next up previous contents index
Next: Intergalactic Helium Absorption Up: QSOsGalaxies, and Previous: DEEP Keck Survey

Science with the Hubble Space Telescope -- II
Book Editors: P. Benvenuti, F. D. Macchetto, and E. J. Schreier
Electronic Editor: H. Payne

Automated Morphological Classification of Galaxies from Ultradeep WFPC2 Fields in BVI

Stephen C. Odewahn, Rogier A. Windhorst
Arizona State University, Dept. of Physics & Astronomy, Tempe, AZ 85287

Simon P. Driver
University of New South Wales, School of Physics, Sydney, NSW 2052, Australia

William C. Keel
University of Alabama, Dept. of Physics & Astronomy, Tuscaloosa, AL 35487



We analyzed deep HST/WFPC2 images in using bandpass-specific artificial neural network (ANN) classifiers, which are based on galaxy surface brightness (SB) and light profile (but not on color nor directly on scale-length). The ANN distinguishes quite well between E/S0's, Sabc's, and Sd/Irr's for B 26.5 mag, with larger classification errors at fainter magnitudes. Galaxy types are fairly well-separated in BVI color-magnitude diagrams down to B26.5, with E/S0's being the reddest and Sd/Irr's generally blue down to the formal detection limit for compact objects (B27.5 mag). We model the B band number counts as a function of morphological type with the E/S0 and Sabc samples following no/mild-evolution predictions down to B26.5 mag (z1). The Sd/Irr B band counts are modeled by a moderately steep local LF undergoing luminosity evolution since z1.


The way in which galaxies form and evolve is a fundamental question which has yet to be resolved. Of particular interest is the long standing problem of the faint blue galaxy (FBG) excess. The majority of explanations for the FBG excess observed in deep ground-based CCD images (Tyson 1988) involve Irregular/dwarf populations, or objects with fainter than -17.0 mag (H80, q0.5, see Driver et al. 1995; D95). The superb resolution provided by the refurbished HST Wide Field Planetary Camera 2 (WFPC2) allows the determination of the morphological types of individual galaxies over a wide range of epochs and hence provides important information for studying the nature the FBG population.

Observations and Reduction

In Cycle 4--5, we imaged a single deep dithered WFPC2 field surrounding the weak radio galaxy 53W002 at z=2.39 (24 orbits in B, 12 in V and I).

Details of these observations are given by D95 and Odewahn et al. (1996; O96). As a consistency check, we compared our counts in this 53W002 (hereafter ``W02'') field to those from 3 shallow WFPC2 fields obtained from the HST archive, and those from the Hubble Deep Field (HDF, Williams et al. 1996) which reach 0.8--1.2 mag deeper than the W02 field in BVI.

The W02 and shallower WFPC2 images were processed following D95. The same analysis was made for the HDF images which are closer to the confusion limit, but we use the HDF data only down to the detection limit of our W02 sample (B27.5 mag), where the WFPC2 images are definitely not confusion limited. An automated software package (MORPHO) was used to determine type-dependent photometric parameters (e.g., Odewahn & Aldering 1995) and assign morphological types using an Artificial Neural Network (e.g., Odewahn et al. 1992). As network training samples we used 173 WFPC2 galaxies down to I22 mag from DWG and 227 galaxies down to I24.5 mag from D95. Comparison of the galaxy types from our V & I band classifiers (for B26) produced a scale difference of only 5%, and a scatter of 1.7 steps (rms) on the 16 step revised Hubble system.


The total B magnitude versus color is shown for each type in Fig. 1, where ``wide'' = was used to increase the S/N in colors. While the 50% completeness in the W02 catalogs occurs at B27.5 mag, for B 26.0 mag Fig. 1 becomes increasingly incomplete for the bluest W02 objects because of the red detection limit at 26.5 mag, as indicated by the slanted dashed lines (the HDF limits are again 1.5 mag fainter). Of significance is the clear segregation between the early and the late type galaxies in Fig. 1 down to B27 mag, even though no color information was used by the ANN classifier. E/S0's are almost without exception the reddest galaxies at any flux level, and Sd/Irr's are generally blue, at least down to the formal detection and classification limit (B 27.0).

In Figure 2 we plot the differential B number counts for the different morphological subsets in our five WFPC2 fields. Within the errors, Figs. 2a--2d, show that the B counts in the W02 field are consistent with the other 4 WFPC2 B fields, as well as with ground-based counts (Metcalfe et al. 1995) down to B27 mag, so that field-to-field variations are no larger than the formal errors. Following the I-band models of D95, we modeled the B number counts for the three main morphological types separately. Fig. 2b shows that the B counts observed for E/S0's follow the predictions for passively or mildly evolving models (solid curves for a non-evolving local Marzke LF and dotted curves for a non-evolving local Loveday LF). Fig. 2c shows that Sabc's are also consistent with these models, or at best 0.5 mag brighter for 22B25 mag. Hence, a scenario invoking strong luminosity evolution is not required for the early-type galaxies out to B26 mag (I24 mag or z1, see D95), suggesting that their formation was largely complete by z1. However, Fig. 2d shows that non-evolving models are clearly inadequate to model our B counts for the Sd/Irr population. Consistent with the I-band results of D95, our Sd/Irr B-counts (Fig. 2d) can be described by a population with a moderately steep local LF that underwent some luminosity evolution since z1 (dashed curves). Following the models of D95, a starburst with 1.5 mag at z0.5--1 would be required to explain the steep Sd/Irr counts.

Figure: The vs. B ( B) color-magnitude diagram for all classified galaxies in the two deep W02 and HDF WFPC2 fields. Symbols indicate membership in the categories E/S0, Sabc, & Sd/Irr, as classified by the V & I ANN's of O96. Note that a clear type segregation is present down to the classification limit, although no color information was used by the ANN classifiers.

Figure: Differential B number counts from the W02 field (filled symbols), the HDF (open triangles), and three shallower WFPC2 B fields (open squares), as well as previous ground-based B-band counts (crosses). Panels show (a) all galaxies, (b) HST ellipticals (E/S0's), and (c) early-type spirals (Sabc's), as classified by the ANN. Models are described in the text, and by D95 and O96.


Driver, S. P. et al. 1995, ApJ, 449, L23 (D95)

Driver, S. P. et al. 1995, ApJ, 453, 48 (DWG)

Metcalfe, N., Shanks, T., Fong, R., & Roche, N. 1995, MNRAS, 273, 257

Odewahn, S. C. et al. 1992, AJ, 103, 318

Odewahn, S. C. & Aldering, G. 1995, AJ, 110, 2009

Odewahn, S. C., Windhorst, R. A., Driver, S. P., & Keel, W. C. 1996, ApJ, submitted (O96)

Tyson, J. A. 1988, AJ, 96, 1

Williams, R.E. et al. 1996, this volume, p. gif

next up previous contents index
Next: Intergalactic Helium Absorption Up: QSOsGalaxies, and Previous: DEEP Keck Survey