Exoplanets in the Antarctic Sky. I. The First Data Release of AST3-II (CHESPA) and New Found Variables within the Southern CVZ of TESS

Wide-field photometric surveys for transit signals have proved to be one of the most effective methods for finding exoplanets. To date, more than 80% of known exoplanets have been discovered by photometric surveys using either ground- or space-based telescopes, with the Kepler project (Borucki et al. 2010) contributing the largest fraction of these transiting exoplanets. Besides just making new discoveries, photometric surveys provide key physical characteristics of exoplanets: the transit depth reveals the physical size of the transiting planet (relative to the host star), which is crucial in determining its nature (i.e., is it a gas giant or a super-Earth?), while the shape of the transit event indicates the orbit's impact parameter, constraining the orbital inclination of the planet and providing insight into the dynamical evolution of the planetary system (Seager & Mallén-Ornelas 2003). When coupled with dynamical masses determined from radial velocities of the host star, photometric measurements reveal the average density of the planet and hence constrain its composition and internal structure (see, e.g., Seager et al. 2007; Baraffe et al. 2008). Ultra-high-precision photometry can reveal secondary eclipses and phase curves. These provide a unique way to measure the reflected and thermal emission of an exoplanet, leading to estimates of its surface albedo and allowing modeling of the heat transport efficiency (see, e.g., Charbonneau et al. 2005; Deming et al. 2005; Knutson 2007). Combining all the information possible from photometry, we can establish a relatively detailed model of the exoplanet's atmosphere and hence its possible habitability.

To search for and study exoplanets using photometry, a survey needs a wide field of view (FOV), high photometric precision, and as complete time coverage of the obrial periods of interest as possible. Some pioneering ground-based projects have been very successful, e.g., WASP/SuperWASP (Pollacco et al. 2006), HATNET (Bakos et al. 2004), HATSouth (Bakos et al. 2013), and KELT (Pepper et al. 2007). Hundreds of transiting exoplanets have been found from such efforts over the last two decades. These surveys have shown that stable instruments (e.g., low systematic errors, long time stability), optimized operations (e.g., well-organized duty cycle, precise autoguiding), and superb observing conditions (e.g., good seeing, fewer interruptions from bad weather) are crucial for producing high-quality light curves and thus reducing the number of false positives in lists of transiting exoplanet candidates.

On one hand, with experience gained and new technologies adopted, new-generation wide-field transit surveys have overcome many obstacles in both hardware and software, e.g., NGTS (Wheatley et al. 2018) and Pan-Planets (Obermeier et al. 2016). On the other hand, a good observing site is still essential to guarantee a fruitful survey, and this is where the Antarctic Plateau has many advantages over traditional temperate-latitude sites. The extremely cold atmosphere in Antarctica from the telescope up leads to very low and very stable water vapor content, which reduces photometric noise from varying water absorption. The decreased high-altitude turbulence above the plateau results in dramatically reduced scintillation noise (Kenyon & Storey 2006). This combination makes the Antarctic Plateau an ideal place to perform optical, infrared, and THz observations. Lawrence et al. (2004) reported a median seeing of 023 (average of 027) above a 30 m boundary layer at Dome C, drawing worldwide attention. Subsequently, many studies have been made of the astronomical conditions at various Antarctic sites; for a summary, see Storey et al. (2005), Burton et al. (2007), and Ashley (2013). Saunders et al. (2009) studied eight major factors, such as the boundary layer thickness, cloud coverage, auroral emission, airglow, atmospheric thermal backgrounds, precipitable water vapor, telescope thermal backgrounds, and the free-atmosphere seeing, at Domes A, B, C, and F and Ridges A and B. After a systematic comparison, they concluded that Dome A, the highest point of the Antarctic Plateau, was the best site overall. Besides the excellent photometric conditions, the polar night in Antarctica provides an opportunity to stare at a target field for over 40 days without interruption from the diurnal cycle. This is a great advantage to enhance the detectability of short-period transiting exoplanets.

Since the first visit to Dome A in 2005 by the 21st CHInese National Antarctic Research Expedition (CHINARE), Chinese astronomers and their international collaborators have performed a series of site-testing studies. The results from multiple experiments have shown that Dome A has a thin boundary layer with a 14 m median height (Bonner et al. 2010), a strong temperature inversion above the snow surface, low wind speed (Hu et al. 2014), low water vapor (Shi et al. 2016), low sky brightness, and a high clear-sky fraction (Zou et al. 2010; Yang et al. 2017). These results are consistent with previous predictions made largely from satellite observations by Saunders et al. (2009). The first-generation telescope at Dome A (the Chinese Small Telescope ARray (CSTAR); Yuan et al. 2008; Zhou et al. 2010) was installed in 2008. The CSTAR produced more than 200,000 continuous images on a fixed FOV of 45 × 45 centered at the south celestial pole in the austral winters of 2008, 2009, and 2010. After a series of refinements (such as corrections for the inhomogeneous effects of clouds and the treatment of ghost images and diurnal effects; Wang et al. 2012, 2014b; Meng et al. 2013), a photometric precision of 4 mmag was obtained for bright stars in CSTAR data. Hundreds of new variable stars were identified and studied from these data (e.g., Wang et al. 2011; Yang et al. 2015; Zong et al. 2015; Liang et al. 2016; Oelkers et al. 2016), and the first six exoplanet candidates around the south celestial pole were identified (Wang et al. 2014a). Based on this successful experience and lessons learned from the CSTAR project, a second generation of telescopes—the Antarctic Survey Telescopes (AST3)—were conceived for a wide-field, high-resolution photometric survey at Dome A. The AST3 was envisaged as three 50 cm aperture telescopes, each with a different fixed filter. The AST3 telescopes not only have larger apertures than CSTAR, they also have full tracking systems and higher angular resolutions (∼1'' pixel⁻¹), which provide wider sky coverage and better precision. The first and second AST3 telescopes—AST3-I and AST3-II—were installed at Dome A in 2012 and 2015 by the 28th and 31st CHINARE (respectively). Observing programs with AST3-I have been described by Liu et al. (2018) and the corresponding data published by Ma et al. (2018). The third AST3 telescope is nearing completion in Nanjing and will be equipped with a Kdark-band near-infrared camera (Burton et al. 2016). Using the CSTAR and AST3 telescopes, we have started a long-term wide-field photometric survey searching for transiting exoplanets in the Antarctic sky: the Chinese Exoplanet Searching Program from Antarctica (CHESPA).

