We have used the Very Large Array (VLA) in C configuration to carry out a sensitive 20-cm radio survey of regions of sky that have been surveyed in the far-infrared (FIR) over the wavelength range 5-200 μm with ISO (Infrared Space Observatory) as part of the European Large-Area ISO Survey (ELAIS). As usual in surveys based on a relatively small number of overlapping VLA pointings, the flux limit varies over the area surveyed: from a 5σ limit of 0.135 mJy over an area of 0.12 deg2 to 1.15 mJy or better over the whole region covered of 4.22 deg2. In this paper we present the complete radio catalogue of 867 sources, 428 of which form a complete sample in the flux range 0.2-1.0 mJy. These regions of sky have previously been surveyed to shallower flux limits at 20 cm with the VLA as part of the VLA D configuration NVSS (full width at half-maximum, FWHM=45 arcsec) and VLA B configuration FIRST (FWHM=5 arcsec) surveys. Our whole survey has a nominal 5σ flux limit a factor of 2 below that of the NVSS; 3.4 deg2 of the survey reaches the nominal flux limit of the FIRST survey and 1.5 deg2 reaches 0.25 mJy, a factor of 4 below the nominal FIRST survey limit. In addition, our survey is at a resolution intermediate between the two surveys and thus is well suited for a comparison of the reliability and resolution-dependent surface brightness effects that affect interferometric radio surveys. We have carried out a detailed comparison of our own survey and these two independent surveys in order to assess the reliability and completeness of each. Considering the whole sample, we found that to the 5σ nominal limits of 2.3 and 1.0 mJy, respectively, the NVSS and FIRST surveys have a completeness of 96+2-3 and 89+2-3 per cent and a reliability of 99+1-2 and 94+2-2 per cent.
The Infrared Space Observatory (ISO) (Kessler et al. 1996), launched in 1995 November, was the successor to the Infrared Astronomical Satellite (IRAS) and provided unparalleled sensitivity at mid- to far-infrared wavelengths (i.e. 5-200 m). The European Large-Area ISO Survey (ELAIS) (Oliver et al. 1997; Oliver et al., in preparation) is a project that used ISO to carry out a deep wide-angle survey at wavelengths of 6.7, 15, 90 and 175 m. The 6.7- and 15-m surveys employed the ISO-CAM camera (Cesarsky et al. 1996) with the aim to reach a 5σ sensitivity of ∼2 mJy at 15 m. The 90- and 175-m surveys used the ISO-PHOT camera (Lemke et al. 1994) with the aim to reach a 5σ sensitivity of ∼25 mJy. At these limits, we expect ISO to be confusion-limited at 90 and 175 m by galaxies and Galactic cirrus emission, and hence this survey should be the deepest FIR survey possible with the satellite.
The area covered in the ELAIS survey is ∼13 deg2 at 15 and 90 m, ∼7 deg2 at 6.7 m and ∼3 deg2 at 175 m.
The ELAIS survey is ∼50 times deeper at 5-20 m than IRAS. Thus our survey will allow us to explore IRAS-like populations to higher redshift and possibly unveil new classes of objects or unexpected phenomena. We expect to detect thousands of galaxies, many of which will be at high redshifts and undergoing vigorous star formation. The expected large number of high-z IR galaxies should provide vital information about the star formation rate out to z=1 and possibly earlier.
The spatial resolution of ISO will be insufficient to identify optically faint objects properly. At 15 m, the survey resolution (full width at half-maximum, FWHM) is ∼10 arcsec and at 90 m it is about 1 arcmin. Complementary radio data will play a crucial role in identifying many of the most interesting objects, as they did in the early days of X-ray astronomy (e.g. Cyg X-1) and in more recent times for IRAS (e.g. IRAS F10214+4714; Rowan-Robinson et al. 1991).
In this paper we report the description of the radio observations obtained in the three ISO ELAIS survey regions in the Northern celestial hemisphere (N11610+5430, N21636+4115 and N31429+3306). The observations are made with the Very Large Array (VLA) radio telescope at 1.4 GHz (20 cm) in the VLA C configuration (maximum baseline 3.4 km) with a resolution (FWHM) of ∼15 arcsec. The aim of these VLA observations was to obtain uniform coverage of the ELAIS regions with an rms noise limit of ∼50 Jy. These VLA observations will be essential in the optical identification phase of the ELAIS sources and in assessing the reliability of the ELAIS source lists.
