2.1. Surveys and operations

The public SkyMapper Southern Survey claims 50% of the observing time on the SkyMapper Telescope, with a focus on good-seeing time, while bad-seeing time since 2015 has been used for the SkyMapper Transient Survey, a supernova survey described by Scalzo et al. (Reference Scalzo2017), and a fraction of time is kept available for third-party programmes. The public Southern survey contains two components with different depth and saturation limits.

The Shallow Survey targets the entire Southern hemisphere with short exposures, ensuring full-hemisphere coverage early in the project and a bright saturation limit. Several dithered visits aim at fully covering the target area despite gaps in the CCD mosaic, and repeat observations to ascertain the static or transient nature of any detected sources. During each visit, a six-colour sequence is observed within less than 5 min. While the first year of SkyMapper survey operations was dedicated mostly to the Shallow Survey, this now continues only in the brightest nights around full moon. The Shallow Survey is complete to nearly magnitude 18 in all six bands uvgriz and provides the calibration reference for the following Main Survey.

The Main Survey targets the same sky area in the same six bands, but aims to be complete to g, r ≈ 22. In combination with the Shallow Survey, the entire SkyMapper Southern Survey will provide calibrated uvgriz photometry from magnitude 9 to 22. The Main Survey includes two six-colour sequences, each taken within a 20-min interval, as well as additional visits to collect pairs of gr and iz images. The gr pairs are acquired under dark and grey sky, while the iz pairs are primarily collected in the period between astronomical and nautical twilight.

SkyMapper’s regular operations model includes daily calibration procedures including bias frames, twilight flatfields, and several visits to dedicated SkyMapper standard fields during the night. Eight such standard fields have been defined, each of which includes a Hubble Space Telescope (HST) spectrophotometric standard star. These standard field observations have not been used for DR1, but will be utilised in future releases. A paper with more complete technical and procedural references is in preparation (Onken et al. in preparation).

2.2. Filters

The SkyMapper filters are described in detail in Bessell et al. (Reference Bessell, Bloxham, Schmidt, Keller, Tisserand and Francis2011). The filters u, v, g, and z all comprise coloured glass combinations; the r-filter was made by combining magnetron-sputtered short- and long-wave passband coatings; the i-filter is coloured glass plus a short-wave passband coating; the z-band long-wavelength cut-off is defined by the cut-off in the CCD sensitivity. All filters had two-layer anti-reflection coatings applied. The diversity in fabrication methods for the SkyMapper griz filters helps explain why their passbands differ in width and placement from the SDSS griz passbands.

After polishing, the u and v bandpass glasses were slightly thinner than anticipated. As a result, they have red leaks near 700 nm, around 1% for u and much less for v. These are evident in u-band observations of K and M stars (see Figure 1).

Figure 1. Normalised transmission curves for SkyMapper filters including atmosphere, telescope, and detector from Bessell et al. (Reference Bessell, Bloxham, Schmidt, Keller, Tisserand and Francis2011). Note that SkyMapper has an ultraviolet u-band and a violet v-band, where SDSS has only a single u-band. Central wavelengths and filter widths of griz-filters vary up to 40 nm relative to their SDSS cousins. Right: u-band filter curve including atmosphere, telescope, and detector, for airmass 1 and 2. The red leak at 700–750 nm wavelength increases relative to the main transmission band with airmass, because UV light is heavily absorbed by the Earth’s atmosphere, while far-red light remains nearly unaffected. As a result, the measured u-band magnitudes of red stars increase with airmass as the relative contribution from the leak increases.

The transmission curves of the filters are shown in Figure 1, they were measured at many places across the aperture and averaged to provide the mean filter passband. The all-glass filters u, v, g, and z have remarkably uniform wavelength transmission with position and with changing field angle. The interference filter defined edges on the r- and i-filters, however, change slightly with angle. The average transmission was then multiplied with one airmass of atmospheric extinction and the mean QE response curve of the E2V CCDs. The CCD response curves and the interference coatings may change over the period of the survey and thus our calculated SkyMapper passbands may need adjustment. This investigation is beyond the scope of DR1 and will be assessed from observations of our spectrophotometric standard stars.

2.3. Detector properties

SkyMapper’s 32 detectors are arranged in four rows of eight CCDs, with horizontal spaces of ~0.8 arcmin between the mosaic columns, a small gap of ~0.5 arcmin between the central rows, and larger gaps of ~3.2 arcmin separating the two upper rows from each other and the two lower rows from each other. The read-out noise of the amplifiers varies broadly between 6 counts and 14 counts. In Shallow Survey exposures, fringes are only apparent in the i- and z-band. As their amplitude is below the read-out noise, we do not defringe any of the DR1 data released here.

Figure 2 shows the efficiency curves of the 32 CCDs after normalising them at 500 nm wavelength. Variations in the blue sensitivity are already evident around 400 nm, which means that the uvg passbands will differ from CCD to CCD, with corresponding subtle differences in the zero-point (ZP) calibration and colour terms. These variations are not considered in this data release, but will be included in a future release.

Figure 2. Quantum efficiency (QE) of the CCDs in the mosaic of the SkyMapper camera. Points indicate the median values after normalising to a common level at 500 nm wavelength, the grey bars indicate the 16th-to-84th-percentile range at each wavelength, and the thin errors bars indicate the minimum and maximum values. A more detailed QE curve for a single typical CCD is shown with a line. Different short-wavelength sensitivities are evident, which will lead to subtle variations in the uvg passbands.

