pipeline Note a postscript file of this document can be found here.
Grism data will account for a significant portion of the ACS science team GTO data (50 WFC orbits). While ST-ECF will provide software to efficiently extract 1-d spectra, we will need to construct a scientific analysis pipeline for the grism data. The goal of the grism science pipeline is to efficiently process the extracted spectra (from ECF software) in order to produce scientifically useful parameters while maintaining an automated approach where feasible. We forsee the following steps (note the first two may be included in the final ECF software but are not part of the current ECF requirements).
We assume the ECF software will identify objects with overlapping spectra. We will want to deblend overlapping spectrum so as to utilize all the information in the grism image. (In our cluster fields, many of the spectra will be overlapping. Simulations are required to estimate the magnitude of this problem.) This can be done automatically utilizing knowledge of the relative fluxes (from direct images) and spectral shape of the overlapping objects (taken from a set of spectral templates). Such deblending has been implemented by ST-ECF for the extraction of NICMOS grism data and the algorithms used should be translatable to ACS (see here for more information.)
The identification of strong emission line sources after 1-d spectra have been extracted is straightforward. One possible procedure (following that used for the KPNO International Spectroscopic Survey, Salzer et al. 2000) is the following. The extracted spectrum should be smoothed using a median filter algorithm. This filters out any strong emission lines. The smoothed spectrum is then fitted by a low-order polynomial. The polynomial fit is then subtracted from the original, unsmoothed object spectrum to create a continuum-subtracted spectrum. This spectrum is then divided by a noise spectrum which is calculated pixel by pixel given the flux level in the spectrum, assuming Poisson statistics apply, and including terms for sky background and read-out noise. Thus, the final output spectrum is in units of the local noise level (a "sigma-spectrum") which can then be searched for pixels above a certain user-specified threshold (e.g., a 5-sigma detection).
Another characteristic spectral feature of galaxies is a strong spectral break such as 4000 Å break in early-type galaxies, and the Lyman continuum break exhibited by high redshift (z > 3) galaxies in the optical. Probably the best way to accomplish this task is via cross-correlation with a set of template galaxy spectra. See the description above for further information.
The best way to automatically measure spectroscopic redshifts is through cross-correlation analysis (e.g., Glazebrook et al. 1998) with a set of template spectra. The cross-correlation function is equivalent to a redshift likelihood which is quite similar to the Bayesian approach of determining photometric redshifts (Benitez 2000a, BPZ). Benitez (2000b) has proposed an extension of BPZ, called "Bayesian Automatic Redshifts" (BAZ) which will use the photometric information to create prior information on the most probably galaxy types and colors which can be used to automatically measure redshifts. Additional output of this program will include an estimate of the galaxy type (or best-fitting template spectra) which provides information on the presence of strong spectral breaks and the stellar population of the galaxy. We note that Glazebrook et al. used a set of eigen-templates (e.g., Connolly et al. 1995) which can be linearly combined to form a best-fitting match to the observed spectra. We will probably want to adopt this technique as well.
Once strong emission-line sources have been identified, we will want to measure the line flux and equivalent width of these objects. This should be a straightforward task involving the fitting of a Gaussian profile to the emission line in question.
Another interesting piece of information we can derive from the grism spectra is the best-fitting galaxy spectral type. This should be a by-product of the automated redshift scheme described in above.
Benitez, N. 2000a, ApJ, 536, 571 (BPZ)
Benitez, N. 2000b, private communication (BAZ)
Connolly, A.J., Szalay, A.S., Bershady, M.A., Kinney, A.L., & Calzetti, D. 1995, AJ, 110, 1071
Glazebrook, K., Offer, A.R., & Deeley, K. 1998, ApJ, 492, 98
Salzer, J.J., Gronwall, C., et al. 2000, AJ, 120, 80
