WFC3 contains three grism elements: the G280 in the UVIS channel and the
G102 and
G141 in the IR channel. The grisms provide slitless spectra over the whole field of view of each channel. These spectroscopic modes present the well-known advantages and disadvantages of slitless spectroscopy. The chief advantage is large-area coverage, which enables spectroscopic surveys. Among the disadvantages are overlap of spectra, high background from the integrated signal over the passband, and modulation of the resolving power by the sizes of dispersed objects.
In the UVIS channel, the G280 grism provides spectroscopy over a useful wavelength range of 200-400nm, at a dispersion of ~14Å per pixel in the first order. The two grisms for the IR channel cover the wavelength ranges 800-1150nm (G102) and 1075-1700nm (G141). The dispersions are 24.5 and 46.5Å per pixel, respectively.
The primary aim of the reduction of WFC3 slitless spectra is to provide one-dimensional wavelength- and flux-calibrated spectra of all objects with detectable spectra. The reduction presents special problems because of the dependence of the wavelength zero-point on the position of the object in the field, the blending of spectra in crowded fields, and the need for flat-field information over the whole available wavelength range.
The aXe software was developed by the Space Telescope European Coordinating Facility (ST-ECF) for use with ACS slitless modes but, can also be used for the reduction of WFC3 spectroscopic data. This package enables automatic and reliable extraction of large numbers of spectra from individual images.
A slitless Spectroscopy Workshop was held at STScI in November, 2010, with a focus on using the aXe software. The Web cast for this event can be found at:
The normal method for taking WFC3 slitless spectra is to take a pair of images, one direct and one dispersed, of each target field, with no shift in position between the two. The direct image provides the reference position for the spectrum and thus sets the pixel coordinates of the wavelength zero-point on the dispersed image.
The WFC3 UVIS and IR grisms have some unique properties that result in different types of issues associated with data for the two different channels. These are discussed in more detail in later sections. There are, however, some common issues associated with all grism observations, which we highlight here.
The brightest objects produce spectra that can extend far across the detector. This is especially problematic for the UVIS G280 grism, where the relative throughput of the higher spectral orders is significant. These spectra provide a strong source of contamination for fainter sources. In addition, the higher order spectra are increasingly out of focus and thus spread in the cross-dispersion direction. Bright stars also produce spatially extended spectra formed by the wings of the PSF.
In slitless spectroscopy the object itself provides the “slit”. The WFC3 PSF has a high Strehl ratio over most of the accessible wavelength range of the grisms and therefore the degradation of point sources beyond the theoretical resolution is minimal. The spectral resolution for an extended object, however, will be degraded depending on the size and light distribution in the object and spectral features will be diluted.
The grism 0th order is only detectable for brighter objects observed with the IR grisms because it contains only a small fraction of the total flux. This faint feature is therefore easily mistaken for an emission line. The direct image can be used to determine the position of the 0th order for each source, which allows the 0th order feature in the dispersed images to be distinguished from emission lines. For the UVIS G280 grism, the 0th order has high throughput and is therefore more readily distinguished from emission features. The high throughput of the G280 0th order also means that it will often be saturated in long exposures, which leads to CCD charge bleeding and potential contamination of adjacent spectra.
The background in a single grism image pixel is the result of the transmission across the whole spectral range of the disperser and can therefore be high, depending on the spectrum of the background signal. The G280 grism, for example, produces relatively low background compared to the IR grisms, because of the faintness of the sky in the near-UV and optical. The IR grism background includes not only signal from the sky, but also thermal emission from the telescope and WFC3 optics. The detected background in the IR grisms shows a distinct two-dimensional structure that is due to the overlapping of the background spectral orders. This background needs to be carefully removed before extracting the spectra of targets.
Because of the high sensitivities of the WFC3 grisms, observations of moderately crowded fields can produce many instances where spectra overlap. It is important to know if a given spectrum is contaminated by that of a neighbor. This can be done by obtaining grism observations of the same field at different telescope roll angles, which improves the chances of cleanly extracting the spectrum for a given target.
There will inevitably be cases where objects outside the field of view result in spectra getting dispersed into the field, resulting in contamination of sources within the field. This is more serious for the G280 where the spectra are long relative to the size of the detector. In such cases reliable wavelengths can not be determined for the extra-field object unless the 0th order is also present. Even then, the wavelength zero point will be relatively uncertain because the 0th order is somewhat dispersed and therefore difficult to localize.
The direct image of a direct-plus-grism image pair can be fully reduced by calwf3, including bias subtraction, dark subtraction, flat-fielding, and computation of the photometric zero-point in the header. However, because of the individual wavelength coverage of each object in the dispersed image, the reduction of the dispersed image by
calwf3 is slightly more restricted. In contrast to direct images, no single flat-field image can be correctly applied to grism images, because each pixel contains signal arising from different wavelengths. Flat-fielding must therefore be applied during the extraction of spectra once the wavelength corresponding to each pixel is known. Each pixel receives a flat-field correction dependent on the wavelength falling on that pixel, as specified by the position of the direct image and the dispersion solution. So during
calwf3 processing the
FLATCORR step is still performed, but the division is done using a special flat-field reference file that only contains information on the relative gain offsets between the different detector amplifier quadrants. This allows the
FLATCORR step to still apply the gain correction (converting the data to units of electrons for UVIS or electrons per second for IR) and thus also corrects for offsets in gain between the various quadrants of the detectors.