Moreover, with a radio survey it will be possible to investigate the radio/far-infrared correlation in star-forming galaxies to flux densities considerably fainter than those reached by IRAS. Helou, Soifer & Rowan-Robinson (1985) noted a strong correlation between radio and far-infrared flux for star-forming galaxies, valid over a very wide range of infrared luminosities, a result that has been confirmed in many subsequent studies (e.g. Wunderlich, Klein & Wielebinski 1987; Condon, Anderson & Helou 1991). The radio emission is interpreted as a synchrotron radiation from relativistic electrons that have leaked out of supernova remnants. It is expected that this correlation should extend below the IRAS flux level since the majority of sub-mJy radio sources have been identified with faint blue galaxies with spectra similar to those of star-forming objects (Benn et al. 1993).
In addition, combining deep radio and optical data with the ISO survey fluxes will provide information on the trivariate IR-radio-optical luminosity function and its evolution, and the contribution of starburst galaxies to the sub-mJy radio source counts. The ratio of the FIR emission and radio emission will also allow us to investigate the physical origin and spatial distribution of the energy sources in the detected objects in the same way that VLA maps have been central to our understanding of the origin of IRAS sources.
Finally, owing to its depth and angular extension, this survey is important also as a radio survey in its own right. In fact, the selected sample is large and deep enough to constitute a statistically significant sample of sub-mJy radio sources, whose nature and characteristics are still a major topic in observational cosmology (see Windhorst, Mathis & Neuschaefer 1990;,Fomalont et al. 1991;,Rowan-Robinson et al. 1993; Gruppioni et al. 1997).
The ISO ELAIS survey is available at http://athena.ph.ic.ac.uk.
2 Radio Observations
2.1 Choice of observing frequency and VLA configuration
The VLA C configuration and the observing frequency of 1.4 GHz give the optimum resolution to acquire the kind of radio data that we need. The VLA D configuration is confusion-limited at the fluxes we wish to attain (the 5σ confusion limit in D configuration is 0.4 mJy/beam). With the C configuration and a frequency of 1.4 GHz the synthesized beam size (full width at half-power, FWHP) is ∼15 arcsec, which should enable us to pinpoint optical identifications to ∼1-2 arcsec, except for asymmetric multicomponent sources. The frequency of 1.4 GHz was chosen because at this frequency the FWHP of the VLA primary beam is 31 arcmin. This allows us to cover the ELAIS survey fields with a relatively small number of pointing centres. It is possible to obtain a mosaic map with nearly uniform sensitivity if the separation is 31/√2 ∼ 22 arcmin. Moreover, at 1.4 GHz there will be contributions from both the steep and flat spectrum populations of radio sources.
Our observations are made in spectral line mode using two different IF (intermediate frequency) channels centred at 1.3649 GHz (IF1) and 1.4352 GHz (IF2). Each IF channel has a bandwidth of 18.75 MHz, which is subdivided into seven spectral line channels evenly spaced in frequency across the bandwidth of the input IF channel. Therefore, we have a total of 14 spectral line channels with a bandwidth of 2.68 MHz each. We decided to use the line mode in order to facilitate wide-field mapping and avoid the effect of interference. The total bandwidth available is thus 37.5 MHz, narrower than the 2×50 MHz used in continuum mode. A narrower bandwidth means a worse sensitivity. In our case we lose about 30 per cent in signal-to-noise ratio near the pointing centres because the total bandwidth is ∼40 per cent of that in the continuum mode. However, we reduce the chromatic aberration (bandwidth smearing) considerably, which reduces the area covered by each pointing in continuum mode. In addition, the line mode is less susceptible to narrow-band interference, since one only needs to excise the channel that is affected rather than losing the whole IF band, which is the case in continuum mode. Since our aim is to obtain a uniform sensitivity over the whole ELAIS fields and not to obtain a single-field deep survey, we have opted to use the line mode. Moreover, the reduced bandwidth smearing in line mode will give us more accurate angular sizes for our sources.