4.1. Ingest

Images from the telescope are transferred each morning via ethernet to long-term storage at the National Computational Infrastructure (NCI) located on the ANU campus. When a night is ready to be processed by the SDP, the data is moved from the NCI image archive to a local disk, and the individual images are then ingested into the SDP, which reorganises the 64 extensions of the FITS file (one per amplifier) into 32 separate FITS images (one per CCD). Preliminary quality assurance (QA) checks are performed to eliminate bad images from further processing. The initial processing during the ingest phase includes: suppression of interference noise; identification of pixels subject to saturation or blooming; overscan subtraction; correction for cross-talk; creation of image masks; and flipping the image axes to restore the on-sky orientation.

4.1.1. Interference noise

The SkyMapper detectors are subject to an intermittent, high-frequency source of interference, which imposes a strong sinusoidal pattern on the images with a wavelength between 6 and 7 pixels. The phase and amplitude can vary from row to row in an image, so each row read out by each amplifier is corrected individually by fitting a sine curve. In rare cases, the procedure does not adequately remove the pattern from a given row. The mean amplitude of the subtracted sine curve was 4.4 counts. An example of the interference noise and its removal are shown in Figure 6.

Figure 6. Left: Interference noise that appears intermittently in SkyMapper images, with amplitudes and row-to-row phase shifts that vary within a single readout. The wavelength along each row is between 6 and 7 pixels. We obtain a sinusoidal fit to each row, which is then subtracted to remove the interference. Right: The same image after removal of the fit to the noise. Residual row-to-row variations in the bias are removed by a separate step.

4.1.4. Amplifier cross-talk and ringing

The SkyMapper mosaic is subject to electronic cross-talk between its 64 amplifiers. The most severe cross-talk, and the only one treated by the current version of the SDP, occurs between adjacent amplifiers on the same CCD, with a fractional amplitude of ~5 × 10⁻⁴. However, in the presence of fully saturated pixels, other amplifiers can display cross-talk artefacts (switching between negative and positive counts on alternative amplifiers) with fractional amplitudes as high as 1 × 10⁻⁴ or about seven counts. Fully saturated pixels also induce amplifier ringing, where the next non-saturated pixel read out has 0 counts and the subsequent two pixels have severely suppressed count levels. Figure 7 shows examples of these behaviours.

Figure 7. Example image showing cross-talk between amplifiers, visible as depressed counts opposite the bright star on the left, and amplifier ringing, visible as dark regions immediately adjacent to the pixel blooming associated with the bright star on the left.

Cross-talk between the two amplifiers on the same CCD is corrected by subtracting a second-order polynomial in the count level, with an additional subtractive component for the counterpart of a fully saturated pixel. The correction is made in a single pass through each CCD, beginning with the left-hand amplifier and then correcting the right-hand amplifier.Footnote ³

The pre- and post-scan pixels are then trimmed from both amplifiers of a given CCD, and the two amplifiers are stitched together into a single FITS image. Finally, the images and masks are flipped in the x-direction to restore the on-sky orientation.

4.1.6. Mask creation

Image-specific masking augments the global bad pixel mask. We identify saturated pixels using a threshold of 55000 counts. The fully saturated pixels and the three neighbouring pixels subject to amplifier ringing are all masked with a saturation flag. Pixels subject to cross-talk from saturated pixels in the other amplifier on the same CCD are masked with a separate flag. Finally, as described in Section 4.4, pixels in which cosmic rays have been identified and corrected are flagged. The various image-mask flags are described in Table 2.

Table 2. Meanings of image mask flags.

Very bright stars leave some strong and curiously shaped optical reflections in the images, which have not been removed (see Figure 8 for an example).

Figure 8. Optical reflections from very bright stars (here, Antares) are not removed from images in DR1. This image is an inverted-colour image from the SkyMapper SkyViewer.

4.2. Astrometry

The astrometric solutions for the survey images are tied to the fourth U.S. Naval Observatory CCD Astrograph Catalog (UCAC4; Zacharias et al. Reference Zacharias, Finch, Girard, Henden, Bartlett, Monet and Zacharias2013). The 32 CCDs are each run independently through the solve-field program from the Astrometry.net software suite (version 0.43; Lang et al. Reference Lang, Hogg, Mierle, Blanton and Roweis2010), using a set of UCAC4 index files, to generate tangent-plane World Coordinate System (WCS) solutions. Each CCD for which solve-field succeeds is then run through Source Extractor to derive (x, y) pixel positions and RA, Dec values for each source, as well as to calculate the median seeing and elongation values for the CCD (using Source Extractor parameters FWHM_IMAGE and ELONGATION). For each source in the image, the closest-matching RA, Dec position from the UCAC4 catalogue is taken as the true position,Footnote ⁴ and we then fit a Zenithal Polynomial (ZPN) projection (see Calabretta & Greisen Reference Calabretta and Greisen2002) to the pixel+coordinate sets using a fixed curvature term of PV2_3 = 3.3, which was determined by an independent fit to a well sampled mosaic image. The ZPN solution provides a good fit to the radial distortions in the SkyMapper focal plane, but the ZPN formalism is not understood by several of the software tools utilised by the SDP. Hence, the ZPN solution is then transformed into a TPV projection (tangent plane with generalised polynomial distortion)Footnote ⁵ , which can be understood by the software packages from AstrOmaticFootnote ⁶ (e.g., SExtractor, SWarp, and PSFEx).