The calwf3 flt products should then be the starting point for all subsequent reduction of slitless data with
aXe or other software. The units of the data in the SCI and ERR extensions of these files are electrons for UVIS and electrons per second for IR. The primary output of
aXe is a file of extracted spectra (
spc). This is a FITS binary table with as many table extensions as there are extracted beams.
The common approach to dithering WFC3 imaging data, in order to improve the sampling of the PSF and to allow for the removal of bad pixels, applies equally well to slitless spectroscopy data. For long grism observations the data taking is typically broken into several sub-orbit dithered exposures.
The MultiDrizzle task, which is normally used to correct for the geometrical distortion of WFC3 and combine dithered exposures, is not generally applicable to grism observations. This is due to the fact that the spatial distortion correction would only be applicable to the cross-dispersion direction of grism images. For similar reasons, the combining of dithered grism images before extracting spectra is not a good idea. Every detector pixel has a different spectral response, which has not yet been corrected in the calibrated two-dimensional images (see the preceding section on flat-fielding). Combining dithered grism images before extraction will combine data from different pixels, making it difficult or impossible to reliably flat-field and flux-calibrate the extracted spectra. Extracted spectra from dithered images can be properly combined into a final spectrum using the
aXedrizzle task in the
aXe package.
MultiDrizzle processing of dithered grism exposures can, however, be useful for simple visual assessment of spectra in a combined image and for the purpose of flagging cosmic-ray (CR) hits in the input
flt images. When
MultiDrizzle detects CR’s in the input images it inserts flags to mark the affected pixels in the DQ arrays of the input
flt files. The
aXe spectral extraction can then be run on these updated
flt images and utilize the DQ flags to reject bad pixels. This is very useful for rejecting the large number of CR hits that occur in long UVIS G280 exposures. It is not as necessary for IR grism images, because the IR
flt files have already had CR’s rejected by the
calwf3 up-the-ramp fitting process.
The filters most often used for obtaining a direct image in tandem with the G280 grism are the F300X and F200LP. The direct image provides the reference position for the spectrum and thus sets the pixel coordinates of the wavelength zero point on the dispersed image. The G280 wavelength zero point is generally calibrated to an accuracy of about 1 pixel. It is not possible to use the 0th-order image of a source in a G280 exposure to establish the source position, because the 0th-order is somewhat dispersed.
Spectra produced by the G280 grism are oriented in WFC3 images with the positive spectral orders to the left (lower x-axis pixel index) of the 0th-order spot, with wavelength increasing to the left. Negative orders are located to the right, with wavelength increasing to the right. The +1st order extends to the left of the 0th order a distance of about 1/4 of the image width. The throughput of the +1st order of the G280 is only somewhat larger than that of higher orders and of the negative orders. This leads to heavy overlap of the orders at wavelengths greater than ~400nm. In addition, there is curvature of the spectra at the blue ends of the orders. The amplitude of the curvature is about 30 pixels in the detector y-axis. Due to the relatively significant throughput of the higher orders, the spectra of very bright objects may extend across nearly the entire field of view of the detector. See WFC3 ISR 2009-01 for more details on the characteristics and calibration of the G280 grism.
As an example, Figure 7.3 shows a G280 image of the Wolf-Rayet star WR-14, which is used as a wavelength calibrator. Superimposed on the dispersed image is a F300X image, which illustrates the relative location of the direct image of the source (circled in
Figure 7.3). The full 4096-pixel x-axis extent of the detector is shown, which is completely filled by the positive and negative orders of this bright source.
Figure 7.4 shows a zoomed view of the first several positive spectral orders of this source, where wavelength increases to the left. Notice how the blue end of each order curves upwards, and that at longer wavelengths (greater than ~400nm) there is significant overlap of adjacent orders. Very bright sources produce spectra in which orders up to 6-8 can be detected. These spectra, which in principle can be analyzed (although dispersion solutions have been determined for only the first few orders), provide a strong source of contamination for the spectra from fainter objects. In addition, the higher order spectra are increasingly out of focus and thus spread in the cross-dispersion direction.
The dispersion of the G102 grism is high enough that only the positive 1st and 2nd order spectra generally lie within the field of the detector. For the lower-dispersion G141 grism, the 0th, 1st, 2nd, and 3rd order spectra lie within the field for a source that has the 1st order roughly centered. The IR grisms have the majority (~80%) of their throughput in the +1st order, resulting in only faint signals from the other orders. The trace of the observed spectra are well described by a first-order polynomial, however the direct-to-dispersed image offset is a function of the source position in the field. The tilt of the spectra relative to the image axes are small, being only 0.5-0.7 degrees. Typical filters used for obtaining companion direct images are the F098M and F105W for the G102, and the F140W and F160W for the G141. Other medium- and narrow-band filters can be used when necessary to prevent saturation of very bright targets.
