Документ взят из кэша поисковой машины. Адрес оригинального документа : http://www.stsci.edu/hst/HST_overview/documents/calworkshop/workshop2005/papers/meurer_grism.pdf Дата изменения: Wed Apr 5 00:02:25 2006 Дата индексирования: Sun Dec 23 23:44:12 2007 Кодировка:

The 2005 HST Calibration Workshop Space Telescope Science Institute, 2005 A. M. Koekemoer, P. Goudfrooij, and L. L. Dressel, eds.

Selection and Characterization of Interesting Grism Spectra
Gerhardt R. Meurer Department of Physics and Astronomy, The Johns Hopkins University, 3400 N. Charles Street, Baltimore, MD 21218 Abstract. Observations with the ACS Wide Field Camera and G800L grism can produce thousands of sp ectra within a single WFC field producing a p otentially rich treasure trove of information. However, the data are complicated to deal with. Here we describ e algorithms to find and characterize sp ectra of emission-line galaxies and sup ernovae using tools we have develop ed in conjunction with off-the-shelf software.

1.

Introduction

The G800L grism combined with ACS's Wide Field Camera is a p owerful combination for obtaining thousands of sp ectra with relatively modest outlay of HST time. However, the resulting images are difficult to interpret due to a numb er of p eculiarities including: (1) strong spatially varying sky background; (2) a p osition-dep endent wavelength solution; (3) the wide sp ectral resp onse: a three-dimensional flat field and modeling of the wavelengths contributing to each pixel is required for precise flat fielding; (4) tilted sp ectra with resp ect to the CCD grid (the tilt varies over the field); (5) each source is disp ersed into multiple orders resulting in much overlap - deep images b ecome confusion limited; (6) zeroth-order images of compact sources can easily mimic the app earance of sharp emission features; and (7) the low resolution (R 90 for p oint sources) means that only high Equivalent Width (EW) features can b e discerned, while most familiar features are blends. The Space Telescop e ­ Europ ean Coordinating Facility has done an excellent job of addressing these p eculiarities with the software package aXe (Pirzkal et al. 2001). Armed with it and a broad-band detection image, users can extract 1D and 2D sp ectra that are sky-subtracted, wavelength-calibrated, flat fielded, and flux calibrated with minimum effort. Here I describ e complimentary techniques I have develop ed to analyze WFC grism images. Sp ecifically, I describ e tools geared to finding emission-line sources and sup ernovae (SNe). Here I concentrate on my work with the ACS GTO team to search the Hubble Deep Field North (HDFN) for Emission Line Galaxies (ELGs) and work with the PEARS team to find SNe. 2. Initial Processing

aXe is designed so that it can work with a stack of individual dithered exp osures (the FLT or CRJ images), where the grism images have not b een flat fielded nor geometrically corrected. However, b oth flat fielding and drizzling can b e very useful. Application of the F814W flat corrects most gross blemishes and removes at least half the amplitude of large-scale sky variations (after geometric correction). Spurious dark sp ots may remain at the blue end of some sp ectra, but their amplitude will b e diluted if there are numerous small dithers. Their presence will have little impact on emission-line searches, while their sharpness means they are unlikely to b e confused with real absorption features.

95 c Copyright 2005 Space Telescop e Science Institute. All rights reserved.


96

G. R. Meurer

a

b

c

d

e

f

Figure 1: Steps in processing the grism and broad band detection images for finding ELGs using method B. Panels a and b show a 50в50 cutout of the grism image b efore and after subtracting a 13в3 median filtered version of the image. Panels c and d show cutouts of the detection image b efore and after the same filtering. The width of the cutout covers the full x range over which the counterpart to the source seen in panel b may reside. Panel e shows the collapsed 1D sp ectra of five rows centered on the emission line extracted from the grism image b efore (black (upp er) line) and after (blue (lower) line) median filtering. Panel f is the same but for the 1D cuts through the detection image. The shaded region with identification is derived from collapsing the the SE segmentation image of the detection image.