The VLA observation of the ELAIS regions were carried out in 1996 April (10 h) and in 1997 July (24 h). A total of 20 pointings were made in 20-cm spectral line mode. Seven of these pointings are in the ELAIS field N1, eight in the field N2 and five in N3. The integration time of each pointing was ∼1 h. This allowed us to obtain, at each field centre, an rms noise of ∼0.05 mJy (a 5σ limit of 0.25 mJy). Moreover, during the 1997 July run, we observed two regions (one in N1 and one in N2) with an integration time of ∼3 h each. In particular, in N1 we re-observed (for ∼3 h) the pointing V1 (already observed for 1 h in 1996 April) while in N2 the deep pointing (N2 VD) was shifted 10 arcmin from the map centre to avoid the presence of a strong (>100 mJy) radio source. In the two deep pointings N1 V1 and N2 VD we reached an rms noise of ∼0.026 mJy (a 5σ limit of 0.13 mJy). Finally, in order to study the degradation of the image quality as a function of the off-axis angle, a strong calibrator (the source 3C 286) was observed at position offsets of 5, 10, 15, 20, 25, 30 and 35 arcmin in two orthogonal directions (north-south and east-west). The result of this test is reported in Section 5.1.2.
In Table 1, we report the position of the centre of each observation, while Fig. 1 shows the sky positions and orientations of the ISO survey regions. The circles show the VLA regions mapped. The circles are drawn with a radius (R∼20 arcmin) where the VLA power sensitivity is ∼30 per cent of the central value.
3 Data Reduction
All the data were analysed with the NRAO AIPS reduction package. The data were calibrated using 3C 286 as primary flux density calibrator [assuming a flux density of 15.04 Jy at 1.3649 GHz (IF1) and 14.70 Jy at 1.4352 GHz (IF2)], and the sources 1549+506 and 1635+381 as secondary calibrators. The task UVFLAG was used to ‘flag’ (delete) the corrupted data (e.g. bad integration, non-operating antennas, high amplitudes caused by interference, etc.) while the tasks VLACALIB and GETJY were used to calibrate the data and to determine the source flux densities. Finally each observation was cleaned using the task IMAGR. The cleaning procedure was stopped when the rms map noise was similar to the expected value (see below).
3.1 Rms map noise of the single pointings
The integration times of each observation after deletion of the corrupted data are reported in Table 1. The rms noise of each pointing was estimated using the amplitude distribution of the pixel values in the cleaned map before correcting for primary-beam attenuation. In Fig. 2 we show this distribution for the pointing N2 V1. This distribution is the sum of a Gaussian noise core with a positive-going tail caused by discrete sources. Note that the y-axis in Fig. 2 is logarithmic.
Thus, the rms of the noise distribution alone should be nearly equal to the rms of that distribution obtainable by reflecting the negative flux portion of the observed amplitude distribution about flux=0. In column 7 of Table 1 we report the theoretical rms noise level (at 1σ) as computed directly from the observing parameters (integration time, number of antennas, observing frequency, bandwidth and number of IF channels), while in column 8 we report the 1σ rms noise level obtained by fitting the noise distribution of each pointing. As shown in Table 1 there is a very good agreement between the observed and the theoretical rms noise levels.
3.2 Mosaic maps
Using the AIPS tasks LGEOM, HGEOM and LTESS, we have combined all the observations, creating a mosaic map for each field (N1, N2 and N3). A special procedure has been adopted for N2. The presence of a deep observation not in the centre of the radio map (see above and Fig. 1) makes the noise of this map highly non-uniform. To simplify source extraction, we have created two different mosaic maps. In the first map we have combined all the observations in N2 excluding the deep pointing, while in the second one we have combined the deep pointing N2 VD with all the surrounding pointings (N2 V1, N2 V2, N2 V6 and N2 V7, see Fig. 1). In this way we have obtained two mosaic maps with smoothly varying noise: lower in the map centre and higher in the outer regions.
An example of our images is given in Fig. 3 where we show the contour map of the central region (52×52 arcmin2) of the mosaic map N1.
3.3 Noise of the mosaic maps
We analysed the noise properties in each mosaic map (N1, N2, N2 Deep and N3). As expected, N1, N2 and N2 Deep have a regular noise distribution: a circular central region with a flat noise distribution surrounded by concentric annular regions, where the noise increases for increasing distance from the centre. No structures or irregularities were found in the rms noise maps. In Fig. 4 we plot the variation of the rms noise as a function of the off-axis angle for N1, N2 and N2 Deep. Owing to the different number of pointings in N3 (five pointings instead of seven, see Fig. 1) the mosaic map in this field has a different shape and the noise distribution in the map has a semicircular shape instead of a circular one. In Figs 4(d)-(f) we plot the variation of the rms value in three different slices. In Fig. 4(d) we plot the rms value in the north-south direction, in Fig. 4(e) the rms value in a slice that connects the centre of the three northern pointings (N3 V1, N3 V2 and N3 V4), and in Fig. 4(f) the rms value in a slice that connects the centre of the two southern pointings (N3 V3 and N3 V5). As for N1 and N2, no structures or irregularities were found in the rms noise map of N3.
