Beijing-Arizona Sky Survey (BASS)

The Third Data Release (DR3)

  • News: including all BASS and MzLS data
  • News: new detection algorithm and PSF modelling
  • News: updated imaging processing pipeline
  • News: new astrometry based on Gaia DR2
  • This page is public, so users don't need to log in(无需注册).


The Beijing-Arizona Sky Survey (BASS) is a wide and deep imaging survey to cover a 5400 deg^2 area in the northern galactic cap with the 2.3m Bok telescope. Two filters of g and r bands are adopted. The Mosaic z-band Legacy Survey (MzLS) cover the same area in z band with the 4m Mayall telescope. These two surveys will be used for spectroscopic targeting of the Dark Energy Spectroscopic Instrument. The survey observations had been completed since the early of 2019 March. This paper describes the third data release (DR3), which contains the data from all BASS and MzLS observations during 2015 January and 2019 March. The median astrometric precision relative to Gaia positions is about 17 mas and the median photometric offset relative to the PanSTARRS1 photometry is within 5 mmag. The median 5σ magnitude depths for point sources are 24.2, 23.6, and 23.0 mag for g, r, and z bands, respectively. The whole survey area presents a high homogeneity, where the difference between the depths of 20% and 80% of the area are less than 0.3 mag. The DR3 data, including raw data, calibrated single-epoch images, single-epoch photometric catalogs, stacked images, and co-added photometric catalogs.

Data Access

Exposure numbers and quality statistics for each filter

The following figure shows spatial distributions of exposure numbers for g, r, and z bands. The regular survey footprint is outlined in cyan. The figure shows the final coverages of all BASS and MzLS observations. Except for the regular survey footprint as enclosed by dashed cyan line, there are a few other patches across the sky, representative for some test regions (e.g. COSMOS and SDSS Stripe82) and additional observations shared by other programs/observers. Most of the survey footprint is covered by three exposures for r and z bands. Those regions covered by more than three exposures are reobserved regions, where some of the three passes are not deep enough. About one third of the g-band footprint is covered by more than three exposures. It is because that the data taken in 2015 were problematic (shallower due to the imperfect ETC and noisier due to bad performance of CCD readout amplifiers) and all related regions were reobserved in the following years.

The following table presents the median observational and quality statistics for calibrated single-epoch CCD images.

Filter number of images airmass exposure times seeing arcsec sky brightness 5σ depth mag
g 101,351 1.08 103 1.58 22.18 23.50
r 79,489 1.10 102 1.43 21.13 22.99
z 243,121 1.11 90 1.02 18.78 22.45

Data Reduction

BASS and MzLS are totally open surveys. The raw data were immediately released once they were taken and transferred. These data are processed by dedicated data reduction pipelines, which provide calibrated data products. The pipelines achieve detrending all sorts of instrumental effects and obtaining astrometric and photometric solutions. The calibrated data are then fed to a photometric pipeline that generates catalog products.

BASS and MzLS data were taken using different combinations of cameras and telescopes, which have their own instrumental effects. More details can be referred to Zou et al. 2017a and Zou et al. 2018

For basic steps of data reduction Read more .

The photometric pipline

For most of the survey area, there are three dithered passes. Different passes are observed under different weather and seeing conditions. The single-epoch exposures from these passes can be combined to form deeper stacked images. Usually, a stacked image is assembled from tens of exposures taken on different nights. The PSF across the stacked image varies dramatically so that it is very hard to model accurately. The photometry in stacked images would be quite coarse. However, the PSFs across single-epoch images change smoothly and can be well modelled. The photometry in single-epoch images can be more accurate compared with the one in stacked images. Our photometric strategy is as follows: (1) sources are detected in stacked images; (2) forced photometric measurements are made in single-epoch images with prior knowledge of the object positions; (3) these measurements are co-added to form the final catalogs. The flowchart for the photometric pipeline is presented in Figure 6. The photometric software is developed in Python4 and the stacking procedure is based on SWarp software.

For details of the pipline Read more .

Quality Assessment of This Data Release

  • Completeness and false rate of source detection

Our sources are detected in the stacked images and are required to be identified in at least two bands. It naturally removes the moving objects (e.g. astroids) and transients (e.g. novae and supernovae). To check the completeness and false detection rate, we use the catalogs from the Canada-France-Hawaii Telescope Legacy Survey (CFHTLS) as the references. The CFHTLS provide two-component imaging surveys: “CFHTLS Deep” and “ CFHTLS Wide”. There are four independent deep pointings for the deep survey, giving 80% completeness limits in AB magnitudes of g = 26.0, r = 25.6, and z = 25.0 for point sources, which are about 2 magnitude deeper than the BASS and MzLS surveys. The median seeing in r band is about 0′′.7. One of the four deep fields, CFHTLS-D3, is within the BASS footprint. It is located at (α = 14:19:27, δ = +52:40:56). We use the photometric catalog for this field in CFHTLS T0007 Release (Hudelot et al. 2012). The completeness and false rate of the source detection are estimated by comparing with the sources in the deep CFHTLS-D3 catalog. As shown in the following Figure, the 80% completenesses are 23.4, 22.9, and 21.7 mag for g, r, and z bands, respectively.

