and the long wavelength part of G750L, the absolute flux calibration is accurate
for observations in 1997 to ~3% for stars centered in the 52x2" slit. Preliminary results on
the vignetting analysis for the first order modes suggest that the smooth changes in sensitivity as a function of distance along the slit are less than ±2 % for all first order modes,
except for the extremities of G140L, where 5% differences in detector QE are observed.
While the 52x2" slit is ideal for photometric continuum determinations, a narrower slit
should be used for line profile measurements in order to minimize the contamination from
the impure light in the wings of the PSF. For the narrowest slits, local inhomogeneities in
the slit width cause a variable throughput along the slit (see the STIS Foibles page on the
STIS Instrument WWW pages).
1. Introduction
A sensitivity curve, i.e. an absolute flux calibration, is used to determine the flux of an
observed point-source spectrum by
F=C/S
1
where F is the calibrated absolute flux spectrum, C is the observed spectrum in counts
pixel-1 second-1, and S is the sensitivity as a function of wavelength (Bohlin, Lindler, &
Keyes 1995). The sensitivity is derived by
S = Cstd / Fstd
where Cstd is the countrate spectrum of a spectrophotometric standard star Fstd is the
known absolute flux spectrum of the same standard in erg cm-2 s-1 Å-1.
The pixel in the units for C refers to the dispersion direction, while the point source
response is integrated over a fixed extraction height in the cross-dispersion direction on a
two-dimensional detector.
Before the 1998 February delivery of these throughputs that are based on in-flight
observations, the STScI CALSTIS pipeline utilized pre-launch estimates in the photometric throughput tables to flux calibrate data. When the rootname_sx2 files for low resolution
first order data calibrated with the prelaunch estimates are used to find the stellar flux as
described in the Data Handbook (page 23-5 version 3.0) or when a point source spectrum
is extracted from the flat fielded data with the X1D IRAF task, the fluxes are generally
within a factor of 1.5 of the correct fluxes. For accurate fluxes, GO’s should recalibrate
their data (X2D or X1D IRAF tasks) using the latest reference files.
2. STIS Observations
Observations of the pure hydrogen white dwarf GD153 (Bohlin, Colina, & Finley
1995) are used to define the sensitivity curve, since these data are the only set of observations of a fundamental standard in all five low-resolution spectral modes. GD153 is a
preferred calibration standard, because the only lines are from HI and because Bohlin
(1996) defined the FOS calibration with a set of four pure hydrogen WD models that
includes GD153. An input spectrum with many lines can confuse the derivation of a sensitivity curve, if the reference spectral line profiles do not perfectly match those of the
observed spectrum.
All input spectra were obtained using the clear 52"x2" aperture. Relative transmission
corrections are required to compute absolute fluxes for other apertures.
Two observations comprise the average input spectrum for the far-UV MAMA mode
G140L. Only one near-UV (G230L) observation of GD153 exists in the 52x2 aperture.
For each of the three CCD modes (G230LB, G430L, and G750L), seven observations
obtained at the end of SMOV produce high signal-to-noise average spectra. The position
of these seven spectra repeated to 1 pixel on the CCD; and the full range of scatter of relative response is better than 1% in broad bands. However, spectra obtained at other times
and at non-standard locations on the CCD show an additional scatter of 1–2%. The component spectra that make up the average spectrum for each mode are listed in Table 1; and
2
the spectral extraction heights are 11 pixels for G140L and G230L, and 7 pixels for
G230LB, G430L, and G750L (Leitherer & Bohlin 1997). The background is extracted 300
pixels above and below the spectrum and is typically much less than 1% of the continuum
signal.
Each of the seven component G750L spectra is corrected for the long wavelength CCD
fringing with a tungsten flat obtained in the 0.3x0.09 slit. This correction starts with the
extracted 1-D spectrum of the tungsten lamp for the three observation that have a 0.3x0.09
contemporaneous flat, O3TT46040, O3TT47040, and O3TT48040. The average of these
three spectra is normalized to unity with a spline fit; the amplitude of the fringes longward
of 6620 Å is reduced by 11%; an offset shift with respect to each stellar spectrum is computed; and this shifted 1-D tungsten flat is divided into each 1-D spectrum of the star.
Below 6620 Å, the tungsten flat is set to unity, since the noise level is worse when the
average flat is applied. This procedure works, because all of the seven GD153 spectra lie
within a pixel of the same location on the CCD. However, the noise level for the earliest of
the seven, O3TT42040, is larger than for the other six, so that the final average GD153
spectrum for G750L is composed of only six observations. Details of the general applicability of this procedure, including a quantitative comparison with alternative techniques
will be presented in a fringe ISR.