4 Source Extraction
4.1 Regions for the source extraction
Using Fig. 4 we determined the regions with constant rms noise to be used for the source extraction. Because of the circular shape of the rms map, we searched for radio sources in a circular central region plus concentric annular regions to an off-axis value of 42 arcmin. Outside this area, the beam attenuation significantly increases the limiting flux (see Fig. 4). In each region the rms noise used for source extraction is the highest rms in the region. In this way we are sure that all the sources extracted are above the fixed threshold in the region (for example five times the rms value). The sizes of these regions are reported in Table 2 and in Fig. 5.
Using the data reported in Table 2 we obtained the solid angle Ω(SP) as a function of the peak flux density SP covered by our survey. This effective survey area is shown in Fig. 6.
4.2 Source detections
Within each region we searched for radio sources up to a peak flux density ≥ 5 times the rms value of the region. The sources were extracted using the task SAD (Search And Destroy), which attempts to find all the sources whose peaks are brighter than a given level. For each selected source the flux, the position and the size are estimated using a least-squares Gaussian fit. However, for faint sources a Gaussian fit may be unreliable [see Condon (1997) for an extensive discussion about errors in Gaussian fits]. For this reason we decided first to run the task SAD with a flux limit of three times the local rms value (see Table 2). Subsequently, we derived the peak flux of faint sources (those detected with 3σ<SP<7σ) using a second-degree interpolation (task MAXFIT). Only the sources with a MAXFIT peak flux density ≥5σ were included in the sample. For these faint sources the total flux density was obtained using the task IMEAN, which integrates the map value in a specific rectangle, while for all the other parameters (major axis, minor axis and position angle) we used the values obtained with the Gaussian fit. For irregular resolved sources the total flux density was calculated using the task TVSTAT which allows use of an irregular area to integrate the map value.
4.3 Multiple sources
Detected sources separated by less than twice the value of our synthesized beam size (i.e., 30 arcsec) and with a flux ratio lower than 2 have been considered as a single source. We adopted this criterion because the component flux density ratio of real doubles is usually small (2) while projection pairs can have arbitrarily large flux density ratios (Condon, Condon and Hazard 1982).
5 The Source Catalogue
Considering all the available observations we detected a total of 867 sources at ≥5σ level (44 of which have multiple components as defined in Section 4.3) over a total area of 4.222 deg2. The catalogue with all the 867 sources (921 components) reports the name of the source, the peak flux density SP in mJy, the total integrated flux density SI in mJy, the RA and Dec. (J2000), the full width at half-maximum (FWHM) of the non-deconvolved major and minor axes θM and θm (in arcsec), the position angle PA of the major axis (in degrees measured east to north) and the off-axis angle in the VLA map (in arcmin). The different components of multiple sources are labelled ‘A’, ‘B’, etc., followed by a line labelled ‘T’ in which flux and position for the total sources are given. For these total sources the positions have been computed as the flux-weighted average position for all the components. Table 3 shows the first page of the catalogue as an example. The full catalogue is available from http://www.ast.cam.ac.uk/elais/.
In Table 4 we report the number of radio sources detected in N1, N2 and N3, while in Fig. 7 we show the distribution of the peak flux density and the total to peak flux ratio as a function of peak flux for all the 867 sources. Contour maps of all the 44 double or multiple sources are shown in Figs 8, 9 and 10.