The total false detection rate is about 7.4% in comparison with the CFHTLS-D3 detections. By visually checking these “false detections” on the images, we find that more than half of them are real objects. Thus we expect the false detection rate would be less than 3%. The COSMOS field was also observed by BASS, but the depths are shallower than the nominal depths. As compared with the ultra-deep COSMOS catalog Leauthaud et al. (2007), the false detection rate is about 1.8%. About 1/3 of them are real objects as checked by eyes.

  • Star-galaxy seperation

The star-galaxy separation relies on the degree of re- semblance to point-like sources. The magnitude difference of point-like and extended sources is adopted as a classifier in famous imaging legacy surveys such SDSS, Hyper Suprime-Cam Subaru Strategic Program (Aihara et al. 2018, HSC-SSP), and PS1. It can be expected that this kind of classification would perform worse at the fainter magnitude end, where distant galaxies can not be resolved so that they look like point sources. Additional information such as colors as well as morphology would definitely improve the classification accuracy. We adopt the above two ways to separate stars from galaxies.

Deeper and higher-resolution imaging surveys are needed to provide fiducial source types and assess the classification accuracy. An examination with the COSMOS classification of Leauthaud et al. (2007) shows that the star/galaxy separation in CFHTLS catalogs is not reliable enough. We adopt a similar method of Leauthaud et al. (2007) to separate stars from galaxies for CFHTLS-D3, which is based on the source distributions on the MU MAX and MAG AUTO plane.

We take the magnitude difference between the Kron and PSF magnitudes as the star/galaxy classifier for each band. The difference is close to zero for point sources and negative for extended sources. The thresholds are set to maximize the total success rate of both star and galaxy types when compared with the classification of the CFHTLS-D3 field. Objects are classified as stars in each band if magnitude differences satisfy We take the magnitude difference between the Kron and PSF magnitudes as the star/galaxy classifier for each band. The difference is close to zero for point sources and negative for extended sources. The thresholds are set to maximize the total success rate of both star and galaxy types when compared with the classification of the CFHTLS-D3 field. Objects are classified as stars in each band if magnitude differences satisfy

gKRON − gPSF > −0.101,
rKRON − rPSF > −0.093, 
zKRON − zPSF > −0.047.

The global type is determined as one of the three types in the order of z, r, and g bands. This classification method based on the magnitude difference can produce a classification even for objects observed in one single band. Figure 8 shows the classification accuracy as function of Kron magnitude relative to the classifications in the CFHTLS-D3. The accuracy for stars (galaxies) is defined as the fraction of stars (galaxies) in our classification that are also classified as stars (galaxies) in CFHTLS-D3.

The following figure shows the classification accuracy as function of Kron magnitude relative to the classifications in the CFHTLS-D3. Star-galaxy classification accuracy is based on the difference of the Kron and PSF magnitudes. The classification from CFHTLS-D3 is used as a reference.

We combine colors and morphologies to produce a more accurate classification by using a supervised machine learning algorithm. The artificial neural networks (ANN) is adopted. It needs a training set to transform the observables to source types. The training set are the objects labelled by reliable classifications in the CFHTLS-D3 field. Equal numbers of stars and galaxies are selected, which are randomly divided into training and testing sets. The training set is used to train ANN to derive a relationship between the observational information and source type. The testing set is used to assess the classification quality. The input observables include the color and morphological measurements in three bands: (1) colors of g−r, r−z, g−z; (2) ellipticity defined as the 1 − B/A, where A and B is half major and minor axis lengths, respectively; (3) KRON RADIUS/FWHM ISO indicating the size difference between extended and point-like sources, where KRON RADIUS is the Kron radius and FWHM ISO is the FWHM of the isophotal profile. The following Figure present the classification accuracy for the objects in the testing set using our trained ANN. The accuracy can be as high as 80% at the magnitude limits. Comparing with the classification based on the magnitude difference, the ANN method needs complete photometric measurements for all three bands but it produces a higher classification accuracy.

  • Astrometric and photometric accuracy

Astrometric statistics are calculated by a 3σ-clipping algorithm. The offsets and RMSs for 20%, 50% and 80% of the survey footprint and corresponding total astrometric accuracy are listed in Table 3. In general, the median astrometric offsets relative to the Gaia DR2 are less than 1 mas and the median astrometric accuracy is about 17 mas. The astrometric accuracy at the higher declination is slightly worse than that at the lower declination. It is mainly caused by the effect of differential chromatic refraction. If taking this effect into account, the astrometric accuracy would be improved to a level of < 15 mas.