3
Table 1. Observations with 52x2 arcsec slit
Optical
Element
Observation
Date
Target
O43J01QAM
09/07/97
GD153
G140L
60.0
2700
7097
O3ZX08HHM
13/07/97
GD153
G140L
187.1
8500
7096
O3ZX08HLM
13/07/97
GD153
G230L
187.1
5200
7096
O3TT42010
21/05/97
GD153
G230LB
600.0
21,400
7063
O3TT43010
28/05/97
GD153
G230LB
600.0
21,400
7063
O3TT44010
04/06/97
GD153
G230LB
600.0
21,400
7063
O3TT45010
10/06/97
GD153
G230LB
600.0
21,400
7063
O3TT46010
18/06/97
GD153
G230LB
600.0
21,400
7063
O3TT47010
25/06/97
GD153
G230LB
600.0
21,400
7063
O3TT48010
01/07/97
GD153
G230LB
600.0
21,400
7063
O3TT42020
21/05/97
GD153
G430L
252.0
16,700
7063
O3TT43020
28/05/97
GD153
G430L
252.0
16,700
7063
O3TT44020
04/06/97
GD153
G430L
252.0
16,700
7063
O3TT45020
10/06/97
GD153
G430L
252.0
16,700
7063
O3TT46020
18/06/97
GD153
G430L
252.0
16,700
7063
O3TT47020
25/06/97
GD153
G430L
252.0
16,700
7063
O3TT48020
01/07/97
GD153
G430L
252.0
16,700
7063
O3TT43040
28/05/97
GD153
G750L
3240
76,400
7063
O3TT44040
04/06/97
GD153
G750L
3240
76,400
7063
O3TT45040
10/06/97
GD153
G750L
3240
76,400
7063
O3TT46040
18/06/97
GD153
G750L
2282
53,800
7063
O3TT47040
25/06/97
GD153
G750L
2282
53,800
7063
O3TT48040
01/07/97
GD153
G750L
2282
53,800
7063
4
Exposure
Time (sec)
Mean
Counts px-1
Rootname
Prop ID
3. Reference Standard Star Flux
The reference spectrum, GD153_MOD_002, is a pure hydrogen white dwarf model
normalized to Landolt’s visual photometry (Bohlin 1996). The spectrum can be obtained
from the Calibration Data Base System (CDBS) at:
http://www.stsci.edu/ftp/instrument_news/Observatory/cdbs/
astronomical_catalogs_alt.html
4. Sensitivity Results
For each optical mode, the reference spectrum is integrated to match the resolution of
each co-added observation and is then divided into the observed spectrum to produce a
sensitivity curve in units of
–1
–1
counts ⋅ pixel ⋅ second
----------------------------------------------------------------–1
–2
–1
ergs ⋅ second ⋅ cm ⋅ Å
At the shortest wavelengths of each mode, the sensitivity drops rapidly with wavelength, so that the uncertainty in the flux calibration in dominated by uncertainties in
wavelength. In order to derive the proper sensitivity calibration, the wavelength scale must
be as observed in the instrumental frame of rest, i.e., with NO velocity correction. The
standard star model spectrum must be shifted to coincide with the observations, i.e., the
reference spectrum wavelength scale must be shifted by the radial velocity of the star plus
the heliocentric correction for the earth and HST velocity vectors. Consequently, the application of derived sensitivity curves must be to the counts/sec with the instrumental rest
frame wavelengths. The observed radial velocity of G191B2B is +22 km/s, which includes
the gravitational component (Reid & Wegner, 1988), while a value of 50 +-30 km/s for
GD153 is estimated from the STIS observations of the Balmer lines. From the proper
motion and the photometric distance of 77 pc, Finley (private communication) calculates a
tangential velocity of 73 km/s for GD153.
Splines with evenly spaced nodes are fit to each ratio in order to obtain a smooth sensitivity curve. Each fit is refined by adding nodes at wavelengths where the slope is large.
Table 2 lists the number of nodes and the nominal wavelength range for each mode. The
wavelength region 1200 Å – 1225 Å in mode G140L is masked to exclude the residuals at
the strong Ly α feature from the fit. Small residuals at the Balmer lines on G430L and
G750L are caused by slight differences in resolution between STIS and the model
spectrum.