5.1 Source parameters and their uncertainties
5.1.1 Positions and angular sizes
Following Condon (1997), the errors on the source parameters reported in the catalogue have been calculated using
where ρ is the signal-to-noise ratio given by
Equations (1) and (2) are the master equations for estimating the variances of the parameters derived from a two-dimensional Gaussian fit on an image with a noise variance σ2map and pixel size h. In our maps h=3 arcsec, while column 5 of Table 2 reports the value of σmap as a function of the off-axis angle. The rms position errors are given by (Condon et al. 1998):
The mean image offsets 〈Δα〉 and 〈Δδ〉 and rms calibration uncertainties ɛ2α and ɛ2δ are best determined by comparison with accurate positions of sources strong enough that the noise plus confusion terms are much smaller than the calibration terms. We used the 37 single compact sources stronger than 10 mJy found in the FIRST (Faint Images of the Radio Sky at Twenty centimetres) radio survey (Becker, White and Helfand 1995). Their offsets 〈Δα〉 and 〈Δδ〉 are shown in Fig. 11. The mean offsets of our map are 〈Δα〉=−0.15±0.08 arcsec and 〈Δδ〉=0.18±0.06 arcsec. The source coordinates reported in the catalogue (see Table 3) have been corrected for these offsets.
From the distribution of 〈Δα〉 and 〈Δδ〉 shown in Fig.11 we have estimated the rms calibration errors: ɛα=0.486 arcsec and ɛδ=0.351 arcsec. Their small values show that our mosaic maps are not affected by relevant geometric distortions resulting from the approximation of a finite portion of the spherical sky away from the instrumental zenith with a bi-dimensional plane (see Perley 1989; Condon et al. 1998). Using the mean offsets and calibration errors obtained using the 37 sources stronger than 10 mJy in common with the FIRST survey, we have calculated the positional errors of all our sources. To verify that our positional errors are realistic also for faint sources, we used the 211 single compact sources stronger than 1 mJy found to be in common with the FIRST survey. The right ascension and declination differences 〈Δα〉 and 〈Δδ〉 were divided by the uncertainties σα and σδ (from equations 3 and 4). As expected, the normalized error distributions (Fig. 12) are Gaussian with nearly zero mean and unit variance, verifying that our positional uncertainties are accurate also for sources down to ∼1 mJy.
However, as shown in the flux distribution of Fig. 7, the majority of our radio sources have a flux lower than 1 mJy. Since we do not have other radio catalogues below 1 mJy to repeat our test, we decided to use our observations to test the reliability of the position errors down to the flux limit of our survey (∼0.135 mJy). In particular, we used the following procedure. Each mosaic map has been obtained by combining five or more individual pointings, whose centres are separated by ∼22 arcmin. Because in each individual map the FWHM of the primary beam is 31 arcmin, there are many overlapping regions where sources are detected in two different pointings. We considered the sources detected in each individual pointing as an independent data set and used the sources in common (i.e. the sources detected in the overlapping regions) to test the reliability of the positional errors. We have a total of 134 sources in the overlapping regions with a flux density lower than 1 mJy. The right ascension and declination differences of these 134 sources were divided by the combined errors σRA and σDEC, where σ2RA=σ2α1+σ2α2 and σ2DEC=σ2δ1+σ2δ2 with σα1,σα2,σδ1 and σδ2 from equations (3) and (4). As shown in Fig. 13, these error distributions for faint sources also agree with the expected Gaussians (smoothed curves).
In conclusion the positional uncertainties obtained using equations (3) and (4) are accurate also for sources down to the limit of our survey (∼0.135 mJy) and they can be used to estimate the reliability of the optical and infrared identifications. The rms position uncertainties σα and σδ are plotted as functions of the flux density in Fig. 14. Typical values are ∼2 arcsec for the fainter sources (∼0.13 mJy) and ∼0.6 arcsec for the brighter sources (>10 mJy).
5.1.2 Bandwidth smearing
The principles upon which synthesis imaging is based are strictly valid only for monochromatic radiation. When radiation with a finite bandwidth is accepted, aberrations, or so-called bandwidth smearing, in the image will result. These take the form of radial smearing, which worsens with increasing distance from the centre. The peak response to a point source simultaneously declines in a way that keeps the integrated flux constant. In other words the bandwidth smearing reduces the peak flux density SP of a source but not its integrated flux density SI. However, in our case, since we used spectral line mode where the bandwidth is narrower than in continuum mode, the bandwidth smearing should be negligible. To verify that this is the case, a strong calibrator (the source 3C 286) was observed at position offsets of 5, 10, 15, 20, 25, 30, 35 arcmin in two orthogonal directions (north-south and east-west) and we made a Gaussian fit of the source in each of the offset positions. The ratio between peak and total flux density (SP/SI) as a function of the off-axis angle is shown in Fig. 15; the bandwidth smearing is negligible up to r≃30 arcmin from the centre of each single pointing. Therefore the resulting mosaic maps, where the distance between the single pointings is ∼22 arcmin, are not affected by relevant bandwidth smearing and point sources should have SP/SI>0.98 everywhere.