In the austral winters of 2016 and 2017, we used the AST3-II telescope to survey a group of selected fields near the southern ecliptic pole and within the southern continuous viewing zone (CVZ) of TESS (Ricker et al. 2009). In this first paper, we describe the observations obtained and data reduction processes used on them, as well as presenting some of the data products obtained in 2016. This data release includes reduced images, calibrated catalogs, and detrended light curves of 26,578 bright (7.5 ≤  m_i ≤ 15.0) stars near the southern ecliptic pole. We also present a catalog of variable stars found in the southern CVZ of TESS. Since wide-field surveys for transiting exoplanets often suffer from large pixel scales and are notoriously plagued by false positives, high-resolution photometric follow-up observations are necessary to filter out objects such as eclipsing binaries. The precision of TESS is expected to be much better than that of ground-based surveys; however, its pixel scale is still relatively large, ∼21'' pixel⁻¹, and a high false-alarm rate caused by blending events such as background eclipsing binaries is expected (Collins et al. 2018). Our catalog of variables will be a good reference for identifying these blending events. In this work, we only present obvious binaries and pulsating stars with regular periods. A detailed classification and analysis of other variables, e.g., rotational spots modulation, semiperiodic/irregular variables, and long-period variables, will be available soon in a series of subsequent papers. The description of the light curve detrending, transit signal search, and a detailed catalog of transiting exoplanet candidates will be presented by Zhang et al. (2018). Results for observations of specific targets, e.g., Proxima Centauri and β Pictoris, will also be released in the near future.

This paper is organized as follows. We introduce the instruments and our survey strategy in Section 2, we describe the survey strategy and observations in Section 3, the detailed data reduction flow is described in Section 4, we present the survey results in Section 5, and we summarize the paper in Section 6.

The AST3-I telescope was designed and built by the Nanjing Institute of Astronomical Optics and Technology (NIAOT); see Cui et al. (2008), Yuan et al. (2014, 2015), and Wang et al. (2017) and references therein for more details. The AST3-II telescope (which was used to acquire all the data used in this work) is the second of the three planned AST3 telescopes. It is almost identical to AST3-I: they share the same modified Schmidt system design (Yuan & Su 2012), entrance pupil aperture of 50 cm, wide FOV of 4.3 deg², Sloan i filter, and CCD camera. Some improvements and innovations have been made in AST3-II in the area of focusing, snowproofing, and the defrosting system based on our experience from the earlier operations of AST3-I during austral winters since 2012. Thanks to these updates, AST3-II worked well during the extremely cold winters of 2016 and 2017 and has acquired over 30 TB of high-quality images. A single raw AST3-II image is 10K × 5K pixels in size, using 122 MB of storage. It is generated by a 10K × 10K frame transfer STA1600FT CCD camera with a pixel scale of ∼1'' pixel⁻¹ over an FOV of 1.5 × 2.9 deg² (R.A. × decl.). The average FWHM of stars obtained by focused AST3 telescopes is around ∼2.0 pixels. The CCD detector operates in frame transfer mode without a shutter and is divided into frame store regions (in its top and bottom quadrants) and an active imaging area in its central half. Although this configuration halved the maximum possible FOV, it eliminates the risk of shutter mechanism failure. Since the telescope has to be operated entirely remotely for 11 months with no possibility of repairs being carried out, any mechanism failure would be fatal to the whole project, and the shutter is one of the most fragile parts of these high-frequency observations. The image area of the CCD has 16 readout channels, each 1500 × 2650 pixels (including overscan regions). A full-frame raw image is shown in Figure 1, and more details on the AST3 CCD performance can be found in Ma et al. (2012) and Shang et al. (2012).

Zoom In Zoom Out Reset image size

During the austral winter observing season, the whole AST3 system is operated remotely, and the scheduled observations are executed in a fully automatic mode. Consequently, safety, reliability, and stability are key issues for the AST3 design. The core systems, consisting of the main control computer, storage disk array, and pipeline computer, were intensively customized to handle almost any conceivable hardware, software, or network failure. Each component in the system had two identical copies for redundancy, minimizing the risk to the system from single-point failures. As an example, the fiber optic data link from the CCD camera was split into two fibers so that two separate computers could be used to control the CCD. The hardware and software for the operation, control, and data (COD) system were developed by the National Astronomical Observatories, Chinese Academy of Sciences (NAOC; Shang et al. 2012; Hu et al. 2016). The electrical power supply and internet communication were provided by a similarly reliable on-site observatory platform, PLATO-A, which is an improved version of UNSW's automated PLATO observatory platform for CSTAR and other earlier instruments. PLATO-A was designed to provide a continuous 1 kW power source for the AST3 telescopes (Lawrence et al. 2009; Ashley et al. 2010) for at least 1 yr without servicing.

The Kepler project has dramatically broadened our horizons in exoplanet science over the last 5 yr. Although it has surveyed only a small part of the sky, its unprecedented photometric precision and continuous observing capability have been superbly successful at discovering planets down to Earth sizes. However, most candidates found by Kepler are just too faint to be followed up with dynamical mass measurements from the ground. Kepler's successor mission, TESS, will be similarly revolutionary, albeit in a different way—by surveying the whole sky for short-period transiting exoplanets orbiting bright and/or nearby stars that can be followed up with dynamical masses from the ground. TESS launched on 2018 April 18 and started surveying all bright stars in the southern hemisphere in 2018 June. Thousands of exoplanet candidates are expected to be detected over the next 2 yr. However, to achieve whole-sky coverage, TESS must use a large pixel scale (∼21'' pixel⁻¹), which may lead to a higher false-alarm rate than Kepler. Many candidates detected by TESS may be astrophysical mimics, e.g., blended eclipsing binaries. Therefore, high-resolution photometric follow-up observations are necessary before undertaking time-consuming and competitive radial velocity observations to measure dynamical masses. The AST3 telescopes were designed to perform wide-field (FOV ∼ 4.3 deg²) and high-resolution (pixel scale ∼1'' pixel⁻¹) time-domain photometric surveys and are well suited to this task. The observational conditions and uninterrupted polar nights at Dome A deliver additional benefits for this science.

To maximize the scientific value of our program, we select 48 fields within TESS's southern CVZ (which is centered at the southern ecliptic pole at R.A. = 06^h00^m00^s, decl. = −66°33'00''). These fields are scientifically important because (in addition to their southerly decl., meaning they can be observed from Dome A at low airmasses) they will be observed by TESS uninterrupted for 12 months. This long observing baseline will greatly increase TESS's sensitivity for short-period exoplanets, as well as allowing the detection of long-period planets in single-star and binary systems (Stassun et al. 2018). Additionally, the CVZ of JWST will also be located in this region, so any targets of interest will be studied by much more intensive detailed follow-up in the future.

