The observation with the Hubble Space Telescope (HST) of the Hubble Deep Field (HDF) (Williams et al. 1996) has yielded numerous new results in observational cosmology. This is due to its depth and spatial resolution in the near-UV and optical, allowing the computation of photometric redshifts of the observed galaxies as well as the determination of their morphological type. With the ground-based follow-up observations, in the near IR (Hogg et al. 1997a, Connolly et al 1997) for example, or at radio wavelength (Fomalont et al. 1997), as well as spectroscopic measurements (Steidel et al. 1996, Cohen et al. 1996), this field now constitutes a very important database for the study of galaxy formation and evolution. Among the striking results that have been already produced (see Ellis (1997) for a review), two have strong implications on the galaxy formation picture. First, the star formation rate (SFR) at high redshifts has been derived (Madau et al. 1996), and together with that computed from Canada France Redshifts Survey (CFRS) data at lower redshifts (Lilly et al. 1996), it seems to indicate that the bulk of star formation arose below z 2. Second, an increase of faint irregular galaxy counts around has been reported (Abraham et al. 1996), as well as the lack of detection of quiescent evolving early types galaxies at high redshifts. These results tend to favor the hierarchical galaxy formation picture.
However, the interpretation of these results is now being widely discussed. For instance, it has already been pointed out that the fainter objects seen in the HDF could merely be star forming regions of galaxies and not entire galaxies, because the distance between detected objects is small with respect to the size of a galaxy (Colley et al. 1996). This idea is supported by new simulations that show that diffuse emission from evolved stellar populations is below the detection limit of the HDF, while star forming regions with high UV surface brightness can be detected (Hibbard & Vacca 1997). For similar reasons, quiescent evolving early-type galaxies at redshift of are also undetectable (Maoz 1996).
Most of the data that were used to derive star formation rates are UV and optical data from the CFRS and the HST itself, which are sensitive to extinction. Indeed, it has been shown that star formation at high redshift is obscured, typically by 2 to 3 mag at Å (Meurer et al. 1997). The SFR inferred might therefore change if the reprocessing of UV by dust into IR is taken into account. Knowledge of the dust content of galaxies is therefore relevant to assess the results obtained so far, both at with the CFRS and at higher redshifts. Additional information on this problem can be obtained by observing the HDF field in the infrared.
Many models have been built to compute the infrared spectra of various types of galaxies (AGN, starbursts, early types, etc ) and make predictions on the number counts for surveys in the infrared (see, for example, Franceschini et al. (1991). Moreover, various groups have derived new IRAS 60 µm counts, going to the extreme limits of the sensitivity of this satellite. Their results show an excess of sources with respect to the predictions of a scenario with only passive luminosity evolution of galaxies (Bertin et al. 1997, Gregorich et al. 1995). However, the exact amount of evolution needed is still a matter of debate, because the number counts derived by the various teams are slightly different. This issue needs to be revisited on deeper samples. From this point of view, the HDF field is a great opportunity to test these models, first because their number count predictions can be checked against observations, and second because all the information at other wavelengths is of great help to test the hypotheses of the various models.
Since the HDF field is rather small ( 5 arcmin2) and very deep, infrared observations require a good spatial resolution as well as a good sensitivity. This can be achieved by the mid-IR imager ISOCAM of the Infrared Space Observatory (ISO) (Kessler et al. 1996), which has proven to be 1000 times more sensitive than IRAS, with a 50 times finer spatial resolution (Cesarsky et al. 1996). Part of the Director's Time was dedicated to complete these observations, at m and m, for a total amount of 12.5 hours, with M. Rowan-Robinson as Principal Investigator (Serjeant et al. 1997, Goldschmidt et al. 1997, Oliver et al. 1997, Mann et al. 1997, Rowan-Robinson et al. 1997. The data of the ISO-HDF are now in the public domain, and we present here the results of our processing.
The ISOCAM observation of the HDF, hereafter ISO-HDF, is at the moment the deepest survey obtained in the 15 µm band. It is also deep in the 6.75 µm band, although a deeper integration on the Lockman Hole was obtained by Taniguchi et al. (1997) with this filter. We have developed a new technique for ISOCAM data reduction, based on wavelet analysis (Starck et al. 1998): the Pattern REcognition Technique for ISOCAM data (PRETI). Compared to data that we have at hand from less deep surveys in the Lockman Hole, from the Guaranteed Time program by Cesarsky et al. (1996), and in the shallow survey of the ELAIS program, this set has a much higher redundancy, and thus is useful for finding the ultimate capacities of our data reduction method. In addition, since the ISO-HDF has been analyzed by other teams such as Rowan-Robinson et al. (1997) and Désert et al. (1998), this observation allows us to compare the results from our method to those of more classical ones. All these methods aim at extracting very faint sources in ISOCAM data, and independent reduction techniques improve the quality of the produced catalogs. In Sect. 2, the characteristics of the emission expected from galaxies in the MIR are outlined, and compared to the bandpasses of the ISOCAM filters that were used for the HDF observation. In Sect. 3 we present the data acquisition and stress the critical steps of the reduction process. The way in which these difficulties are handled by the PRETI method is described in Sect. 4. The efficiency of the method for source detection and its photometric accuracy was also tested on simulated data, which are presented in a Sect. 5. Finally, we give our source catalog and number counts, compare them with those of other groups, and discuss the results. A more detailed interpretation of the results, source by source, will be presented in a forthcoming paper (Elbaz et al., in prep. ).
© European Southern Observatory (ESO) 1999
Online publication: February 22, 1999