Astronomy is an observational, rather than an experimental, science, and its practitioners carry out the equivalent of experiments by discovering, and then studying, astrophysical systems with a variety of ages and initial conditions. Surveys generate the list of available laboratories for such studies, and, as such, are central to progress in the discipline. Complete, unbiased surveys are the best technique we have both for discovering new and unexpected phenomena (Harwitt 1981; Kellerman and Sheets 1983), and for deriving the intrinsic properties of source classes so that their underlying physics can be deduced. Recognizing this, optical astronomers began the first modern effort at a complete unbiased survey in 1949. The National Geographic-Palomar Observatory Sky Survey (POSS I) provides data on optical emitters over the entire northern sky down to a flux density limit of mJy and with arcsecond positional accuracy; it has been a fundamental resource ever since its creation, and the impending release of digital images and source catalogs based on this survey will only serve to enhance its importance to the discipline.
The radiation emitted by most astronomical objects is not, however, restricted to narrow ranges in the electromagnetic spectrum defined by atmospheric transparency or instrumental sensitivity limits. Thus, surveys in widely separated wavelength bands are essential in constructing accurate and complete pictures of various source populations. As each new window in the spectrum has opened over the past 30 years, major space missions have had as their first priority unbiased sky surveys: from SAS-2 to the Compton GRO in the gamma ray regime, UHURU to ROSAT in X-rays, and IRAS in the infrared, the source catalogs and sky maps produced by survey missions form a core resource for all of modern astrophysical research. With recent improvements in computing speed, storage capacity, database systems, and networks, the availability and utility of this resource is further enhanced. Projects currently being planned to carry out new ground-based surveys in the optical (the Sloan Digital Sky Survey, Gunn and Knapp 1992) and the near-infrared (the 2-micron Survey, Kleinmann 1992) will have equally major impacts on the field.
Radio astronomers have produced a large number of surveys over the fifty-year history of the discipline although, in the last three decades, their approach has often been to build special purpose telescopes for this task, while using the ``premier" instruments of the day for detailed studies of individual objects and source classes. The series of Cambridge survey arrays, the Mills Cross and its successors, and the Texas 80-cm array are examples of special purpose instruments which have made major contributions to mapping the radio sky; the Texas survey catalog is the largest radio source list in existence with 80,000 entries (Douglas, private communication), although it has never been formally published. In the late 80's, the Green Bank 300-foot telescope, considered by some (prescient) astronomers to be in its dotage, was used by Condon and his collaborators to conduct perhaps the most widely used radio sky surveys at 6 cm and 20 cm (Condon and Broderick 1985, 1986; Condon Broderick and Seielstad 1989). Catalogs derived from the higher frequency survey have been produced by Becker, White, and Edwards (1991) and by Gregory and Condon (1991), and contain sources at 6cm down to a flux density threshold of mJy with positions; White and Becker (1992) produced the analogous catalog at 20 cm which includes 30,329 sources with positions and a limiting flux density of 100 mJy. Finally, a southern complement to the Green Bank surveys has recently been completed with the Parkes telescope (Wright et al. 1994); it has a flux density limit of 30 mJy at 6cm and positions good to . The maps and catalogs from this work represent the state of our knowledge of the radio sky at centimeter wavelengths.
These surveys do not, however, represent the best that is possible with today's radio telescopes. Pencil beam surveys made with the interferometric arrays at Westerbork and the VLA have reached flux densities a factor of lower (Donnelly, Partridge, and Windhorst 1987; Windhorst et al. 1985; Mitchell and Condon 1985), revealing new populations of radio sources below a few mJy (e.g., Benn et al. 1993 and references therein). In addition, the hundred-fold improvement in angular resolution made possible with these instruments has shown that of the sources in single-dish catalogs (S mJy) are resolved on scales of (Windhorst, Mathis, and Neuschaefer 1990). More importantly, since optical counterparts to centimetric sources are typically faint - of sources at 3 mJy have counterparts fainter than (§ 3.1) - optical quality positions are required for determining source distances and studying their evolution. Until a few years ago, computing capacity was simply insufficient to contemplate a major sky survey using these instruments. This is no longer the case. Thus, both Westerbork and the VLA are now engaged in large-scale sky surveys. FIRST is the most sensitive of these.
FIRST represents a factor of improvement in limiting sensitivity over the best available sky survey at any radio wavelength. This is equivalent to the four-magnitude improvement in limiting magnitude that the Sloan Digital Sky Survey will achieve over the POSS I plates. In addition, it is a fully digital survey whose maps and catalogs will be instantly accessible to the astronomical community. More importantly, however, FIRST also represents a factor of 50 improvement in angular resolution and concomitant positional accuracy. As we show below, even the weakest sources will be located to . As a consequence, 15%of all detected objects will be immediately identifiable on the POSS I survey, and over 50%will have optical counterparts at the SDSS limit with a low chance coincidence rate. The final catalog will contain approximately one million radio sources.
On August 8, 1990, we submitted a proposal to the NRAO which began: ``We propose to produce a centimeter-wave counterpart to the Palomar Observatory Sky Survey by using the VLA to map the northern sky." Our stated goals were a flux density limit of mJy and positional accuracies of . The timing of this proposal was a consequence of several factors. We were in the process of completing a multiwavelength survey of the Galactic plane at 6 cm, 20 cm, and 90 cm, and had worked through some of the problems attendant on producing and analyzing thousands of VLA images (Zoonematkermani et al. 1991; Becker et al. 1991; Helfand et al. 1993, and Becker et al. 1994); in addition, our generation of source catalogs from the Green Bank 6 cm and 20 cm surveys of Condon and Broderick (1986) and Condon et al. (1989) provided experience in the automated production of survey source lists. More importantly, the VLA was in the process of installing a new generation of 20 cm receivers which would lead to a sensitivity improvement of a full factor of two for all observations at this frequency. Also critical to the implementation of such a large-scale project was the rapid advance in workstation computing and storage capacity which generated affordable options for reducing the terabyte datasets involved in a high-resolution all-sky survey.