Our selected fields were numbered and divided into three groups, with each group being scheduled to be surveyed for a whole austral winter in 2016, 2017, and 2018, respectively (see Figure 2). Fields close to the Large Magellanic Cloud (LMC) were excluded because they are too crowded and most bright stars in the LMC will be giants, making transiting exoplanet signals exceedingly weak. From 2016 May 16 to June 22, we scanned the first group of 10 adjacent fields (AST3II004—AST3II013; see Table 1). The available dark time increased day by day during the first half of our observation campaign, and it finally became 24 hr a day at the end of June. A part of the very dark nights were spent on the other programs searching for transient targets, e.g., supernovae. After excluding interruptions caused by bad weather and regular instrument maintenance, we observed for more than 350 hr spanning 37 nights (Figure 3). The overall operation coverage is ∼40%, which is less than the coverage we expected: ∼75%. In 2017, we spent most of the nights where weather and instrumentation were suitable on CHESPA, and its working coverage reached ∼80%. To avoid saturation by bright stars and to reduce the electromagnetic interference in our CCD images (see Section 4.1.4), we adopted a short-exposure stacking strategy. In each observing night, we started from field AST3II004 and took three consecutive 10 s exposures, then moved to field AST3II005 and repeated, then moved to AST3II006, and so on. When the last field—AST3II013—completed, we jumped back to the first field—AST3II004—and started a new loop. The dead time between each field was about 72 s, including three 16 s readout intervals and a 24 s slewing operation. Every three adjacent images of the same field were median combined to achieve a high signal-to-noise ratio (S/N). Thus, the overall sampling cadence was about 12 minutes for each field.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Twilight-sky frames were taken at each dawn and dusk during the periods when the Sun was still rising each day at Dome A. These sky frames were combined to produce a master flat-field image (see Section 4.1.3). However, fluctuations in the super-sky flat field are a major source of systematic noise for wide-field surveys. To reduce this, we first adopted a moderate defocusing of the telescope. The average seeing on the ground of Dome A is estimated to be around ∼1'' or better. Convolved with the optics of a telescope, the actual size of the point-spread functions (PSFs) of stars is usually larger than the seeing. So the AST3-II pixel scale of 1'' pixel⁻¹ is designed to be better than the average FWHM of stars acquired by the focused AST3-II at Dome A, which is around ∼2.0 pixels. During the exoplanet survey, we adjusted the focus to produce star images with PSFs between FWHM = 3 pixels and FWHM = 5 pixels. The shapes and sizes of the star images vary with the environment temperature, airmass, and etc. Sometimes, the PSF deviates from the standard Gaussian profile and forms a flat top. But it does not cause much of a problem, since we adopt aperture photometry. For each star, the variations in the FWHM, elongation,²⁴ and fraction-of-light radius²⁵ are recorded. In the later external parameter detrending (EPD) process, these variations will be detrended from the light curve. Second, a semi-autoguiding method was adopted. For each field, a template image with a well-determined astrometric solution is constructed before the survey begins. Every time the telescope points to a new field, the first image of this field is cross-matched with its template, and offsets are fed back to the pointing model and the two following images acquired after this pointing correction. The interval between two pointing corrections is the same as the dead time between moving fields, i.e., every ∼72 s. Within this short time, the position variation of a star is still less than the average size of a PSF, FWHM ∼ 4 pixels. Even so, it is still very difficult to fix stars to the same pixels during the entire observation campaign. Finally, we detrend the brightness variation of each star against its inter- and intrapixel position changes and variances to further reduce any systematic error caused by position shifting of stars.

The telescopes working at Dome A are entirely unattended during the year, apart from about 3 weeks over summer when servicing is performed on-site. The internet connection to Dome A relies on the Iridium satellite system, which has a typical latency of about 2 s and can occasionally drop out and take up to 1 minute to reconnect. The operation of the telescope and CCD is therefore controlled automatically from scripts running on-site at Dome A. The whole system can be shifted to manual control from China if required (e.g., if the telescope optics and/or gear mechanisms need defrosting). Images taken by AST3-II are stored on a disk array on-site. The bandwidth of the Iridium connection makes it impractical to download even a single whole image, so we only perform relatively simple data reduction processes on-site—e.g., to check data quality and set up alarms for specific events. The detailed reduction of the entire data set waits until the hard drives are physically returned by the expedition team in April or May of the following year.

At the end of the exoplanet survey program in 2016, more than 35,000 science frames were taken by the AST3-II telescope. Some statistics on the data quality are shown in Figure 4. We removed images with flaws including high sky backgrounds, a large fraction of saturated area, large average FWHM (>6 pixels), large average elongations (>2), and small numbers of detectable stars (<500). After this quality filtering, 30,889 high-quality images remained. They were further reduced and coadded to produce 18,729 science images. Catalogs of these images were matched with a master catalog generated from the APASS database (Henden et al. 2015, 2016). Finally, light curves for 26,578 stars brighter than Sloan  m_i = 15 mag were extracted.

Zoom In Zoom Out Reset image size

The on-site power budget and limited computational resources mean we only run a simple on-site data reduction pipeline including only standard image processes, such as overscan subtraction, flat-field correction, and simple-aperture photometry using Sextractor (Bertin & Arnouts 1996). Reduced catalogs are then downloaded through the Iridium satellite connection while observing. However, the limited connection bandwidth and monthly quota mean only the catalogs of selected fields are downloaded. We perform further astrometric solutions and flux calibration on these catalogs and extract light curves of selected targets. Although the precision of this on-site data reduction is not optimal, it does return rms ∼1% light curves, which are useful for inspecting the daily data quality and triggering alarms of highly valuable targets (e.g., potential transit events of β Pictoris and Proxima Centauri). As described above, full data processing awaits the return of the data on hard drives from Dome A.

A detailed data reduction flowchart for this full processing is shown in Figure 5. The pipeline consists of five major components, which are described below: (1) image reduction, which produces cleaned, WCS solved, and coadded science images; (2) catalog processing, which produces flux-calibrated catalogs; (3) light-curve detrending, which produces light curves for general scientific usage; (4) periodic signal searching, which searches for periodic variables and generates the variables catalog that is published in this work; and (5) transit signal searching, which polishes the light curves further, searches for transit-like signals, and validates exoplanet candidates. The pipeline operates in the MATLAB environment and has been fully parallelized. Some functions from well-tested open-source packages are used, including Sextractor (Bertin & Arnouts 1996), Swarp (Bertin et al. 2002), VARTOOLS (Hartman & Bakos 2016), and Astrometry.net (Lang et al. 2010).