6 The Source Counts
The complete sample of 867 sources with SP>5σ was used for the construction of the source counts. Sources with multiple components are treated as single radio sources. For all the sources we have used the total flux in computing the source counts. Every source was weighted for the reciprocal of its visibility area, Ω(SP)−1 (see Figure 6), which is the area over which the source could have been seen above the adopted limit. The 1.4-GHz source counts of our survey are summarized in Table 5. For each flux density bin, the average flux density in each interval, the observed number of sources in each flux density interval, the differential source density (in sr−1 Jy−1), the normalized differential counts nS2.5 (in sr−1 Jy1.5) with estimated errors (as n1/2S2.5) and the integral counts (in sr−1) are given. The normalized 1.4-GHz counts are plotted in Fig. 16 where, for comparison, the differential source counts obtained with other 1.4-GHz radio surveys are also plotted, while the integral source counts (deg−2) are plotted in Fig. 17. The solid line in Fig. 16 represents the global fit to the counts obtained by Windhorst et al. (1990) by fitting the counts from 24 different 1.4-GHz surveys. The open circles represent the counts obtained using the 1.4-GHz survey FIRST (White et al. 1997). Finally, the open stars represent the 1.4-GHz counts obtained by Gruppioni et al. (1997) in a recent radio survey in the Marano Field (RA=03h 15m, Dec.=−55 13′).
As shown in the figure, there is a very good agreement between our counts and those obtained with other surveys. In particular, the points at the fainter flux level confirm the well-known flattening observed in the normalized differential source counts below a few mJy (Windhorst et al. 1990). The roll-off of the sources in the FIRST survey (open circles in Fig. 16) at flux densities less than ∼2 mJy arises because of the low peak flux densities of many faint, extended sources, which make them undetectable in the FIRST survey. As noted by White et al. (1997) this is a clear indication of the incompleteness of the FIRST survey at the faint limit. On the other hand, the very good agreement between our counts and those obtained with other surveys (also at very faint flux levels) and the lack of any roll-off at faint flux levels in our counts indicate that our procedure for the source extraction (see above) yields very good results and that our sample is not affected by incompleteness at the faint limit.
A maximum-likelihood fit to our 1.4-GHz counts with two power laws:
gives the following best-fitting parameters: α1=1.71±0.10, α2=2.36±0.20, Sb≃ 0.5 mJy. These values suggest that the re-steepening of the integral counts towards a Euclidean slope starts below 1 mJy, in agreement with Gruppioni et al. (1997), Condon & Mitchell (1984) and Windhorst, van Heerde & Katgert (1984), while Windhorst et al. (1990), by fitting the counts of several 1.4-GHz surveys, found that the change in slope starts around 5 mJy.
7 Comparison with the NVSS and First Radio Catalogues
The regions that we are observing with the VLA are also covered by the NVSS (Condon et al. 1998) and FIRST (Becker et al. 1985; White et al. 1997) radio surveys. The NVSS (NRAO VLA Sky Survey) covers the sky north of J2000 δ=−40 at 1.4 GHz with θ=45 arcsec resolution and a nearly uniform sensitivity of 0.45 mJy (1σ), while the FIRST (Faint Images of the Radio Sky at Twenty centimetres) will cover over 10 000 deg2 at 1.4 GHz with a typical 1σ rms of 0.15 mJy and a resolution of 5.4 arcsec.
It was therefore natural to make a comparison between our results and the results obtained by the NVSS and FIRST surveys on the same regions of the sky. In Section 7.1 we discuss the reliability of the software that we are using to extract the radio sources (SAD in the AIPS version 1996 October), while in Section 7.2 we report the comparison between our VLA observations and the NVSS and FIRST source catalogues.
7.1 SAD on the NVSS and FIRST maps
As a first step we used the NVSS and FIRST radio maps and the published catalogues to test the reliability of the software that we are using to extract radio sources. From the NVSS and FIRST archives, we retrieved the maps covering the region of the sky N2 (see Table 1).
From the NVSS archive we retrieved the map I1640P40 while from the FIRST archive we retrieved the maps F16330+40417, F16330+41132, F16360+40417, F16360+41132, F16390+40417 and F16390+41132.