After all 32 CCDs have been analysed, two QA tests are performed: the plate scale near the middle of each CCD is required to be between 0.487 and 0.507 arcsec (0.483 and 0.507 arcsec for the four corner CCDs); and the offset from the middle of the CCD to the middle of one of the central CCDs is required to be within 10 arcsec of the nominal value. Later, a third test is performed on the WCS solutions, requiring that the corner-to-corner lengths of each CCD are within 2 arcsec of a reference value, where the reference is dependent on the CCD position within the mosaic, the filter used in the observation, and the fractional Modified Julian Date (MJD). The dependence on the fractional MJD, which contributes 0.5 arcsec of variation across the observed fractional MJD range, is thought to reflect temperature gradients in the Corrector Lens Assembly, which evolve as a night progresses. This final test ensures that the WCS solutions do not extrapolate to unrealistic representations of the sky in regions where there may be too few stars to constrain the fit.

Overall, 98.6% of mosaic images used in DR1 yield valid WCS solutions for all 32 CCDs, and amongst the few images that yield between 1 and 31 valid WCS solutions, the mean number is over 28. The success rate is lowest in the u- and v-filter with 96.7% each, because the density of stars bright enough to match to UCAC4 is the lowest. Whenever a valid WCS solution is not available for a CCD, it has further calibrations applied as normal, after which a second attempt is made to obtain a WCS solution.

As an indicator of the quality of our WCS solutions in DR1, we show the median offset distances between the sources in our DR1 master table and the nearest source in Gaia DR1 (Lindegren et al. Reference Lindegren2016) as a function of RA and Dec in Figure 9. The overall median offset (restricted to distances less than 10 arcsec) is 0.16 arcsec and the [10%, 90%] range is [0.06 arcsec, 0.45 arcsec]. For cleanly detected objects with r-band magnitudes between 9 and 14 mag, the median offset is 0.12 arcsec, or about 1/20th of the typical PSF FWHM.

Figure 9. Median positional offset between the DR1 sources and the closest match from the Gaia DR1 catalogue, as a function of Right Ascension and Declination. The grayscale ranges from 0 (white) to 0.5 arcsec (black). The overall median offset is 0.16 arcsec.

4.3. Calibration data

The individual bias and flatfield exposures are combined into a series of master frames with variable durations of validity. Hard boundaries on the input frames considered for each master calibration frame are imposed at each instance of detector warming (when we find subsequent rapid evolution of the flatfield shape), as well as at any other significant change to the image properties, such as modification of detector voltages (Section 4.3.3), changes to flatfield position angles (PAs), or cleaning of the telescope optics.

4.3.3. Detector ‘tearing’

In early 2014, it was discovered that all 32 of the CCDs exhibited curvilinear features in which approximately 5% of the counts appeared to be shifted to neighbouring columns. An example is shown in Figure 10. The position of these features differed from amplifier to amplifier and changed position from image to image, and hence we leave them uncorrected in DR1. The features were eventually identified as being instances of ‘tearing’ (cf. Doherty Reference Doherty2013; Rasmussen Reference Rasmussen2014) and were traced to the detector bias voltage settings being used at the time. From 2014 July, an updated configuration, modelled on the settings used by the European Southern Observatory, eliminated the features from subsequent SkyMapper images. However, the relative-strength nature of the tearing means that in DR1 the features may be visible in flatfields with few input frames or in high-background images taken prior to 2014 July.

Figure 10. Detector ‘tearing’ on four amplifiers across four CCDs from an r-band flatfield image from UT 2014 March 31. The amplitude of the features is approximately 5% of the nominal count level. The position of the features varied with amplifier and with each image. Modifications to the detector bias voltages eliminated the features from 2014 July onward.

4.4. Application of calibrations

For each CCD frame of a science image, we first subtract the associated master bias frame. Then we prepare the science frame for PC-fitting by creating a background map using SExtractor and identify sources in the image for masking from the fit. After subtracting the background, we fit unmasked pixels with a weighted least-squares algorithm, using the PCs associated with the master bias for that night (Figure 11). Again, each half of the CCD is treated separately. The number of PCs included in the fitting procedure is determined by the fraction of the pixels in the row that remain unmasked: all 10 PCs for at least 25% unmasked; five PCs if 10–25% of the row is unmasked; one PC if 2–10% of the row is unmasked; and no fitting is performed if less than 2% of the row is unmasked.

Figure 11. Example of bias residuals after traditional overscan correction used in the SkyMapper Early Data Release (EDR, left) vs. PCA model-based bias removal used in DR1 (right).

After the bias PCA treatment, the master flatfield is divided out.

Next, cosmic rays are detected and removed from the science frames using the lacosmicx software package,Footnote ¹⁰ a fast Python implementation of the L.A.Cosmic routine (van Dokkum Reference van Dokkum2001). Each amplifier is treated separately to allow for the use of specific read noise settings. From over 7 000 bias frames taken during the DR1 period, we determine the read noise in each frame, and determine for each amplifier a mean level and an rms variation over time. To avoid removal of spurious cosmic rays from uncorrected interference noise in higher noise frames, we use a high fiducial value for the read noise, setting it to 〈RON〉 + 3σ_RON + 1 counts for each amplifier. The range of input read noise values is 7.36–17.04, with a median of 9.95 counts. The saturation level is set to 60 000 counts, with a Laplacian-to-noise limit of 6σ and a fractional detection limit for neighbouring pixels of 1.35σ. The image mask is provided as an input to lacosmicx in order to avoid detecting known bad pixels.