Zoom In Zoom Out Reset image size

4.1. Image Reduction

The image reduction component includes some standard steps (e.g., overscan subtraction, dark-current subtraction, and flat-field correction), as well as some special steps to correct specific features within our data (e.g., cross-talk effects and electromagnetic fringes).

4.1.1. Overscan Subtraction

As shown in Figure 1, the 10K × 10K STA1600FT CCD camera installed on AST3-II works in a frame transfer mode and produces an image size of 12,000 × 5300, including overscan regions. At both the top and the bottom boundaries, there are 60 rows that are insensitive to light; these are trimmed off before further reductions. The rest of the image is divided into 16 readout channels, with eight at the top and eight at the bottom. The vertical readout direction of the channels at the top or bottom of an image is from the center to the top or bottom, respectively. The horizontal readout direction is always from left to right within each channel. Within a readout channel, the first 10 rows and the last 180 columns compose a reverse "L"-shaped overscan region surrounding the photosensitive area, which has a size of 1320 × 2580 pixels. Instead of simple bias subtraction, we perform a 2D overscan subtraction for each channel: horizontal subtraction first and then vertical subtraction. Taking a top channel as an example, its horizontal overscan region is located at its bottom. This region, which is composed of 10 rows and 1500 columns, is median combined to a single row, and this combined row is fitted using a second-order polynomial function and subtracted from the other 1320 rows (including the vertical overscan region). Then, the vertical overscan region at the right, which is now 1320 rows by 180 columns, is median combined, second-order polynomial fitted, and subtracted from each column of the final photosensitive area. The channels at the bottom are then processed in a vertically flipped direction. After the 2D overscan subtraction, the photosensitive areas of all the channels are fitted together to form a new image (see Figure 6).

Zoom In Zoom Out Reset image size

4.1.2. Cross-talk Correction

In the overscan-subtracted image, serious dark spots can be seen at the same positions along the readout direction in each channel (see Figure 6). These are the cross-talk effects caused by saturated stars in other channels. When one amplifier is reading out a saturated pixel, the video signal can contaminate the pixel streams that are being read out simultaneously from the other amplifiers. In the AST3-II images, the cross-talk effects appear as negative ghosts of their saturated source stars. The depth of such a ghost is several tens of ADUs below the sky background and varies from channel to channel. Since there are 16 readout amplifiers in our CCD camera, these cross-talk effects are significant. Each saturated star results in 15 negative ghosts in the other 15 readout channels, and any real stars at the positions of these ghosts are made to appear artificially faint. A detailed discussion of the causes of this cross-talk is beyond the scope of this work. Here we only describe how we eliminate its influence from our data. The basic idea can be summarized in the following steps.

• 1.  
  Mark all saturated pixels (by setting the value of unsaturated pixels as 0) in every readout channel and combine all channels into a single template channel (flipping the top channels vertically) that reflects the positions of all of the saturated stars/pixels in the entire image.
• 2.  
  Redistribute this template channel into all 16 channels (flipping the top channels vertically) to produce a copy of the original image showing the locations of all of the ghosts.
• 3.  
  Find the difference between the bottom of each ghost spot and the local sky background, then add this back to the pixels inside the spot. Repeat this process for every readout channel.

However, there are some complicating factors. First, some ghosts may overlap with each other, and the overlapped region is darkened twice. Second, a saturated star always generates a halo that also causes cross-talk, even if the halo itself is not saturated. Thus, the corresponding ghost may have a halo, and this halo region may escape from the very first marking step (which marks only saturated pixels). As a result, every ghost has its own structure, and the depth within a ghost is not constant. Finally, the gains of amplifiers are different from channel to channel. So the depths of the ghosts caused by the same saturated star are different in different channels. The depths of the ghosts within the same channel but caused by sources in different channels are different as well. To solve these problems, we decompose each ghost into three parts: the core region, which is caused by the saturated star itself; the halo region, which is caused by charge bleed from the saturated star; and the overlapped region of two or more saturated stars. The pixels within the halo region are not saturated but contain extra charge due to charge-bleed issues caused by the neighboring saturated pixels. The template channel is also decomposed into three corresponding layers (see the captions of panels (a)–(d) in Figure 7 for more details.). Within each layer, we identify all connected pixels and group them as a "subghost." This is similar to identifying stars within an image when performing photometry. We then redistribute each layer into all 16 channels to show the locations of all "subghosts" in the original image. Finally, we examine every identified "subghost," fit its median depth, and add it back to the pixels affected by this "subghost" in the original image. This process is repeated for all readout channels. Figure 8 shows the result of our cross-talk correction. Darkening caused by the cross-talk effects is reduced and less than the variation of the local sky background in the corrected image.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

4.1.3. Flat-field Correction

During twilight periods before the polar winter in 2016, hundreds of twilight-sky images were taken. These twilight images are combined to produce the master flat field. Owing to the relatively large FOV of 4.3 deg², a sky brightness gradient of ∼1%–10% remains in individual twilight images after preprocessing for cross-talk, overscan, and dark current (for a detailed treatment of dark current, see Ma et al. 2014). Two hundred twilight images were selected to model this sky brightness gradient. For each of the selected images, the brightness gradient was first fitted with an empirical function based on the Sun's altitude and position angle. Then, the fitted gradient was divided from each twilight image, and the resulting 200 twilight images were median combined to produce a master flat field. The final flat-field error is well below 1%. More details of the brightness gradient model and uncertainties in the flat field can be found in Wei et al. (2014).