Using the AIPS software SAD on these maps, we obtained two lists of sources (one for the NVSS and one for the FIRST) in the same region covered by our VLA observations. These two lists were compared with the list of sources obtained from the NVSS and FIRST catalogues in which are reported the sources above the 5σ limits of each survey. The results are reported in Table 6 and in Fig. 18.
All three NVSS sources missing in the list obtained by us using SAD on the NVSS map are present in our deeper VLA map. Therefore they are real sources not detected by SAD. Their flux densities are lower than 3 mJy, close to the 5σ limit of the NVSS map.
Three (FIRST 163637.9+410511, 163707.5+405125 and 163815.4+405840) of the eight sources present in the FIRST catalogue but not detected by us using SAD on the FIRST map are also not detected in our deeper VLA maps to a flux limit of 0.1 mJy, so they are spurious sources. The other five sources missing in the FIRST list (as well as the four sources detected by SAD but not in the FIRST catalogue) have a peak flux density 1.2 mJy, i.e. very close to the FIRST flux limit. These differences are probably the result of the unreliable Gaussian fit of faint sources as discussed above. Moreover we have to consider that the source lists (our SAD list and the catalogue lists) are obtained using different source extraction software. In fact, both NVSS and FIRST sources are extracted with special AIPS-based routines written exclusively for each survey (see Condon et al. 1993 and White et al. 1997 for more details).
Therefore, in conclusion, we are confident in using SAD as source extraction software, keeping in mind that for faint sources the flux obtained with a Gaussian fit may be unreliable.
7.2 Comparison between our VLA observations and the NVSS and FIRST source catalogues
The primary difficulty in comparing the three surveys is the difference in angular resolution. A summary of the main properties of the NVSS, FIRST and our VLA survey is reported in Table 7.
The high-resolution surveys will partially resolve extended sources that appear point-like in the low-resolution surveys, so that the flux density of the sources will be smaller in FIRST and our survey than in the NVSS. To minimize this problem, we restricted the comparison to compact sources (i.e. sources with SI/SP<2) with an off-axis angle in our VLA map less than 34 arcmin. The restriction on the off-axis angle allows us to use only the sources detected in regions where the 5σ limit is well below the limit of the FIRST survey (see Tables 2 and 7). For the NVSS and FIRST surveys we used the public source catalogues as available on 1998 February 1.
7.2.1 Our VLA survey versus the NVSS survey
Using the restriction of an off-axis value lower than 34 arcmin and a peak flux density SP>2.3 mJy (∼5σ rms limit of the NVSS survey) we have 109 compact sources in common with the NVSS survey. In Fig. 19 we show a comparison between our VLA and NVSS total flux densities. A comparison between the flux density scales is discussed below. Besides these 109 common sources, there is one NVSS source (NVSS J142753+330837) that we did not detect in our survey plus five sources (with SP>2.3 mJy) detected in our maps but not in the NVSS survey. In Figs 20 and 21 we display a contour plot of these sources. For comparison, a contour plot of the FIRST image in the same region of the sky is also shown. As shown in Fig. 20 the NVSS source missing in our survey is also not detected in the FIRST survey. Its NVSS radio flux is 2.3 mJy, very close to the 5σ detection limit of the survey. It is probably a spurious NVSS source but it could also be a variable radio source. For the five sources missing in the NVSS survey, Fig. 21 clearly shows that one (ELAISR20 J142917+332626) is missing because our higher-resolution survey has resolved an extended NVSS source, one (ELAISR20 J142940+330552) is completely undetected on the NVSS map, and the other three sources are detected also in the NVSS map but below the 5σ limit (∼ 2.25 mJy).
All these five sources are present in the FIRST catalogue (and in the FIRST map as clearly shown in Fig. 21). Their absence in the NVSS catalogue is an indication of the incompleteness of the NVSS survey near that limit. A summary of the comparison between our survey and the NVSS and FIRST catalogue, the relative completeness and reliability is reported in Table 8 as a function of various flux density limits.