We then still observe residual differences in the sky level between the two halves of each CCD due to non-linearity at low count rates. We call SExtractor again to create a background map for each half of the image. From the two background maps, the median value for seven columns near the centre of the CCD (omitting the three columns closest to the middle) are compared, and the difference is then applied as an additive offset to all pixels in the lower of the two halves.

Next, an existing WCS solution is inserted into the image header. The pixel values are then converted to 16-bit integers, with a BZERO value of 32 668 and a BSCALE value of 1, to give a resulting count range of −100 to 65 435 (preserving the noise near zero counts), and the images and masks are compressed using the lossless fpack FITS Image Compression Utility (Pence, Seaman, & White Reference Pence, Seaman and White2009) from the CFITSIO library.Footnote ¹¹

Any images for which the raw image could not generate a WCS solution are tried again after the calibrations have been applied. If still no WCS could be obtained, the CCD is discarded for further analysis.

4.5. Photometry

For any mosaic image in which at least one CCD has yielded a valid WCS solution, the suitable CCDs are then run through our photometry stage.

First, SExtractor is run on each CCD. We use a detection threshold of 1.5σ, the default (pyramidal) filtering function, a minimum area of 5 pixels, a minimum deblending contrast of 0.01, employ a local background estimate, utilise cleaning around bright objects, and measure photometry in a series of apertures (diameters of 4, 6, 8, 10, 12, 16, 20, 30, 40, and 60 pixels). We provide SExtractor with the image-specific mask as a Flag image and a version of the global bad pixel mask as a Weight image. The parameters measured by SExtractor are shown in Table A1 of the Appendix.

4.5.2. Zero-point calibration

For SkyMapper DR1, the ZP calibration is referenced to the AAVSO Photometric All Sky Survey (APASS DR9; Henden et al. Reference Henden, Templeton, Terrell, Smith, Levine and Welch2016) and 2MASS. From APASS DR9, we select suitable calibrator stars with clean flags,Footnote ¹² magnitude errors below 0.15 and magnitudes in the ranges of g = [11.5, 16.5] and r = [10.5, 16]. From 2MASS, we select stars with clean flags, no neighbours within 10 arcsec and K < 13.5. This leaves approximately 12.8 million APASS DR9 sources and 2.5 million 2MASS sources at δ < +30°. Then we compute predicted magnitudes in all SkyMapper filters with a method trained on photometry from the first data release of the Pan-STARRS1 3π Steradian Survey (PS1; Chambers et al. Reference Chambers2016; Flewelling et al. Reference Flewelling2016; Magnier et al. Reference Magnier2016).

We first dereddened stars in all three catalogues using the reddening maps by Schlegel, Finkbeiner, & Davis (Reference Schlegel, Finkbeiner and Davis1998; hereafter, SFD) and bandpass coefficients derived from a Fitzpatrick (Reference Fitzpatrick1999) extinction law. Then we used a large sample of stars that are well-measured in all three catalogues to fit linear relations predicting PS1 magnitude from APASS and 2MASS. As a final step, we apply a linear PS1-to-SkyMapper transformation that was derived using the filter curves of PS1 and SkyMapper together with the unreddened F- and G-type main-sequence and sub-giant stellar templates from Pickles (Reference Pickles1998).

From PS1, we find that trends of mean stellar colour with reddening are minimised if we adopt for the mean reddening of the average star, a prescription of

$$\begin{equation*} \frac{\text{d}E(B-V)_{\rm eff}}{\text{d}E(B-V)_{\rm SFD}} = \left\lbrace \begin{array}{@{}l@{\quad }l@{}}0.82, & \text{if}\ E(B-V)_{\rm SFD}<0.1 \\ 0.65, & \text{otherwise}. \end{array}\right. \end{equation*}$$

Schlafly & Finkbeiner (Reference Schlafly and Finkbeiner2011) had already found that the SFD maps overestimate reddening on average and suggest to generally scale them by a factor 0.86. We also find that we need to reduce more strongly the reddening at E(B − V)_SFD > 0.1, and fit a coefficient in order to remove trends of the mean star colour with reddening. We find the same coefficient 0.65 as previously suggested by Bonifacio, Monai, & Beers (Reference Bonifacio, Monai and Beers2000) for stars of V = [11, 15]. This scaling probably accounts for the fact that a fraction of the stars reside in front of some of the dust that contributes to the IR emission in the SFD maps.

We estimate reddening coefficients for our filters using a Fitzpatrick (Reference Fitzpatrick1999) extinction law with R _V = 3.1. This appears to best match the empirical bandpass coefficients found by analysing trends in the observed colours of stars (Schlafly & Finkbeiner Reference Schlafly and Finkbeiner2011) and QSOs (Wolf Reference Wolf2014) with reddening. Our mean bandpass reddening coefficients R _band are listed in Table 1, and while they apply strictly only to flat spectra, as R _band depends on the object SED, this effect is mostly relevant in the u-band.

We restrict calibrator stars to certain colour ranges after de-reddening in order to ensure sufficient linearity of the colour transformation. For calibrating iz images, we use stars with (J − K)₀ = [0.4, 0.75] and (g − r)₀ = [0.2, 0.8]; for gr images, we require only (g − r)₀ = [0.2, 0.8] and for uv images, we use (g − r)₀ = [0.4, 0.8].