4.1.4. Electromagnetic Interference Correction

Another feature of our CCD images is electromagnetic interference. Straight and inclined fringes can be easily identified in each channel (see panel (a) in Figure 9). The inclined angles are different in the top and bottom halves of each image; thus, a series of "<"-shaped fringes are formed and spread across the entire image. Pixels affected by these fringes will be brightened or darkened by about 20–50 ADU, which has to be corrected before we can achieve millimagnitude precision. These interference fringes are thought to be caused by noise from the camera's power supply being electrically and/or magnetically coupled into the CCD readout signal chain (though this has not yet be confirmed). The best way to remove this noise would be at the source by improving the grounding and shielding of the entire power supply system. In the absence of such a solution, we model the positions and influences of each noise fringe and correct them in software. However, this has turned out not to be easy. While the pattern of the fringes remains very similar, they move across the image with time. So, the affected pixels and impact on them vary from frame to frame. Even within a single fringe, only a fraction of the pixels are affected, and the amplitudes of the affected pixels also vary. It is very difficult to identify the affected pixels, especially those that are also overlapped by stars. Our adopted solution is not yet ideal, but it is simple. We take three 10 s exposures for the same field, resample, and match the last two frames to the first one to guarantee pixel alignments. Finally, all three pixel-aligned images are median combined by the Swarp code (Bertin et al. 2002) to produce a new image with lower background variation (Zackay & Ofek 2017). Since the fringe pattern is moving, a pixel is usually affected in only a single frame of the three. Therefore, the median combination may have a good chance to reject the polluted value as an outlier and reveal the real sky background. This simple process helps to reduce the influence of the fringe noise to an acceptable level. A comparison of images before and after stacking is presented in Figure 9.

Zoom In Zoom Out Reset image size

4.2. Photometry and Calibration

The cleaned images from the preceding steps are then registered and coadded by Astrometry.net (Lang et al. 2010) and Swarp (Bertin et al. 2002), respectively. As we mentioned above, the cadence of stacked images is about 12 minutes with an effective exposure time of ∼30 s. These images are labeled as "scientific images" and are a major component of our data products. Aperture photometry is then performed on these scientific images, and star catalogs are generated using Sextractor (Bertin & Arnouts 1996). For each image, the total photometry process is divided into two steps. First, we output a simple catalog that contains only the positions and fluxes using optimal apertures (corresponding to the keyword FLUX_AUTO). This simple catalog is then cross-matched with the AAVSO Photometric All-Sky Survey (APASS) catalog (which is conducted in five filters: the Johnson B and V and Sloan g, r, and i bands; Henden et al. 2015). The i-band magnitudes and fluxes of all matched stars are extracted from the APASS and AST3 catalogs, respectively. The zero-point magnitude is then fitted from the power-law relation of magnitude and flux (Figure 10). Second, we adopt the fitted zero-point magnitude and run Sextractor again to produce the final calibrated catalog. Although we use a Sloan i-band filter, there may be some systematic offsets or color dependences between our results and that from the APASS database. The major sources may include the manufacturing difference between our filter and the one adopted by APASS, the absence of a strong water feature in the i band due to the extremely low water vapor in Antarctica, and the differences in the magnitude cutoff and colors of the different ensembles of reference stars. To demonstrate these issues, we fit a simple relation, ${\rm{\Delta }}{{\boldsymbol{m}}}_{i}={{\boldsymbol{m}}}_{i,\mathrm{ast}3}-{{\boldsymbol{m}}}_{i,\mathrm{apass}}=c{({{\boldsymbol{m}}}_{r}-{{\boldsymbol{m}}}_{i})}_{\mathrm{apass}}+b$, in each field to show the color dependence of our magnitude calibration. Here $c\equiv d{\rm{\Delta }}{{\boldsymbol{m}}}_{i}/d({{\boldsymbol{m}}}_{r}-{{\boldsymbol{m}}}_{i})$ is the slope of the color dependence and b is a magnitude offset. Results from our 10 target fields are shown in Figure 11. The fitted slope and uncertainty of the color dependence in each field is also labeled in each panel. We find a median value of the slope, c_median = −0.0771, among all 10 target fields. The slopes of the color term are close to zero in all target fields, which indicates that the color issue is insignificant within our result. However, it is still worth noting that this color term has not been corrected in this data release, and it will be considered in future data releases.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In the calibrated catalog, we output image positions, J2000.0 equatorial coordinates and their variances, photometry results of optimal aperture and three fixed apertures (8, 10, and 12 pixels, which are, according to our plate scale, 8'', 10'', and 12'', respectively), parameters listing observation conditions, and parameters from image statistics. All of these pieces of information are inherited by the light curves and used as external trend parameters in the following light-curve detrending process. An example of a catalog header is shown in Table 2.

Table 2.  An Example Catalog Header of an Image Taken on "2016–05–03T16:34:44.34."

Name	Example/Description	Units
Observation start time	2016–05–03T16:34:44.34	UT
Exposure time	30	s
CCD temperature	217.92	K
Airmass	1.07	mag
Solar altitude	−23.87	deg
Moon distance	90.82	deg
Zero-point mag	21.93	mag
# 1 XWIN_IMAGEa	Image position in x axis	pixel
# 2 YWIN_IMAGE	Image position in y axis	pixel
# 3 X2WIN_IMAGE	Position variance in x axis	pixel2
# 4 Y2WIN_IMAGE	Position variance in y axis	pixel2
# 5 XYWIN_IMAGE	Position covariance between x and y axis	pixel2
# 6 ERRX2WIN_IMAGE	Error of position variance in x axis	pixel2
# 7 ERRY2WIN_IMAGE	Error of position variance in x axis	pixel2
# 8 ERRXYWIN_IMAGE	Error of position covariance between x and y axis	pixel2
# 9 CXXWIN_IMAGEb	First ellipse parameter of object's isophotal shape	pixel−2
# 10 CYYWIN_IMAGE	Second ellipse parameter of object's isophotal shape	pixel−2
# 11 CXYWIN_IMAGE	Third ellipse parameter of object's isophotal shape	pixel−2
# 12 XWIN_WORLD	R.A. in J2000.0	deg
# 13 YWIN_WORLD	Decl. in J2000.0	deg
# 14 ERRX2WIN_WORLD	Error of position variance in R.A.	deg2
# 15 ERRY2WIN_WORLD	Error of position variance in decl.	deg2
# 16 ERRXYWIN_WORLD	Error of position covariance between R.A. and decl.	deg2
# 17 MAG_AUTO	Kron-like elliptical aperture magnitude	mag
# 18 MAGERR_AUTO	rms error for AUTO magnitude	mag
# 19 FLUX_AUTO	Flux within a Kron-like elliptical aperture	count
# 20 FLUXERR_AUTO	rms error for AUTO flux	count
# 21 MAG_APER	Fixed-aperture magnitude vector	mag
# 24 MAGERR_APER	rms error vector for fixed-aperture mag	mag
# 27 FLUX_APER	Flux vector within fixed circular aperture(s)	count
# 30 FLUXERR_APER	rms error vector for aperture flux(es)	count
# 33 FWHM_IMAGE	FWHM assuming a Gaussian core	pixel
# 34 ELONGATION	Ellipticity of object's isophotal shape	⋯
# 35 BACKGROUND	Background at centroid position	count
# 36 THRESHOLD	Detection threshold above background	count
# 37 FLUX_RADIUS	Fraction-of-light radius	pixel
# 38 FLAGSc	Extraction flags	⋯

Notes. Position- and flux-related measurements (keywords) are generated by Sextractor (Bertin & Arnouts 1996).