Our original proposal called for a survey covering the whole sky visible from the VLA () using either the C-configuration, or, preferably, a specially designed hybrid configuration between B and C which would provide the resolution and positional accuracy of the B-configuration (maximum spacing 57 kilolambda and a synthesized beam of ), while retaining 6 of the antennas in their more compact, C-configuration positions in order to enhance our sensitivity to low surface brightness objects and sources extended on scales . The plan required a total of 123,750 1.5-minute snapshot observations in bandwidth synthesis mode to achieve a typical map rms of 0.20 mJy at field center; the projected source catalog would contain in excess of one million radio sources with positions accurate to . We proposed carrying out the survey by including one extra month that would be exclusively devoted to the survey in each of the six VLA configuration cycles following completion of the 20cm receiver upgrade; this would have allowed for completion of the survey by the end of the decade with minimal disruption to the rest of the VLA science program.
Following consultation with senior NRAO staff members, the Director informed us of his decision not to have the proposal refereed. During the ensuing eight months, we engaged in extensive discussions with the NRAO and with members of the community, and redesigned our proposal to meet some of the objections raised following the original submission. We adopted the standard C-configuration as the baseline array (with CnB for southern declination fields). As in our original proposal, we committed to releasing the final maps and catalogs within two years of each observing run (to allow for reobservation of problem fields); to demonstrate our ability to reduce and distribute the data in a timely manner, we proposed a ten-day pilot survey for the 1992 C-configuration cycle which we promised to deliver to the community within six months. In our cover letter for the resubmission of our proposal on 9 May 1991, we pointed out that during the period 1988-1990, of all allocated VLA time was used for 20 cm snapshot observations; in that our survey would obviate the need for the vast majority of these observations, it would actually increase the amount of telescope time available for other science projects in the next decade.
Two weeks prior to this resubmission, the NRAO received another proposal from an internal team led by Dr. J. Condon to survey the sky north of using the VLA at 20 cm. This proposal suggested a larger number of shorter pointings (190,000 30-second snapshots vs. 120,000 90-second snapshots). The most important difference, however, was that it proposed to use the D-configuration ( synthesized beam). The advantages cited for the lower-resolution survey were increased surface brightness sensitivity and simplification of the computational requirements, since the use of bandwidth synthesis mode would not be required to achieve the requisite wide-field mapping, and the maps would contain nine times fewer pixels owing to the larger beam size. The costs of going to the D configuration were an increased survey flux density threshold ( mJy vs. mJy[NOTE: The NVSS is actually achieving a flux density limit of 2.5 mJy. ]), lower accuracy positions, and decreased resolution for exploring source structure. The wider usable bandwidth in this array decreased the total survey time required by a factor of two (albeit, at the loss of a factor of 1.5 in threshold sensitivity). In addition, all observations need to be conducted at night in order to minimize man-made interference to which the D configuration is more susceptible. Furthermore, using a small number of broad frequency channels places strong limitations on one's ability to excise narrow-band interference.
Both proposals were refereed in the summer of 1991. There was widespread support among the referees for a VLA sky survey, but no consensus as to which survey should be done. During the 1992 C and D configurations, a 25 region was surveyed twice with the respective survey parameters in order to provide some empirical data with which to make a judgement concerning the relative merits of the two approaches. In November 1992, a special panel including NRAO and non-NRAO scientists was convened to hear presentations from both survey teams. Their recommendation, accepted by the Director, was that two surveys be done. The D-configuration survey, now known as NVSS, was extended to cover the entire sky north of and was scheduled to be completed over the next three configuration cycles. In addition, convinced of the desirability of accurate source positions for optical followup, the virtue of a deeper survey unaffected by source confusion, and the importance of high angular resolution for morphological studies, the committee recommended that our team conduct a survey using the B-configuration (synthesized beam ) that would cover the same 10,000 of the north Galactic cap scheduled for observation with the Sloan Digital Sky Survey. Both teams accepted this charge. The high-resolution survey team chose FIRST as its moniker.
An enlarged and reconstituted Survey Oversight Committee met in March of 1993 to review this decision and to begin their task of monitoring its detailed implementation. A common pointing grid designed by one of us (RLW) was adopted for both surveys and is described in detail below (§ 4.1). The integration time per field was set at 3 minutes for the B-configuration survey in order to achieve a submilliJansky point source threshold. The first FIRST observations were scheduled in April 1993, with a total of 144 hours completed by the end of the B-configuration in May. Apart from a few isolated observations designed to overlap existing deep survey fields and to test the daytime interference environment, coverage included a contiguous strip degrees wide near the local zenith (see § 5). Recently completed observations during the 1994 season extend this strip by 10 degrees in declination, and bring the total area covered to .
As we had originally proposed, the standard VLA proprietary period of one year is waived for all FIRST data; i.e., the raw UV data are available from the VLA archive the day they are taken. In addition, we have agreed to provide fully self-calibrated UV datasets for each observation to the NRAO for distribution to any interested users; all data taken in 1993 and 1994 are now available. More importantly, we are creating a public archive of all survey images and catalogs as soon as they are fully tested for quality and accuracy. This includes the individual CLEANed maps from each pointing as well as the coadded images (see below) which form the principal survey product. In addition, an annotated source catalog and individual ``postage stamp" images of each source are being made available. Information on retrieving any of these data products can be obtained from the FIRST home page on World Wide Web (http://sundog.stsci.edu/). Images from the 1993 observations are now available.