Final SkyMapper AB magnitudes are estimated from APASS AB and 2MASS Vega magnitudes using

$$\begin{eqnarray*} u_2 & = & g_{\rm AP} + 0.622 + 1.757(g-r)_0 + 0.953 E(B-V)_{\rm eff}; \\ u_1 & = & g_{\rm AP} + 0.575 + 2.050(g-r)_0 + 1.002 E(B-V)_{\rm eff}; \\ v & = & g_{\rm AP} + 0.033 + 2.240(g-r)_0 + 0.734 E(B-V)_{\rm eff}; \\ g & = & g_{\rm AP} - 0.010 - 0.295(g-r)_0 - 0.306 E(B-V)_{\rm eff}; \\ r & = & r_{\rm AP} - 0.003 + 0.042(g-r)_0 + 0.014 E(B-V)_{\rm eff}; \\ i & = & J_{\rm 2M} + 0.270 - 0.060(g-r)_0 + 0.060 E(B-V)_{\rm eff} \\ && +\, 0.31(r-J)_0 + 0.207(g-J)_0 + 0.062(J-K)_0; \\ z & = &J_{\rm 2M} + 0.550 + 0.062(g-r)_0 + 0.510 E(B-V)_{\rm eff} \\ && +\, 0.880(J-K)_0. \end{eqnarray*}$$

Because of the red leak in the SkyMapper u-band, atmospheric extinction makes the filter transformations airmass-dependent (see Figure 1). We calculate predicted u-band magnitudes for airmasses of 1 and 2, and interpolate between those based on the airmass of each exposure.

For each SkyMapper source having a positive 15 arcsec-aperture flux, a 15 arcsec-aperture magnitude error less than 0.04 mag, SExtractor FLAGS < 4 (i.e., not saturated), NIMAFLAGS = 0 (i.e., no masked pixels within the isophotal area), and semi-minor axis size of at least 1 pixel, we find the closest match on the sky from the sample of calibrator stars, with a maximum positional difference of 10 arcsec.

We compute the differences in observed and predicted magnitudes for each star on all available CCDs for the mosaic image, along with the whole-of-mosaic (x, y) positions for those stars (adopting the median CCD-to-CCD offsets from ~10 000 images with clean WCS solutions). With the positions and ZP estimates from the entire mosaic, we then simultaneously fit for linear gradients in x and y, in order to take atmospheric extinction gradients across the large field of view into account.

We perform four passes of 2.5σ-clipping to determine the final ZP gradient, and fix the gradients to be zero when five or fewer stars remain unclipped. The resulting ZP plane is then applied to the magnitudes for all sources detected in the mosaic. The RMS around the final ZP fit is recorded, but not applied to the individual magnitude errors. Most images in DR1 have between 300 and 1 000 calibrator stars per frame that are used for the ZP calibration, and in all bands at least 95% of frames have more than 100 calibrator stars.

We adjust all u-band magnitudes globally 0.05 mag brighter, which makes our measurements consistent with SkyMapper u-band predictions derived from measured PS1 g- and r-bands (see Section 5.1 for details).

Prior to uploading the photometry into the database, we perform a small amount of cleaning. Because SExtractor does not provide error estimates for the PA, we adopt the PA error used for the Chandra Source Catalog:

$$\begin{eqnarray*} e_{\rm PA} &=& {\rm atan2}((b+e_b),(a-e_a))\\ && -\, {\rm atan2}((b-e_b),(a+e_a)), \end{eqnarray*}$$

where a and b are the semi-major axis and semi-minor axis lengths, respectively, and e _a and e _b are their errors;Footnote ¹³ and we impose a minimum PA error of 0.01°. We also impose magnitude error floors of 0.0033 mag, and we remove sources having negative fluxes in their 30 arcsec apertures.

5.1. Photometric comparison: SkyMapper vs. PanSTARRS1 DR1

First, we assess the internal reproducibility of measured SkyMapper ‘PSF’ magnitudes of point sources from repeat visits and find that it ranges from 8 mmag rms in griz-bands to 12 mmag in the uv-bands. For a comparison against external data, we chose to use Pan-STARRS1 DR1 and estimate magnitudes in SkyMapper bands from existing measurements in PS1 bands. After checking colour trends with magnitude, we restrict this comparison to calibrator stars that appear well-measured in both surveys, thus ensuring that the transformation from PS1 to SkyMapper bands is reliable.Footnote ¹⁴ We deredden the PS1 SED using the modified mean reddening E(B − V)_eff specified in Section 4.5.2, and apply colour terms fitted on synthetic PS1 and SkyMapper photometry of Pickles (Reference Pickles1998) stars with luminosity class IV and V and colour range (g − r)₀ = [0.2, 1.0]. We apply the following equations to transform measured PS1 griz magnitudes into SkyMapper magnitudes, using a de-reddened PS1 colour of (g − i)₀ = g − i − 1.5 E(B − V)_eff:

$$\begin{eqnarray*} u_2 & = & g_{\rm PS} + 0.728 + 1.319(g-i)_0 + 1.075 E(B-V)_{\rm eff}; \\ u_1 & = & g_{\rm PS} + 0.722 + 1.461(g-i)_0 + 1.124 E(B-V)_{\rm eff}; \\ v & = & g_{\rm PS} + 0.280 + 1.600(g-i)_0 + 0.856 E(B-V)_{\rm eff}; \\ g & = & g_{\rm PS} - 0.011 - 0.162(g-i)_0 - 0.184 E(B-V)_{\rm eff}; \\ r & = & r_{\rm PS} + 0.000 + 0.021(g-i)_0 + 0.028 E(B-V)_{\rm eff}; \\ i & = & i_{\rm PS} + 0.013 - 0.045(g-i)_0 - 0.082 E(B-V)_{\rm eff}; \\ z & = & z_{\rm PS} + 0.022 - 0.054(g-i)_0 - 0.114 E(B-V)_{\rm eff}. \end{eqnarray*}$$