Selection and Characterization of Interesting Grism Sp ectra

97

Drizzle combining multiple dithered exp osures is feasible as long as the dither offsets are all within 6 ; then the alignment across the sp ectra will all b e correct to within 0.5 pixels. The resultant geometrically corrected images have first order sp ectra that are nearly horizontal across the image, and greatly decreased spatial variation in the wavelength solution. Drizzle combining also allows improved CR rejection, esp ecially when done with the ACS GTO pip eline Apsis (Blakeslee et al. 2002). A mask is used to mark or remove the zeroth order images. First the zeroth order sources in the grism image are matched to those in a broad band detection image. The sources are found with SExtractor (hereafter SE; Bertin & Arnouts, 1996) which is used to catalog the sources in b oth the detection and grism images. Only compact sources are matched. Their p ositions are used to define a linear transformation b etween the detection image and the zeroth order. The scaling ratio b etween the matched detection and zeroth order images is determined and used to model which pixels to mask. In the HDFN the images in F775W and F850LP are typically 32 and 21 times brighter, resp ectively, in count rate than the zeroth order counterparts. This scaling ratio is used to determine which pixels will have zeroth order counterparts that are brighter than sky noise level. The p osition of these pixels in the detection image are transformed to p opulate a mask for the grism image which is then grown by three pixels to account for the slight disp ersion in the zeroth order. Masked pixels are set to zero at the appropriate stage of the analysis.

3.

Finding Emission Line Galaxies

The ACS Science team observations centered on the HDFN consist of 3 orbits with G800L and F850LP (z850 ) and two orbits with F775W (i775 ). Two complimentary techniques for finding ELGs were employed on this field. A: Search 1D sp ectra. aXe is used to extract sp ectra of all SE cataloged sources in the detection image down to i775 = 26.5 AB mag. The flux calibrated sp ectra are then filtered by subtracting 13 pixel median smoothed sp ectra leaving only sharp features. Sources with p eaks having S/N > 4 are flagged as likely ELGs. The flagged sp ectra are classified by eye - broad absorption line sources are also flagged by this algorithm. These are usually M or K stars, but also include the two SNe in this field (Blakeslee et al. 2003) . The true ELGs have their emission lines fitted with Gaussians to derive line wavelength and flux. B: Search 2D grism image. The basis of this method is the observation that most emission line sources app ear to corresp ond to compact knots, not necessarily at the center of galaxies. Here we find the line emission in the grism image first and then pinp oint the emitting sources in the detection image, as illustrated in Fig. 1. Sharp ened versions of b oth the grism and direct images are made by subtracting a 13в3 median smoothed version of themselves. In the grism image, this effectively subtracts the continuum and removes crossdisp ersion structure. This image is then cataloged with SE . Ribb ons, typically covering five rows, centered on the y p osition of each source are extracted from b oth the sharp ened grism and direct images. Since the disp ersed sp ectrum lies primarily to the right of the direct image, the extracted ribb ons extend more to the left so that all p ossible sources that could have created the emission line are in the direct ribb on. The ribb ons are collapsed down to 1D sp ectra and cross correlated after the regions b eyond ± 13 pixels from the source in the grism image are set to 0.0. This is done so they do not contribute to the cross-correlation amplitude. Any knot within the detection ribb on will produce a p eak in the cross-correlation sp ectrum. The p osition of the p eak yields the offset b etween the knot and the line emission in the grism image. Using the wavelength solution for the grism, in principle one could derive the line wavelength from this offset. Instead, final measurements of the emission line quantities are obtained from 1D sp ectra of each knot extracted with aXe using the cross-correlation determined p osition of the star forming knot. As with method A, the emission line prop erties are measured with Gaussian fits to the sp ectra.


98

G. R. Meurer

Figure 2: Histogram of i775 AB magnitudes of the grism selected ELG sample in the HDFN (top panel) compared with the sp ectroscopic redshift sample of Cowie et al. (2004; middle panel) and the photometric redshift sample of Capak (2004; b ottom panel). In the upp er left of each corner we rep ort the total numb er of sources in the sample and the 25th, 50th (median), and 75th p ercentile i775 AB magnitudes.