^aAll position-related measurements, e.g., XWIN_IMAGE, XWIN_WORLD, etc., are performed through a Gaussian window function, i.e., the centroid coordinates are calculated through an iterative process updated by a Gaussian-weighted integration of the coordinates of isophotal pixels. The start point is set to be the first-order moment of the source. More details can be found in Bertin & Arnouts (1996). All "Windowed" measurements are denoted with a "WIN," "_IMAGE" means in pixel coordinates, and "_WORLD" means in equatorial coordinates. ^bCXXWIN_IMAGE, CYYWIN_IMAGE, and CXYWIN_IMAGE are three ellipse parameters describing the shape of an object's isophotal ellipse. $\mathrm{CXXWIN}\_\mathrm{IMAGE}{(x-\bar{x})}^{2}$ + CYYWIN_IMAGE ${\left(y-\bar{y}\right)}^{2}$ + CXYWIN_IMAGE $(x-\bar{x})(y-\bar{y})={R}^{2}$, where x and y are XWIN_IMAGE and YWIN_IMAGE, and $\bar{x}$ and $\bar{y}$ are mean values of XWIN_IMAGE and YWIN_IMAGE, respectively. The last constant R = 3. ^cFLAGS contains, coded in decimal, all the extraction flags as a sum of powers of 2. 1: The object has neighbors bright and close enough to significantly bias the MAG_AUTO photometry or bad pixels (more than 10% of the integrated area affected). 2: The object was originally blended with another one. 4: At least one pixel of the object is saturated (or very close to). 8: The object is truncated (too close to an image boundary). 16: Object aperture data are incomplete or corrupted. For example, an object close to an image border may have FLAGS = 16 and perhaps FLAGS = 2 + 4 + 8 + 16 = 30.

Download table as:  ASCIITypeset image

4.3. Light-curve Detrending

The general concept behind our light-curve "polishing" process is to perform blind detrending that removes the common variations among all targets. Two such methods have been commonly employed in data reduction for wide-field photometric surveys: systematic effects remove (SYSREM; Tamuz et al. 2005) and trend-filtering algorithm (TFA; Kovács et al. 2005). We implemented a TFA-like algorithm with some minor adjustments. First, the reference stars of a target light curve were selected according to their distances to the target star (e.g., they must be in the same readout channel as the target star). Second, we built an individual trend matrix (whose columns are magnitudes of reference stars and external parameters) for each target light curve and updated it dynamically according to the correlation coefficients between the brightness variations of the target and its reference stars. Those stars with large variations were then clipped from the reference star list after iteration. Third, the external parameters of the target star, e.g., its airmass, distance to the moon, FWHM, local background variation, etc. (see Table 3 for details), were recorded simultaneously with the brightness measurements and are now appended to its trend matrix as additional columns. The variations present in these external parameters will be used to detrend the target light curve. This is usually called the EPD process. Here we combine the EPD- and TFA-like processes together and detrend the target light curve using its external parameters and brightness of reference stars simultaneously. Finally, instead of the least-squares fit to the linear combination of the reference light curves in the standard version of TFA, the detrending process in our pipeline is achieved by a multivariable linear regression with each column in the trend matrix marked as a "variable." We noticed that some outliers, caused by mismatches of stars or bad weather, caused serious problems in the detrending processes, and occasionally they may crash the box least-squares (BLS) fitting searching function that follows. To eliminate these outliers and retain large physical variations at the same time, we performed a Gaussian regression to model the time-dependent variations within each light curve, and we then clipped all measurements that are more than 3σ away from the model before using them in the detrending process.

Table 3.  Columns in Raw Light Curves

Column No.	Name	Explanation	Units
1	Time	AST3 observation time in UT (JD—2,456,000)	day
2	XWIN_IMAGE	Image position in x axis	pixel
3	YWIN_IMAGE	Image position in y axis	pixel
4	X2WIN_IMAGE	Position variance in x axis	pixel2
5	Y2WIN_IMAGE	Position variance in y axis	pixel2
6	XYWIN_IMAGE	Position covariance between x and y axis	pixel2
7	ERRX2WIN_IMAGE	Error of position variance in x axis	pixel2
8	ERRY2WIN_IMAGE	Error of position variance in y axis	pixel2
9	ERRXYWIN_IMAGE	Error of position covariance between x and y axis	pixel2
10	CXXWIN_IMAGE	First ellipse parameter of object's isophotal shape	pixel−2
11	CYYWIN_IMAGE	Second ellipse parameter of object's isophotal shape	pixel−2
12	CXYWIN_IMAGE	Third ellipse parameter of object's isophotal shape	pixel−2
13	XWIN_WORLD	R.A. in J2000.0	deg
14	YWIN_WORLD	decl. in J2000.0	deg
15	ERRX2WIN_WORLD	Error of position variance in R.A.	deg2
16	ERRY2WIN_WORLD	Error of position variance in decl.	deg2
17	ERRXYWIN_WORLD	Error of position covariance between R.A. and decl.	deg2
18	MAG_AUTO	Kron-like elliptical aperture magnitude	mag
19	MAGERR_AUTO	rms error for AUTO magnitude	mag
20	FLUX_AUTO	Flux within a Kron-like elliptical aperture	count
21	FLUXERR_AUTO	rms error for AUTO flux	count
22–24	MAG_APER	Fixed-aperture magnitude vector	mag
25–27	MAGERR_APER	rms error vector for fixed-aperture mag.	mag
28–30	FLUX_APER	Flux vector within fixed circular aperture(s)	count
31–33	FLUXERR_APER	rms error vector for aperture flux(es)	count
34	FWHM_IMAGE	FWHM assuming a Gaussian core	pixel
35	ELONGATION	Ellipticity of object's isophotal shape	⋯
36	BACKGROUND	Background at centroid position	count
37	THRESHOLD	Detection threshold above background	count
38	FLUX_RADIUS	Fraction-of-light radius	pixel
39	FLAGS	Extraction flags	⋯
40	exposure_time	Exposure time	s
41	CCD_temperature	Temperature of CCD chip	K
42	airmass	Airmass	⋯
43	Sun_altitude	Altitude of the Sun	deg
44	moon_distance	Angular distance to the Moon	deg
45	zeropoint	Zero-point magnitude	mag

Note.