The g-band filter shows the most significant colour term, while the u- and v-band must be extrapolated from g, which is inherently less accurate. The v-band magnitude in particular also depends on metallicity and shows clear trends between the halo and disk population. Indeed, the motivation for the SkyMapper v-band is to act as a metallicity and surface gravity indicator for stars, and below we demonstrate the first step in our search for EMP stars (see Section 6.3).

In Figure 12, we show the difference between the measured and PS1-predicted SkyMapper photometry in histograms after binning in foreground reddening, as well as in sky maps. We restrict this comparison to the four non-extrapolated filters griz, and find an rms scatter in the magnitude difference of 23 mmag. The mean offset is generally less than 10 mmag, with the exception of a tail at higher reddening in z-band, where the median difference increases to 0.06 mag for E(B − V) = [0.5, 1.5]. The sky maps also show the bias in z-band towards low Galactic latitude where reddening tends to be higher. Apart from that the sky maps show low-level discontinuities at the SkyMapper field edges. SkyMapper fields are arranged in declination stripes with small overlaps in RA and Dec. However, additional passes of the Shallow Survey after the DR1 imaging period provide field overlaps to support a more homogeneous calibration.

Figure 12. Difference between SkyMapper magnitude as measured in DR1 and as predicted from PS1 magnitudes with bandpass transformations applied. This comparison is restricted to calibrator stars that are well measured and have reliable transformations. Top: histograms in bins of interstellar foreground reddening up to E(B − V) = 1.5. The bottom four panels show spatially resolved sky maps.

Despite the uncertainties introduced by extrapolating the SkyMapper u- and v-band, we can formally compare the photometry. Again we derive colour terms for SkyMapper u at airmass 1 and 2, although in this comparison, we use the median airmass of 1.11 to represent the whole comparison sample. We show the histograms of magnitude differences in Figure 13 and note that the rms scatter is 0.12 mag in both bands. After finding a mean offset of 0.05 mag in u we adjusted all our u-band magnitudes globally 0.05 mag brighter. We assume that this offset has originated from our calibration procedure, where we extrapolated the APASS DR9 gr photometry to the SkyMapper u-band. Even though we see slight offsets in v-band and a ~0.05 mag difference between regions of reddening less than E(B − V) = 0.1 and those with more reddening, we do not correct these. The shift with reddening is likely to be a metallicity effect, where halo and disk stars are expected to have different v − g colours by roughly the observed amount.

Figure 13. Difference between SkyMapper magnitudes and those predicted from PS1 photometry, as in Figure 12, but for the highly extrapolated filters u and v. The u-band has been globally adjusted by −0.05 mag, and the v-band shows a shift between regions of low vs. high reddening, which likely reflects the metallicity difference between the disk and halo populations in the Milky Way.

5.2. Morphology indicators

Separating point sources from extended sources is a challenge limited by the spatial resolution set by observing conditions and by PSF variations within a given frame. Any indicator of morphology will work best with bright sources but be increasingly ambiguous for noisier faint sources. We consider two morphology indicators: (i) the SExtractor stellarity index CLASS_STAR, which approaches 1.0 for point sources and typically has values between 0 and 0.9 for extended sources, and (ii) the difference between the PSF and Petrosian magnitudes.

In Figure 14, we show these two morphology indicators vs. the r-band magnitude for a large sample of objects including bright galaxies from the hemispheric 6dFGS and faint galaxies from the deeper and smaller area 2dFGRS (shown together with black points). We restrict the plotted sample to objects with good flags and no nearby neighbours. We note that both galaxy surveys contain compact galaxies, e.g., emission-line galaxies with high star-formation densities and active galaxies with dominant Seyfert-1 nuclei. In grey, we show a random sample of all DR1 objects matched in brightness to those two galaxy surveys. The two measures of morphology appear largely similar except for high-stellarity sources at the faint end: SExtractor tends not to attribute a high likelihood of stellarity to faint, noisy sources, while the difference between PSF and Petrosian magnitude for true point sources always scatters around zero, no matter the level of noise. Thus, CLASS_STAR tends to select proven point sources with low contamination and is incomplete at the faint end, while the magnitude comparison selects possible point sources with high completeness but increasing contamination at the faint end.

Figure 14. Morphology indicators vs. r-band magnitude: Left: SExtractor stellarity index CLASS_STAR. Right: Difference between PSF and Petrosian magnitude. Plotted are mag-limited galaxy samples from 6dFGS and 2dFGRS (black) and matching lower density samples of random DR1 objects (grey).

The fs_photometry table also contains for each separate detection our point-source likelihood measure CHI2_PSF, which is closer to the magnitude comparison in behaviour. It essentially expresses in χ² units, how well the sequence of aperture magnitudes, i.e., growth curve, follows the local model of a PSF growth curve.