Selection and Characterization of Interesting Grism Sp ectra

99

In the HDFN field we found 30:39 ELGs with methods A:B. For the most part, the same galaxies are found; 7:16 ELGs were uniquely found with methods A:B. For three ELGs we identified multiple emission line knots with method B. Figure 2 compares the i775 apparent magnitude distribution of the merged list of ELGs from our analysis compared to galaxies with sp ectroscopic and photometric redshifts in the same field. The grism ELGs, found with three orbits of HST time, are on average fainter than the galaxies with sp ectroscopic redshifts gathered over several years from the Keck 10m telescop es. 4. Line Identification

Line identification is a ma jor concern. Only seven of the ELGs in HDFN have two emission lines in our data. In those cases the lines can b e identified using the ratio of wavelengths which remains invariant with redshift. However one must b e careful with this technique since H /[OIII] = 1.3138 is close to H /[OII] = 1.3041. A one pixel (42°) uncertainty in A b oth line wavelengths could result in an incorrect line identification. The remaining sources only have one line. The disp ersion is too low to split the [O II] doublet, the [O III]4959,5007° lines are also blended, as is H and the [N II] doublet. With A only one line, at the grism's resolution, then a good first guess redshift is crucial for line identification. Drozdovsky et al. (2005) tackle this problem, in part, by looking at the size of the host galaxies. However, size alone is not a great indicator of redshift - there is little evolution in angular size for z > 0.2. Our approach is to use photometric redshifts as the first guess redshift. This results in line identifications for 37 of the 39 single line ELGs. Figure 3 compares grism redshifts with sp ectroscopic redshifts, in panel a, and sp ectroscopic versus photometric redshifts in panel b. Taking the sp ectroscopic redshifts as "truth" results in 1/15 : 3/19 misidentified lines with methods A:B. This is similar to the error rates from photometric redshifts, as can b e discerned from Fig. 3b. The disp ersion ab out the zgrism versus zspec unity line, excluding the outliers is 0.016:0.009 for methods A:B. Method B is probably more accurate b ecause it b etter pinp oints the location of line emission. This compares to a disp ersion ab out the unity line in zphot versus zspec (after clipping outliers) of 0.073, 0.107, 0.082 for zphot estimates from Capak (2004), FernandezSoto et al. (1999), and our own photometric redshifts resp ectively. Thus grism redshifts are nearly an order of magnitude more accurate than photometric redshifts. 5. Finding Sup ernovae

The two SNe discovered in the HDFN have broad absorption features, distinctly different from Galactic stars, and are easily visible in our grism sp ectra (Blakeslee et al. 2003) demonstrating the viability of grism surveys for SNe searches. The PEARS team has obtained 200 orbits of HST time primarily to characterize high-redshift ob jects in the two GOODS fields using the WFC and G800L grism. An additional aim is to search for SNe on a rapid turn-around basis. The total exp osure time at each p ointing/roll angle is ab out twice as long as the HDFN observations describ ed ab ove. However, only shallow broadband images are obtained concurrently with the grism exp osures. These are used to align the grism images to the astrometric grid of the GOODS fields. But they are not as deep as the grism image, hence they may not reveal SNe. So although aXe sp ectra are generated of the prior GOODS cataloged sources, they are not useful for finding SNe at later ep ochs. What is needed is a method to find SNe using only the grism images. To this aim I have develop ed an IDL package SHUNT (Sup ernovae Hunt) to find and classify the first order sp ectra of all compact sources in a grism field. As b efore, the starting p oint is geometrically corrected, combined grism images. Since most source cataloging codes (i.e. SE ) have b een develop ed to find compact blobs, they do not work so well for finding grism sp ectra which are very extended, often at near the noise level of the image. Rather than optimizing


100

G. R. Meurer

Figure 3: Comparison of grism redshifts (left) and photometric redshifts (right) with sp ectroscopic redshifts from Cowie et al. (2004). In the left panel, the unity relationship is shown as a solid line, sources outside the dashed lines at z = ±0.105 are outliers. Only photometric redshifts were used for the first guess redshift. If sp ectroscopic redshifts are used as priors there is still one outlier. Here, measurements from method A are shown with solid symb ols, measurements from method B are shown as op en symb ols. The symb ol shap e and color indicate the grism line identification: H emitters are (red) circles, [O III] emitters are (green) triangles and [O II] emitters are (blue) squares. In the right panel the photometric redshifts from Cowie et al. (2004), Fernandez-Soto et al. (2004), and our measurements are shown as (green) filled circles, (red) asterisks, and (blue) hollow diamonds resp ectively.