The dispersion direction of the IR grisms is opposite to that of the G280, with the positive spectral orders appearing to the right of the 0th order and wavelength also increasing to the right. Examples of G102 and G141 observations of the flux calibration standard star GD-153 are shown in
Figure 7.5 and
Figure 7.6, respectively.
The software package aXe provides a streamlined method for extracting spectra from WFC3 slitless spectroscopy data.
aXe is distributed as part of the
STSDAS software package at:
There is a detailed aXe manual and a cookbook specific to WFC3 grism data reduction, both of which are available from the
aXe Web pages, so only a brief outline of its use is presented here.
|
•
|
Make a direct image source catalog. This step consists of identifying and cataloging sources in the direct image of the direct-grism image pair. The source positions and sizes are used later to define extraction boxes and calculate wavelength solutions in the extraction step. The source information is often derived from a MultiDrizzled combination of direct images.
|
The starting point is always a set of dispersed slitless images and the derived catalog of objects in the images. Information about the location of the spectra relative to the position of the direct image, the tilt of the spectra on the detector, the dispersion solution for various orders, the name of the flat-field image and the sensitivity (flux per Å/e
−/sec) table are stored in a configuration file, which enables the full calibration of extracted spectra. For each instrumental configuration the configuration files and all necessary calibration files for flat-fielding and flux calibration can be downloaded from the
aXe Web pages.
aXe has two different strategies for removal of the sky background from the spectra.
The first strategy is to perform a global subtraction of a scaled ``master sky'' frame from each input grism image at the beginning of the reduction process. This removes the background signature from the images, so that the remaining signal can be assumed to originate from the sources only and is extracted without further background correction in the
aXe reduction. Master sky frames are available for download from the
aXe Web page.
The second strategy is to make a local estimate of the sky background for each beam by interpolating between the adjacent pixels on either side of the beam. In this case, an individual sky estimate is made for every beam in each science image. This individual sky estimate is processed (flat-fielded, wavelength calibrated) parallel to the original beam. Subtracting the 1D spectrum extracted from the sky estimate from the 1D spectrum derived from the original beam results in the pure object spectrum.
The primary output of aXe is the file of extracted spectra (SPC). This is a multi-extension FITS binary table with as many table extensions as there are extracted beams. The table contains 15 columns, including wavelength, total and extracted and background counts and their errors, the calibrated flux and error, the weight and a contamination flag. The primary header of the SPC table is a copy of the header of the frame from which the spectrum was extracted.
aXe can also create a 2-d “stamp” image for each beam, for the individual inspection of single beams. The stamp images of all beams extracted from a grism image are stored as a multi-extension FITS (STP) file with each extension containing the image of a single extracted beam. It is of course also possible to create stamp images for 2-d drizzled grism images.
The output products from aXe consist of ASCII files, FITS images and FITS binary tables. The FITS binary tables can be accessed using the tasks in the stsdas.ttools package and wavelength-flux plots, with error bars, can be plotted using stsdas.graphics.stplot.sgraph.
When there are many detected spectra on a single image, then a dedicated task aXe2web is available. aXe2web creates html pages consisting of direct image cut outs, stamp images and 1-d spectra for each extracted beam. This enables convenient browsing of large numbers of spectra or the publishing of
aXe spectra on the Web with minimal interaction.
The ST-ECF ACS/WFC Grism Final Release (2010, July 6) on the HLA page above contains 47919 extractions from 32149 unique objects that have been uniformly reduced by the ECF team using the aXe software.
The WFC3 grism dispersion solutions were established by observing both astronomical sources with known emission lines (e.g., the Wolf-Rayet star WR-14 and the planetary nebula Vy2-2; see WFC3
ISR 2009-17 and
ISR 2009-18) and ground-based monochromator sources (see WFC3
ISR 2009-01 and
ISR 2008-16). The field variation of the dispersion solution was mapped by observing the same source at different positions over the field. The internal accuracy of these dispersion solutions is good to ~0.25 pixels for the IR grisms (~6Å and ~9Å for the G102 and G141, respectively), and to ~1 pixel (~14Å) for the UVIS G280.
For a given object the accuracy of the assigned wavelengths depends most sensitively on the accuracy of the zero point and the transfer of the zero point from the direct to the slitless spectrum image. Provided that both direct and slitless images were taken with the same set of guide stars, systematic pointing offsets less than 0.2 pixels can be expected. For faint sources the error on the determination of the object centroid for the direct image will also contribute to wavelength error. Realistic zero point errors of up to 0.3 pixels are representative.
The sensitivity of the dispersers was established by observing a spectrophotometric standard star at several positions across the field. The sensitivity (
aXe uses a sensitivity tabulated in ergs/cm
2/sec/Å per detected e
−) was derived using data flat-fielded by the flat-field cube. Results for the IR grisms show 4-5% differences in the absolute flux of spectra located near the center of the field as compared to those near the field corners. This is clear evidence for a large-scale variation in the overall illumination pattern in the grism flat-field data cubes. Additional field-dependent flux calibration observations are planned, which will enable such corrections to be implemented.