5.3. Survey depth and number counts

As DR1 includes only images from the Shallow Survey, it is expected to be ~4 mag shallower than the full SkyMapper survey. In Figure 15, we show number counts per filter from our master table. These typically turn over around 18 mag in each band, with the exception of the shallower, relatively narrow, v-band that appears complete to 17.5 mag. Galaxies are included in this plot although they will be incomplete already at much brighter magnitudes given that they are more extended than a PSF and their outer parts will get lost in the noise of the Shallow Survey images.

Figure 15. Number counts in PSF magnitude for each band. Except for the v-band, the completeness limit is nearly 18 mag.

A given mean population of stars can be roughly characterised by a horizontal line in Figure 15. The horizontal offset of the number count lines then shows the mean colours of the stellar population. This suggests, e.g., that a v-selected population with v < 17.5 will only be largely complete to z < 15.

5.4. Colour space and colour terms for stars

In Figure 16, we explore the locus of point sources in several dereddened colour–colour diagrams. We select stars from two magnitude bins (r = [12, 12.5] and r = [14, 14.5]) with low reddening E(B − V) < 0.1. The first two panels demonstrate how SkyMapper colours alone can differentiate between cool dwarfs and cool giants using the u − g colour, and between AF-type main-sequence stars and horizontal-branch (HB) stars using the u − v colour. The SkyMapper-specific u and v bands bracket the Hydrogen Balmer break at 365 nm, and lower surface gravity causes narrower Balmer absorption lines and thus higher v-band flux and a redder u − v colour.

Figure 16. Dereddened SkyMapper colours measured for stars with E(B − V) < 0.1 and r _PSF = [12, 12.5] (black) or r _PSF = [14, 14.5] (grey). Top: Internal SkyMapper colours—we can easily see the bifurcation between cool dwarfs and cool giants in the top left panel, and separate horizontal branch (HB) stars from main-sequence (MS) stars in the top right panel; also shown is a region where metal-poor (MP) stars can be found easily. Bottom: Combined colour between a SkyMapper band and a 2MASS band or a GALEX band, respectively.

The bottom two panels exploit the fact that our database holds copies of the 2MASS and GALEX BCS catalogues (Bianchi, Conti, & Shiao Reference Bianchi, Conti and Shiao2014), to which we have pre-cross-matched all DR1 sources. Hence, multi-catalogue colours can be obtained with simple and fast table joins. In dereddening GALEX NUV magnitudes, we used R _NUV = 7.04, which we derived by adding the empirical E(NUV − g) coefficient determined by Yuan, Liu, & Xiang (Reference Yuan, Liu and Xiang2013) to the R _g of SDSS. The separation of the cool dwarf and giant branches is also clearly visible in optical-NIR colours such as i − K. Optical-to-near-UV colours such as NUV − g will further help to differentiate stars from galaxies and QSOs, when morphology or other indicators are unreliable.

Finally, in Figure 17, we show colour terms between the SkyMapper and SDSS bandpasses, from synthetic photometry performed on all Pickles stars (from main-sequence stars to supergiants) as well as on a grid of white dwarf spectra from Detlev Koester (private communication). We compare the SkyMapper u- and v-filters both to the SDSS u′-band, and the other four filters to their eponymous (g′r′i′z′) SDSS cousins. Between those cousins, the g-band shows the strongest colour terms, which reach up to 0.4 mag difference for cool stars.

Figure 17. Colour terms between SkyMapper and SDSS for the spectral libraries of main-sequence (IV/V, black points) and giant (I–III, light grey points) F–M stars of Pickles (Reference Pickles1998), and a grid of DA/DB white dwarf spectra from T _eff = [6000, 40000] K by Detlef Koester (priv. comm; dark grey points). The line shows a reddening vector for A _V = 1 using a Fitzpatrick (Reference Fitzpatrick1999) law.

6.1. Properties of known variable stars

We investigated 256146 known variable stars from the AAVSO International Variable Star Index (VSX; Watson, Henden, & Price Reference Watson, Henden and Price2006, Reference Watson, Henden and Price2017) with cross-matches in DR1 and look for brightness variations seen by SkyMapper repeat visits. We obtain this information, e.g., for the r-band, from the database using the query:

Here, we avoid objects where the photometry may be affected by a range of effects captured in the FLAGS column, by bad pixels as captured in the NIMAFLAGS column, and by a merger of child objects (NCH_MAX).

In Figure 18, we show variability amplitudes vs. the known period or variability time scale of the stars, for all matches that have over 1 mag difference between the brightest and faintest good-quality detection in DR1. This information is of course available for all six SkyMapper bands, but we show only the v- and r-band to illustrate how the variability amplitude depends on the type of variable star and its temperature evolution.

Figure 18. Variability amplitude of ~3 000 known variable stars from the AAVSO International Variable Star Index (VSX) with matches in DR1 and amplitudes in either v-band or r-band of over 1 mag. For clarity, this figure is restricted to U Geminorum stars (UG), RR Lyrae stars (RR), Cepheids (CEP), and Mira variables (M).

The combined high-amplitude list from these two bands contains 39 U Geminorum stars, 2,200 RR Lyrae stars, 55 Cepheids, and 520 Mira variables. Here, we omit further objects from rarer types and those with less characteristic time scales, such as eclipsing binaries. The variables thus end up clearly grouped by their types, which have characteristic and different time scales.