The sensitivity curves and their residuals are shown in Figures 1 and 2. The spline fits
are represented by the dashed line in each plot, while the spline nodes are plotted with diamonds. The large residual at Ly α is caused by neighboring continuum light in the far
5
wings of the PSF. Figure 3 is a plot of this line, where the model line profile (smooth solid
line) agrees well with FOS observations (dashed line). However, the central depth of the
line in the STIS spectrum (heavy jagged solid line) is about 10% higher in units of the
continuum. Except at wavelengths beyond ~8500Å where there is considerable scattering
within the substrate of the CCD, the scattered light level shown in Figure 3 approaches a
worst case, because the PSF from the OTA has more energy in the wings from the microroughness of the mirrors, because the effect of the voltage on the repeller wire in the FUV
MAMA is to add signal to the PSF wings, and because the wide 2" slit includes the OTA
PSF wings. See Figure 6 in Leitherer & Bohlin (1997) for plots of the STIS PSF in the
direction perpendicular to the dispersion. Since the fit in Figure 1 ignores this contaminating light, the shape of the sensitivity curve in the 1216Å region is the same as a pointsource continuum calibration based on a STIS spectrum corrected for the impure light.
However, at wavelengths shortward of ~1160Å and at the short wavelength cutoff of
G230L, out-of-band light dominates the observed spectrum and makes absolute fluxes
unreliable.
Out-of-band light is substantially reduced with the 0.2" and smaller slits. Additional
data are required to quantify the impure light vs. slit width relation and guide the GO in
choosing the right slit width for measuring line profiles and equivalent widths. A subsequent ISR will discuss the choice of the optimum slit in detail; and in the interim, GOs are
directed to the STIS Foibles page on the STIS WWW page for the latest information.
Preliminary results on the vignetting analysis for the first order modes suggest that the
smooth changes in sensitivity as a function of distance along the slit are less than ±2 % for
all first order modes, except for the extremities of G140L, where ±5% differences in detector QE are observed. While the 52x2" slit is ideal for photometric continuum
determinations, a narrower slit should be used for line profile measurements in order to
minimize the contamination from the impure light in the wings of the PSF. For the narrower slits, local inhomogeneities in the slit width can cause a variable throughput along
the slit (e.g., see the STIS Foibles page on the STIS Instrument WWW pages).
5. Uncertainties
The residuals shown in Figures 1 and 2 are determined by dividing each curve by its
spline fit and are plotted for the wavelength ranges listed in Table 2. Table 2 lists the percent root-mean-square residuals averaged over all wavelengths for each mode. Average
RMS residuals are listed for three wavelength ranges for mode G750L to show how the
scatter increases in the long wavelength region, where fringing is the worst.
Unless a peakup is done in a small aperture, pointing errors can cause wavelength
errors of up to a pixel (~2-3 σ) on the MAMAs and up to a half pixel on the CCDs. The
error in absolute fluxes that is caused by wavelength errors depends of the slope of the sensitivity curve vs. wavelength but can be over a percent near the steep sensitivity cutoffs,
6
Table 2. Nodes and Average Percent RMS Deviation for Each Mode
Optical
Element
Wavelength Range (Å)
Spline Nodes
Avg
σ rms
G140L
1140–1200, 1225–1712
30
1.35a
G230L
1601–3139
27
1.01
G230LB
1671–3065
31
0.40
G430L
2889–5676
34
0.79
G750L
5249–8000
42b
0.44
G750L
8000–9000
0.51
G750L
9000–10219
0.73
a. Residuals in the region of the strong Ly α feature (1220–1225 Å) are
not included in the computation of σrms.
b. Number of nodes for full G750L wavelength range.
for example below ~1200Å on G140L. If no wavecal spectrum is taken with the data
before moving the Mode Select Mechanism, wavelength errors of a few pixels are possible
because of MSM repeatability limits. Wavelength errors can often be corrected, using the
stellar radial velocity and a spectral feature of known wavelength in the data spectrum.
In general, wavelength uncertainties and the counting statistics of the calibration
observations, result in negligible errors in the smooth STIS sensitivities derived here for
the 1997 epoch. Uncertainties in the absolute sensitivities are dominated by systematic
uncertainties in the reference stellar spectra, which are based on model atmosphere calculations for pure hydrogen WDs (Bohlin, Colina, and Finley 1995). Assuming no error in
the model atmosphere calculation, the uncertainty is determined by the shape of the model
spectrum vs. wavelength and by the normalization to the V magnitudes of Landolt, as
explained in (Bohlin, Colina, and Finley 1995). To recap: Landolt photometry is repeatable to 1σ of ~0.004 mag. Allowing for some uncertainty in the normalization of a flux
distribution by the broadband V filter photometry, the reference standard star spectra and
derived STIS calibration should be accurate to ~1% (1σ) in the V band wavelength region.