We compare the co-added PSF photometry with the PSF magnitude in the PS1 catalog to check the photometric homogeneity. The bright unsaturated stars with PSF magnitudes between 16 and 20.5 mag are used and the offsets are calculated using a 3σ clipping algorithm. In addition, the PS1 magnitudes are transformed to the BASS/MzLS photometric system through the transformation equations as given in Section 6. Figure 11 displays the spatial distributions of the photometric offsets relative to the PS1 photometry. The median offsets for g, r, and z bands are within 5 mmag. The overall magnitude offsets are within 10 mmag.

The following figure shows the spatial distributions of the astrometric offsets and RMSs in the g, r, and z bands relative to the Gaia DR2 catalog positions. Spatial distributions of the astrometric offsets and RMSs are relative to the Gaia DR2 catalog positions. The regular survey footprint is outlined in cyan.

  • Photometric depths

The 5σ magnitude limits are estimated as the median PSF magnitude with photometric error at ∼ 0.21 mag. Figure 12 shows the 5σ depths for g, r, and z bands. The depths look quite homogenous over the whole footprint, except for some deeper patches covered by relatively large number of exposures.

Figure 13 shows the cumulative distributions of the magnitude limits and following table lists the depth values for different fractions of the survey area. The steep slopes of the distributions in Figure 13 indicate the survey homogeneity. The median depths are 24.2, 23.6, 23.0 mag for g, r, and z bands, respectively.

Percentage Depth(g) Depth( r) Depth(z)
20% 24.09 23.43 22.89
50% 24.23 23.56 23.03
80% 24.37 23.68 23.16

The number count of sources also reflects the imaging depth. Figure 14 shows the number counts of BASS objects in the CFHTLS-D3 field for g, r, and z bands. The count peaks are located at 23.6, 23.1, and 22.4 mag for BASS g and r and MzLS z bands, respectively. The number counts for the CFHTLS and PS1 catalogs are also plotted for comparison. The objects in the CFHTLS catalog is regarded to be complete in the magnitude range of Figure 14 . The number counts for PS1 g, r, and z bands peaks at 22.2, 22.0, 21.0 mag , respectively, which are more than 1 mag shallower than BASS.

System Transformation Equations

We provide the system transformation equations that can be used to convert the magnitudes of other wide or deep surveys to the BASS/MzLS photometric system. These transformation equations are derived using PSF magnitudes and suitable for normal stars with 0.3<g−i<2.7or0.3<g−z<4.0. Theyshouldbe also reasonable for normal galaxies without strong emission lines. The surveys include PS1, SDSS, HSC-SSP, CFHTLS, and DECaLS and corresponding transformation equations are shown as follows.


(g−i) ≡(gPS1−iPS1),
gBASS =gPS1+0.01327+0.09322(g−i)−0.01296(g−i)²−0.00123(g−i)³,
rBASS =rPS1−0.01132−0.06871(g−i)+0.02802(g−i)²−0.01032(g−i)³, 
zMzLS =zPS1+0.02294−0.13983(g−i)+0.06771(g−i)²−0.01754(g−i)³.


(g−i) ≡(gSDSS−iSDSS),
gBASS =gSDSS+0.01475-0.05067(g−i)−0.00703(g−i)²−0.00185(g−i)³,
rBASS =rSDSS−0.02990−0.09839(g−i)+0.05239(g−i)²−0.01459(g−i)³, 
zMzLS =zSDSS+0.00009−0.06104(g−i)+0.02899(g−i)²−0.00714(g−i)³.


(g−i) ≡(gHSC−iHSC),
gBASS =gHSC+0.01475-0.05067(g−i)−0.00703(g−i)²−0.00185(g−i)³,
rBASS =rHSC−0.02990−0.09839(g−i)+0.05239(g−i)²−0.01459(g−i)³, 
zMzLS =zHSC+0.00009−0.06104(g−i)+0.02899(g−i)²−0.00714(g−i)³.


(g−i) ≡(gCFHTLS−iCFHTLS),
gBASS =gCFHTLS-0.06107+0.06023(g−i)+0.02627(g−i)²−0.01317(g−i)³,
rBASS =rCFHTLS−0.09247−0.06329(g−i)+0.03335(g−i)²−0.01186(g−i)³, 
zMzLS =zCFHTLS-0.04811−0.10862(g−i)+0.05005(g−i)²−0.01204(g−i)³.


(g−i) ≡(gDECaLS−iDECaLS),
gBASS =gDECaLS-0.01478+0.07557(g−i)−0.01977(g−i)²+0.00232(g−i)³,
rBASS =rDECaLS-0.03701+0.02209(g−i)-0.00420(g−i)²+0.00074(g−i)³, 
zMzLS =zDECaLS-0.02578−0.01494(g−i)+0.00544(g−i)²−0.00103(g−i)³.
datarelease/dr3/home.txt · Last modified: 2019/08/20 11:16 by Zou Hu
CC Attribution-Share Alike 4.0 International
Powered by PHP Driven by DokuWiki Recent changes RSS feed Valid CSS Valid XHTML 1.0 Valid HTML5