6.2. Long-term variability of QSOs

We investigated the variability of QSOs and Seyfert-1 galaxies identified by the Hamburg-ESO Survey (HES; Wisotzki et al. Reference Wisotzki, Christlieb, Bade, Beckmann, Köhler, Vanelle and Reimers2000). The public HES catalogue lists B _J magnitudes measured on photographic plates exposed over 20 yr ago, which we compare to the SkyMapper g-band magnitudes observed in 2014/5 to study long-term variability. We can obtain a list of HES QSOs and their approximate magnitude change with a simple query.

However, with a more specific aim in mind, we look for candidates of low-redshift changing-look Seyfert galaxies; given that HES only includes broad-line AGN, this search is only sensitive to transitions from Sy-1 to Sy-2, which would manifest themselves by a large drop in brightness. We first restrict the list to redshifts below 0.3 and find 133 objects with a mean B _J − g _petro difference of +0.27 mag due to Vega-to-AB and bandpass differences, and an RMS scatter of 0.51 mag.

Taking the mean difference as indicating no variability, we search for objects with a brightness decline of more than 1 mag by selecting for B _J − g _petro − 0.27 < −1 using the query:

This query lists five candidates (see Table 3), which we have all followed-up spectroscopically with the ANU 2.3-m telescope and the WiFeS spectrograph (Dopita et al. Reference Dopita2010) in late 2017. All five Sy1-galaxies show a drop in UV continuum and the brightness of their broad Hydrogen emission lines, when compared to HES spectra taken two decades earlier. Three out of the five show little or no visible broad Hβ-emission any more, and thus have changed their type from Sy-1 to Sy-2.

Table 3. Candidates of low-redshift changing-look Seyfert galaxies identified by comparing SkyMapper DR1 g-band magnitudes with B _J photometry reported by the Hamburg–ESO Survey (Wisotzki et al. Reference Wisotzki, Christlieb, Bade, Beckmann, Köhler, Vanelle and Reimers2000). Three out of five objects have changed from type-1 to (nearly) type-2 (quoted 2017 types are provisional).

6.3. Selecting metal-poor star candidates

Identifying a large sample of chemically pristine stars in the Milky Way is a prime science driver for SkyMapper and a key motivation behind the distinct u- and v-filters (e.g., Keller et al. Reference Keller2007; Bessell et al. Reference Bessell, Bloxham, Schmidt, Keller, Tisserand and Francis2011). The v-filter, in particular, was designed to be metallicity-sensitive, especially at low metallicities where the greatly decreased significance of the Calcium H and K lines translates directly to enhanced flux. The release of SkyMapper DR1 facilitates a uniform search for EMP stars across the entire Southern hemisphere; the comparatively bright limit of the Shallow Survey is ideally suited to this task as high-resolution spectroscopic follow-up of stars fainter than magnitude ~16 is prohibitively expensive even on 8–10-m class facilities.

To identify EMP candidates we utilise a two-colour plot involving the g − i colour, which serves as a proxy for effective temperature and allows us to isolate stars of the desired spectral type, and the v − g colour, which indicates metallicity. On the plane defined by these two colours, stars with [Fe/H] < −2.5 are separated from the vast bulk of more metal-rich sources by more than ~0.5 mag; for giants with T_eff ≈ 5000K, the spread between stars with [Fe/H] ≈ −2.5 and [Fe/H] ≈ −4.0 is ~0.1 mag. This selection strategy facilitated the discovery, from SkyMapper commissioning data, of the most iron-deficient star known (Keller et al. Reference Keller2014), which has [Fe/H] ⩽ −6.5 (Nordlander et al. Reference Nordlander2017). Our current DR1-based EMP survey is yielding stars at [Fe/H] ⩽ −3.0 with ~35% efficiency (see Da Costa et al. in preparation).

Before separating stars according to their colours, we select stellar objects with high-quality photometry from an area with reddening limited by E(B − V) < 0.2. We impose cuts on the g-band magnitude (<16), the photometric uncertainties in v, g, and i (0.05, 0.03, and 0.03 mag, respectively), the CLASS_STAR index (>0.9), the flags indicating any potential issues with the measurement (FLAGS <3 and NCH_MAX =1), and the location of the nearest neighbour (>7.5 arcsec), and we insist on at least one clean detection in all six filters and two in v (given its central importance). This translates into the query:

This defines a sample of 7.7 million sources, which cannot be downloaded in one step due to the current million-row limit for TAP queries. A subsequent colour analysis thins the list to ≈25000 high-quality candidates, those where the v − g colour suggests [Fe/H] ⩽ −2.5. Their spatial distribution is shown in the all-sky map of Figure 19 and exhibits the concentration towards the Galactic centre expected for a halo-like population, as well as some degree of sub-structure—although whether this is real or an artefact of residual imperfections in the photometric calibration is not yet understood (for example, note a deficiency in the number of candidates along a clear strip near α ≈ 280° and δ ≈ −65°).

Figure 19. Map of EMP candidates with DR1 photometry suggesting [Fe/H] < −2.5: the shaded grey area marks the region from which candidates were selected after applying a selection for stars with high-quality photometry (see Section 6.3). The plane of the Milky Way is indicated with a solid line, and the Galactic centre with a large open circle. Known globular clusters are marked with blue crosses.