Additional uncertainty at other wavelengths is introduced by uncertainty in the continuum
slope of the model spectrum, which is dominated by uncertainty in the temperature of the
model. For the worst case of GD71 at 1150Å, the formal 1σ uncertainty is 0.35% due to
the temperature uncertainty of 100K. The main uncertainty in the physics is the HummerMihalas occupation probability formalism, which causes uncertainties of up to 1%
between the Balmer lines. So an assignment of 1σ=1% to flux uncertainties that are
caused by slope error is conservative and leads to to a formal 3σ=4% uncertainty outside
the V band region and 3% at 5000–6000Å. Future revisions to the STIS flux calibration
7
should be based on three stars, instead of just GD153, so that the uncertainties would be
formally reduced by a factor of 1.7.
There is no evidence for changes in the sensitivity or scatter in the STIS spectrophotometry beyond ~1% through the end of 1997, except for G140L where initial results
suggest a drop in sensitivity by up to 7% per year. Fringing in the CCD limits photometric
precision to ~2% longward of 8500Å. By mid-1998, the sensitivity monitoring observations should be sufficient to more precisely quantify any changes in sensitivity.
6. Application of the Sensitivity Curves to Observations of G191B2B
Figures 4a-c show the residuals of STIS fluxes of GD153 and G191B2B calibrated
with the sensitivity curves described in this paper. The comparison is with the standard
star models from Bohlin (1996). Residuals for the G191B2B fluxes confirm the above estimates of uncertainty in STIS fluxes; but since GD153 provides the flux calibration, its
residuals are unity, except for noise and small differences in the Balmer line cores. The
small (<1%) systematic deviations from unity for the G230LB and G430L observations of
G191B2B in the lower panels of Figures 4a-b may be caused by sensitivity changes over
the approximately 5 months between the epochs of the GD153 baseline observations in
SMOV and the G191B2B observations in 1997 October-November or else by different
sensitivities at the different locations on the CCD. This difference in position is 5–6 px in
the dispersion direction and 2-3 px in the perpendicular direction.
The systematic deviations from unity for G750L have additions to the list of possible
causes: Beyond 6220 Å, contemporaneous flats are used with a normalization to unity that
may be in error by ~1%. (No flat field correction is done below 6220 Å, for G750L.)
Beyond the Paschen jump at 8200 Å, the models that define the standard stars are uncertain by an additional 1–2%.
Acknowledgments
We thank Don Lindler and Phil Plait for essential data reduction advice. D. Finley and
R. Saffer found the reference for the radial velocity of G191B2B. Thanks to S. Baum for
careful review and suggested improvements to this ISR.
References
•
Bohlin, R. C. 1996, AJ, 111, 1743.
•
Bohlin, R. C., Colina, L., & Finley, D. S. 1995, AJ, 110, 1316.
•
Leitherer, C. & Bohlin, R., 1997, Instrument Science Report, STIS 97-13, (Baltimore:
STScI).
•
Reid, N., & Wegner, G., 1988, ApJ, 335, 953.
8
Figures
Figure 1: Sensitivity curve and residuals for optical element G140L, derived from 2
averaged observations of GD153.
9
Figure 2a: Sensitivity curve and residuals for optical element G230L
10
Figure 2b: Sensitivity curve and residuals for optical element G230LB
11
Figure 2c: Sensitivity curve and residuals for optical element G430L
12
Figure 2d: Sensitivity curve and residuals for optical element G750L at central wavelength 7751 Å
13
Figure 3: Line profile at Ly α of GD153, illustrating the good agreement of the model
(solid line) and FOS (dashed line) in comparison to the STIS data (heavy jagged line)
through a wide 2” slit. The excess scattered light will be much less for observations
through narrow STIS slits.
14
Figure 4a-4c: Residuals of STIS fluxes for GD153 and G191B2B in comparison with the standard star model for G230LB (4a), G430L (4b), and G750L (4c). Since the flux calibration is
based on GD153, its residuals are unity, except for noise and small differences in the Balmer
line cores.
Figure 4a
Figure 4b
15
Figure 4c
16
17