Figure 4: Example of a sup ernova identified with SHUNT . The top left panel shows the geometrically corrected grism image. The b ottom left panel shows the extracted 1D sp ectrum found by collapsing the ab ove 2D image b etween the dashed lines. The top right panel shows a 1D cut along the cross disp ersion of the sp ectrum. The b ottom right panel shows the squashed grism image with the source identified.


Selection and Characterization of Interesting Grism Sp ectra

101

Figure 5: Portion of a grism image containing a faint late typ e SN. Panel a (left) shows the geometrically corrected and combined image. Panel b (right) shows the same image after subtracting the aXe model image based on prior GOODS observations, and then subtracting a smoothed version of the residuals. The SN now clearly stands out. the code to fit the data, SHUNT makes the data suit the code by squashing (rebinning) the image 25 в 1 p er output pixel b efore cataloging with SE . This results in first order sp ectra b eing close to critically sampled in the x direction. The resultant catalog is filtered to remove small sources (typically zeroth order images) and extended sources (galaxies). The remaining 250 sources are then classified. Five rows centered on each source are collapsed to form a sp ectrum (which is not wavelength calibrated). Classification is by eye where the classifier (me) examines figures such as Fig 4 showing the 2D sp ectrum, the collapsed 1D sp ectrum, a cross disp ersion trace and the squashed grism image. Each source is classified as either SN, unidentified absorption sp ectrum, probable M or K star, break sp ectrum, emission line source, featureless, non-first order sp ectrum, or spurious (the order is of decreasing interest, and roughly of increasing occurrence rate). Direct p ostage stamp images from GOODS (or the shallow broad-band images) are generated with a rectangular error b ox plotted which should contain the source. An empty error b ox in the prior GOODS image is a second indication of a transitory source. It typically takes 0.5 to 1 hour to classify all the ob jects in a field. One problem with this approach is that it can miss SNe blended with galaxy sp ectra. This is more likely to occur for late time SNe sp ectra which can have low S/N and/or b e featureless at grism resolution. An example is shown in Fig 5. One way such ob jects can b e found is to subtract model sp ectra of the sources cataloged by GOODS using aXe v1.5 (Kummel, this volume). The models are very good but not p erfect. However, subtraction of Ё the smoothed residuals is sufficient to isolate faint transient ob ject sp ectra from the model residuals.

6.

Summary

The ACS grism produces amazingly rich datasets. While the data are somewhat difficult to interpret, tools have b een develop ed to make the most use of these data. Public access tools like aXe are readily available to remove most of these complications and extract 1D and 2D sp ectra. Here I have shown how some common manipulations of the data (such as geometric correction and flat fielding) allow interesting sources such as emission line galaxies and sup ernovae to b e efficiently found.


102

G. R. Meurer

Acknowledgments. Many memb ers of the ACS and PEARS Science teams as well as others have contributed to the work presented here. I particularly thank Zlatan Tsvetanov, Holland Ford, Caryl Gronwall, John Blakeslee, Peter Capak, Sangeeta Malhotra, Norb ert Pirzkal, Chun Xu, Txitxo Benitez, James Rhoads, Jeremy Walsh, and Martin Kummel. Ё References Blakeslee, J. P., Anderson, K. R., Meurer, G. R., Benitez, N., & Magee, D. 2002, ASP Conf. Ser. 295: ADASS XI I, p. 257 Blakeslee, J. P., et al. 2003, ApJ, 589, 693 Bertin, E., & Arnouts, S. 1996, A&AS, 117, 393 Capak, P. L. 2004, Ph. D. Thesis, U. Hawaii Cowie, L. L., Barger, A. J., Hu, E. M., Capak, P., & Songaila, A. 2004, AJ, 127, 3137 Drozdovsky, I., Yan, L, Chen, H.-W., Stern, D., Kennicutt, R., Spinrad, H., & Dawson, S. 2005, AJ, 130, 1324 Fernґndez-Soto, A., Lanzetta, K. M., & Yahil, A. 1999, ApJ, 513, 34 a Pirzkal, N., Pasquali, A., & Demleitner, M. 2001, ST-ECF Newsletter, 29, "Extracting ACS Slitless Sp ectra with aXe", p. 5 (http://www.stecf.org/instruments/acs)