Keywords have the same meaning as in Table 2.

Download table as:  ASCIITypeset image

In this data release, we present two versions of light curves for general science usage: the raw light curves that contain all the original information (see Table 3) and the detrended light curves that contain only time, magnitude, and the error in the magnitude. The cadence of these light curves is ∼12 minutes, and the best rms after detrending is ∼4 mmag at the bright end, ${{\boldsymbol{m}}}_{i}=10$. This precision is good enough to search binaries or pulsating variables; however, to search for transiting exoplanets, we need higher precision. To achieve this, the light curves are then further "polished" and binned to 36 minutes, after which they achieve a best rms of 2 mmag at ${{\boldsymbol{m}}}_{i}=10$ (see Figure 12). The curves of the expected photon-limited rms against magnitude are also overplotted in Figure 12. Note that when we calculate the photon noise, the sky background noise, readout noise, saturation limit, and average number of measurements falling in the binned intervals are also considered. As a result, the photon limit of light curves with a 36 minute cadence is not simply improved by a factor of $\sqrt{3}$ but rather a factor of $\sim \sqrt{2}$, to that of a 12 minute cadence. The rms of the binned light curves is improved a lot at the bright end; however, there are still hidden systematic errors that prevent us from reaching the photon-limited precision. The detrending process and further polishing determine the final precision of our light curves and directly influence the detection rate of transit signals. The detailed algorithm and parameter settings, e.g., the number of reference stars and selection of external parameters, will be presented in a following paper (Zhang et al. 2018) that concentrateson transiting exoplanet searching.

Zoom In Zoom Out Reset image size

4.4. Periodic Signal Searching

The "L" variability index (Welch & Stetson 1993; Stetson 1996) is used to select potential variable candidates. For each light curve, we calculated its L index using the −Jstet command of VARTOOLS (Hartman & Bakos 2016). The overall distribution of L index in our sample is shown in Figure 13. Two components can be fitted from this distribution, one of which shows a Gaussian profile and corresponds to stars with insignificant variation and the other of which shows an exponential tail and is likely to correspond to variable stars. The intersection of these two profiles is at L ≈ 0.19. Stars with L ≥ 0.19 are selected as variable-star candidates and delivered to the "periodic signal-searching" procedure. Periodic and sinusoidal variations are then identified using both the generalized Lomb–Scargle (LS; Press et al. 1992; Zechmeister & Kürster 2009) and the analysis of variance (AoV) (Schwarzenberg-Czerny 1989; Devor 2005) methods. In practice, we integrate the −LS and −aov commands of VARTOOLS into our pipeline. This module produces two data sets from the LS and AoV processes. Periodic signals with S/Ns above 1.5σ in both data sets and above 3σ in single data sets then proceed to visual inspection. Some targets detected by the BLS method (Kovács et al. 2002) in the latter transit signal-searching module, most of which are eclipsing binaries, are also appended to the results of this module. Only obvious and regular periodic signals that could be classified as binaries or pulsating stars are selected and published in this work. Signals that are classified as rotating spots, rotating ellipsoidal variables, and other semiperiodic/irregular variables will be studied and released in future works.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To determine the period of a variable that has multiple candidate periods, we fold its light curve with the period found from each of the LS, AoV, and BLS methods and select the one that results in the smallest rms. Then we rerun the corresponding method at a range of ±0.1 days around the selected period with a time step of 0.0001 days. We further fold the light curve according to the updated period and inspect the plot visually. Confusion is usually caused by some binaries whose primary and secondary eclipses are similar. In this case, the half period often mimics the true period. To distinguish the tiny difference between eclipses, we double the period when we fold the light curve. Therefore, the plot shows the phase from 0.0 to 2.0, and the second phase is not identical to the first one. This ensures that the plot shows two or four deep eclipses, no more and no less (if it has only one or more than four deep eclipses, then we know the original period is wrong; see panels (b1), (b2) and (c1), (c2) in Figure 14), and makes it easier to compare the odd and even eclipses. In most cases, we will find a slight but distinguishable difference between the depth of the odd and even eclipses (e.g., AST3II096.5462-68.5901 in panel (b1) of Figure 14) or a tiny secondary eclipse between two adjacent deep eclipses (e.g., AST3II106.3396-69.1610 in panel (c1) of Figure 14). For contact eclipsing binaries, if the odd and even eclipses are just identical, then we assume the primary and secondary eclipses are both present but indistinguishable and select the period that makes four eclipses from phase 0.0 to 2.0 (e.g., AST3II108.2786-71.7907 in Figure 16). However, for detached eclipsing binaries, if there is no obvious secondary eclipse found, we assume the secondary eclipse is present but undetectable at our precision and select the period that makes two eclipses from phase 0.0 to 2.0 (e.g., AST3II098.1909-70.8216 in Figure 17).

We scanned 10 fields within the southern CVZ of TESS in the polar nights of 2016. The raw image data acquired by the AST3-II telescope were reduced. Each catalog that is released along with this work is first cross-matched with the APASS point-source catalog in the Sloan i band and calibrated by the fitted zero-point magnitude (see Figure 10). The matching radius is set to be 2''. Although we defocused the telescope (FWHM ∼ 5'') and adopted a photometry aperture up to 12'', this is no significant blending issue. The reasons are: first, we have to avoid the crowded field, e.g., fields close to the LMC; second, we adopted a short exposure time (10 s), and not too many faint stars were caught; and last, we have a high spatial resolution (pixel scale ∼1'' pixel⁻¹), so neighboring targets can be easily distinguished even if they fall in the same aperture. Light curves of stars brighter than 15th magnitude are extracted, detrended (see Figure 12), and released along with this work. From these light curves, we have found 221 variables within the southern CVZ of TESS, including 117 binaries and 104 pulsating stars. Their phase-folded light curves are shown in Figures 15–23. Our final variable catalog is cross-matched with the latest AAVSO database and the TESS Input Catalog (TIC; Stassun et al. 2018) using a matching radius of 5''. This large matching radius is in consideration of our defocused PSF. In practice, if multiple matches happened within a matching radius, we select the ones with the nearest magnitude. Fortunately, the space distribution of known variables in AAVSO and preselected bright stars in TESS is quite sparse. So, no confusion in cross-matching was reported by our matching process. We list the IDs of matched targets from both the AAVSO database and the TIC in Table 4. The target stellar properties, such as stellar radius, mass, and variable-star type, are also listed. We note that there are 67 targets (identified with a candidate target list (CTL) flag of 1 in Table 4) found in the exoplanet CTL for TESS. Targets in the CTL are preselected objects that have higher priorities to search for transiting exoplanets using short-cadence (∼2 minutes) light curves. Our detection of variability in these targets indicates that additional care needs to be taken when searching for transit signals in these targets. There are 42 variables in common between our catalog and the AAVSO database. Panel (a) in Figure 14 shows the comparison between the periods of common variables found by AST3-II and in the AAVSO database. There are only two mismatches in period: AST3II096.5462-68.5901 (ASAS J062611-6835.4) and AST3II106.3396-69.1610 (YZ Vol). We plot their phase curves in panels (b1), (b2), (c1), and (c2) according to the periods from AST3-II and the AAVSO database. The periods found from the AST3-II observations (panels (b1) and (c1)) are clearly more convincing.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