7.2.2 Our VLA survey versus the FIRST survey
Using the restriction of an off-axis value lower than 34 arcmin and a peak flux density SP>1.0 mJy (the limit of the FIRST catalogue), we have 216 compact sources in common with the FIRST survey. Six (FIRST J142953+330541, J143027+332249, J160636+543249, J160647+541511, J163607+410000 and J163730+ 405951) of these 216 common sources are flagged in the FIRST catalogue as a possible sidelobe of a nearby bright sources. In Fig. 22 we show a comparison between our VLA and FIRST total (integrated) flux densities.
Besides these 216 common sources, there are 14 FIRST sources (three of which are flagged: FIRST J163640+410016, J163646+ 405442 and J163654+404757) that we did not detect in our survey and 28 sources present in our survey but not in the FIRST catalogue. A flux distribution and contour maps of these last 28 sources are shown in Figs 23 and 24 respectively, while contour maps of the 14 FIRST sources that we did not detect are shown in Fig. 25. As shown in Fig. 23, all the radio sources missing in the FIRST catalogue have a peak flux density lower than 2 mJy. However, as shown in Fig. 24, many of them appear on the FIRST maps. This result confirms the incompleteness of the FIRST survey below ∼2 mJy, as already indicated by the differential source counts reported in Fig. 5. On the other hand, Fig. 25 shows that many of the 14 FIRST sources missing in our survey probably are not real. Some of them may be uncleaned residual sidelobes around strong sources (see, for example, FIRST J161318+541607 and J163815+405840) or simply spurious sources in a not well cleaned map (see, for example, the FIRST maps of FIRST J161351+543258, J163634+413013 and J163706+405125). The completeness and reliability of the FIRST survey as a function of various flux limits are reported in Table 8.
7.2.3 Flux comparison between the three surveys
As shown in Figs 19 and 22 the flux densities of our survey are in good agreement with the FIRST and NVSS flux densities over two orders of magnitude. However, as expected, the high-resolution surveys tend to estimate lower fluxes than the lower-resolution survey. This effect is evident in the lower panels of Figs 19 and 22: our VLA flux densities are, on average, lower than the NVSS flux densities but higher than the FIRST flux densities. High-resolution surveys, with their smaller synthesized beam size, lose flux as a result of the resolution surface brightness effect.
Using the Very Large Array (VLA) radio telescope, we observed at 1.4 GHz a total area of 4.222 deg2 in the ISO/ELAIS regions N1, N2 and N3. The lowest flux density reached by our observation is 0.135 mJy (at 5σ level) over an area of 0.118 deg2, while the bulk of the observed regions are mapped with a flux density limit of 0.250 mJy (5σ). The data were analysed using the NRAO AIPS reduction package. The source extraction has been carried out with the AIPS task SAD. The reliability of SAD has been tested using the maps of the radio surveys FIRST and NVSS.
Considering all the available observations, we detected a total of 867 sources at 5σ, 44 of which have multiple components. These sources were used to calculate the normalized differential source counts. They provide a check on catalogue completeness and reliability plus information about source evolution. A comparison with other surveys shows very good agreement, confirming the presence of the well-known flattening of the counts below 1 mJy, the completeness of our catalogue, and the reliability of our procedure for source extraction.
A comparison with the FIRST and NVSS radio surveys quantifies the completeness of these two surveys near their flux density limits (5σ) of 1 and 2.3 mJy respectively. Considering the whole sample, we found a completeness of 96+2-3 per cent for the NVSS and 89+2-3 per cent for the FIRST survey. The probable cause of the lower completeness in the FIRST survey is the ‘missing flux’ caused by the higher spatial resolution as discussed by Becker et al. (1995) and White et al. (1997). Near their flux density limit the incompleteness reaches values of ∼40 per cent for the NVSS and ∼22 per cent for the FIRST survey, although these values are based on quite small numbers of sources. A flux comparison between the three surveys has shown that our survey with the VLA array in C configuration is a good compromise between high- and low-resolution radio surveys for our present purposes. The positional errors of the radio sources are ∼2 arcsec for the fainter sources (∼0.13 mJy) and ∼0.6 arcsec for the brighter sources (>10 mJy). This small value will enable us to obtain reliable optical and far- infrared identification of the radio sources.
This work was supported by the EC TMR Network programme (FMRX-CT96-0068). RGM thanks the Royal Society for support. We thank David Helfand for a careful reading of the manuscript, discussions and suggestions; Rick White for discussions on the optimal pointing grid; and both Bob Becker and Jim Condon for discussion on the optimal observing strategy; and Bob Becker for the provision of a FIRST observe file.