As shown in Figure 12, the valid magnitude range for our survey is from ${{\boldsymbol{m}}}_{i}=10$ to ${{\boldsymbol{m}}}_{i}=15$. Most stars brighter than ${{\boldsymbol{m}}}_{i}=10$ mag are likely to be saturated. However, on occasion, the extinction caused by bad weather or frosting on the front window of the telescope can be as high as 4 mag. During such times of poor photometric conditions, only the brightest stars can be observed, which has the fortunate result that some parts of the light curves of very bright stars ($7.5\lt {{\boldsymbol{m}}}_{i}\lt 10$) are still usable. After filtering out the remaining saturated measurements, the photometric precision from these highly extincted images is sufficient to identify some variables with large periodic variations. For example, target "AST3II111.7546-69.5154" has a magnitude of 8.45 in the i band and a variation period of 0.84726 days. It has been labeled as an "EW/KE" variable²⁶ in the AAVSO database with a variation period of 0.84743 days. Target "AST3II092.1221-71.1200" is 9.09 mag in the i band and has a variation period of 1.71539 days. In the AAVSO catalog, its name is "ASAS J060829-7107.2," which is labeled as an "ACV" variable²⁷ with a period of 1.71443 days.

Another thing worthy of notice in Figure 12 is that the rms values of faint stars are not improved by binning from 12 to 36 minutes. A possible reason is that the fainter a star is, the less data points are in its light curve. Most of the valid measurements are usually crowded into those days with good observing conditions. For the other days with large extinction, light curves of faint stars are very sparse, say 1–2 points per hour, on average. When we divide these time series into equally spaced intervals with 12 or 36 minutes, many intervals are actually empty or have only one measurement. As a result, the rms of the entire light curve may be dominated by these sparse parts and will not be improved by increasing the binning time from 12 to 36 minutes. Another reason is that some systematic errors may still be buried in the photometry of stars fainter than ${{\boldsymbol{m}}}_{i}=14.0$. This effect raises an uncertainty in the rms at a level of several percent, which is not a serious problem for most variables stars, since the amplitudes of their brightness variations are usually larger than 10%. For transiting exoplanet searching, we abandon stars fainter than 14.0th magnitude.

5.1. Variables Found by AST3-II

We present some of the latest results from CHESPA. This program is based on the AST3-II telescope located at the Chinese Kunlun station at Dome A, Antarctica. This first data release contains a data set obtained in the austral winter of 2016 with target fields within the southern CVZ of TESS. The data set is available to the community through the website of the School of Astronomy and Space Science, Nanjing University,²⁸ and the Chinese Astronomical Data Center.²⁹ The released data consist of three parts: calibrated catalogs, light curves for general science usage, and a catalog of newfound variables. In addition, science images can be made available via email requests to . The science images include reduced FITS images with headers containing the observation information, including target name, world coordinate system, observation date, and exposure time.

The calibrated catalogs contain all sources found by our pipeline with an S/N above a threshold of 3σ. Observation conditions and statistics of image quality are also included in the header of each catalog (see Table 2). The light curves for general science usage include sources that are cross-matched with the sources in the APASS catalog with a limiting magnitude of ${{\boldsymbol{m}}}_{i}\leqslant 15$. There are two subsets of light curves: (1) original light curves that inherit all the information from the corresponding catalogs and (2) polished light curves that are detrended and binned to a cadence of 12 minutes. The catalog of variables shows coordinates, magnitudes, periods, and stellar properties cross-matched with the AAVSO database and the TIC for all of the variables we have found within the southern CVZ of TESS (see Table 4). There are 42 variables that are matched in the variable catalog of AAVSO. Although classification is not the major goal of this work, we have achieved a high consistency with the AAVSO database. Only three of these 42 variables are not consistent: target "AST3II092.1221-71.1200" ("ACV" or "CEP"), "AST3II096.3232-71.3623" ("ROT" or "RR"), and "AST3II102.4829-71.7105" ("YSO" or "EW"). From the shape of our light curves, we think our classification is more convincing. Note that we have included the contact eclipsing binary, "EC," the detached eclipsing binary, "ED," and the semidetached eclipsing binary, "ESD," in the AAVSO database in "EW," "EA," and "EB," respectively. The two types in each pair, e.g., "EC" and "EW," are basically the same type in two different classification systems. There are 179 targets that are newly found variables, and 67 targets are also listed in the potential planet candidate list of TESS. Since we have only selected variables with regular shapes and obvious periods, many irregular and/or long-term variables are not included in our variable catalog. These objects will be studied and released in future works. The minimum variation reliably detected by our survey is below 5 mmag, showing that we have the ability to find transiting exoplanets. A detailed catalog of transiting exoplanet candidates found by our survey is in preparation.

This work was supported by the Natural Science Foundation of China (NSFC grants 11673011, 11333002, and 11273019) and the National Basic Research Program (973 Program) of China (grant Nos. 2013CB834900 and 2013CB834904). The authors deeply appreciate all the CHINAREs for their great effort in installing/maintaining CSTAR, CSTAR-II, AST3-I, AST3-II, and PLATO-A. This study has also been supported by the Chinese Polar Environment Comprehensive Investigation & Assessment Program (grant No. CHINARE2016-02-03), the Australian Antarctic Division, and the Australian National Collaborative Research Infrastructure Strategy administered by Astronomy Australia Limited. HZ is also grateful to the High Performance Computing Center (HPCC) of Nanjing University for reducing the data used in this paper. This research was made possible through the use of the AAVSO Photometric All-Sky Survey (APASS), funded by the Robert Martin Ayers Sciences Fund.

Software: Sextractor (Bertin & Arnouts 1996), Swarp (Bertin et al. 2002), VARTOOLS (Hartman & Bakos 2016), Astrometry.net (Lang et al. 2010), MATLAB.