Photometric calibration of a wide-field sky survey data from ...

Received 29 April 2016; Revised 6 June 2016; Accepted 13 June 2016

DOI: xxx/xxxx

ARTICLE TYPE

Photometric calibration of a wide-field sky survey data fromMini-MegaTORTORA

S. Karpov*1,2,3 | G. Beskin2,3 | A. Biryukov4 | S. Bondar5 | E. Ivanov5 | E. Katkova5 | N.Orekhova5 | A. Perkov5,3 | V. Sasyuk3

1CEICO, Institute of Physics, CzechAcademy of Sciences, Prague, CzechRepublic

2Special Astrophysical Observatory, RussianAcademy of Sciences, Nizhniy Arkhyz,Russia

3Institute of Physics, Kazan FederalUniversity, Kazan, Russia

4Sternberg Astronomical Institute, MoscowState University, Moscow, Russia

5Research and Production Corporation“Precision Systems and Instruments”,Nizhniy Arkhyz, Russia

Correspondence*Sergey Karpov, Institute of Physics ofCzech Academy of Sciences, Na Slovance1999/2, 182 21 Praha 8, Czech RepublicEmail: [email protected]

Present AddressSergey Karpov, Institute of Physics of CzechAcademy of Sciences, Na Slovance 1999/2,182 21 Praha 8, Czech Republic

Funding InformationCzech Ministry of Education, Youth andSports, CZ.02.1.01/0.0/0.0/15 003/0000437.Russian Science Foundation, 14-50-00043.RFBR, 17-52-45048.

Mini-MegaTORTORA is a 9-channel wide-field camera that continuously monitorsthe sky looking for rapid optical transients since mid-2014. It is also performing aregular sky survey, and has already acquired nearly half million images coveringevery point of northern sky hundreds to thousands times. Photometric analysis ofthese data may provide a huge amount of information useful for detection and char-acterization of different types of variable objects. Here we present a brief descriptionof our activities related to acquisition, processing and calibration of these data, aswell as examples of uncatalogued variable stars of various types detected during theanalysis.

KEYWORDS:surveys, techniques: photometric, methods: data analysis, stars: variables: general

1 INTRODUCTION

Mini-MegaTORTORA is a 9-channel optical wide-field sys-tem intended for continuous monitoring of the sky with sub-second temporal resolution in order to detect rapid opticaltransients of various types, from cosmic gamma-ray burststo meteors and artificial satellites (Beskin et al., 2010, 2017,2014, 2013; Karpov et al., 2017a; Karpov, Katkova, et al.,2016; Karpov, Orekhova, et al., 2016). Its every channel has10x10 degrees field of view provided by the Canon EF85/1.2lens and 5.5 megapixel Andor Neo sCMOS detector, a set of

installable color and polarimetric filters, and a coelostat mirrorfor a rapid repointing, which allows for either mosaicking thelarger field of view, or for pointing all the channels in one direc-tion. The latter regime together with the filters are typicallyused for rapidmulti-regime follow-up of the transients detectedby the system (Karpov et al., 2017b), while most of the timethe sky is observed in a “monitoring” wide-field regime withno filters installed, i.e. in white light, covering ∼900 squaredegrees simultaneously with temporal resolution of 0.1 s.Mini-MegaTORTORA started its operation in June 2014,

and routinely monitor the sky since then. The observations aregoverned by the dedicated dynamic scheduler optimized forperforming the sky survey. The scheduler works by selecting

2 S. Karpov ET AL

0h3h6h9h12h 15h18h21h

0

500

1000

1500

2000

2500

3000

3500

4000

Num

ber o

f sur

vey

fram

es

FIGURE 1 Number of survey frames acquired by Mini-MegaTORTORA over last four years covering every point ofthe sky. Contours correspond to the density of Tycho-2 cat-alogue stars, basically tracing the Galactic plane. The peakof Mini-MegaTORTORA coverage corresponds to the skyregions close to zenith at the observatory site during the winterwhen majority of observations occur.

the next pointing forMini-MegaTORTORAby simultaneouslyoptimizing the following parameters: distances from the Sun,Moon and the horizon should be maximized, distances fromthe current pointings of Swift and Fermi satellites should beminimized, and the number of frames already acquired on agiven sky position that night should be minimized. In thisway more or less uniform survey of the whole visible skyhemisphere1 is being performed while maximizing the prob-ability of gamma-ray bursts observations simultaneous withthese satellites. The Mini-MegaTORTORA typically monitorsevery sky field continuously for 1000 seconds before mov-ing to the next pointing. Before and after observing the fieldwith high temporal resolution, the system acquires deeper“survey” images with 60 seconds exposure (replaced sincemid-2015 with 3 consecutive 20 s exposures in order to min-imize the impact of cosmic rays on the survey) in white light,having significantly deeper (down to ∼14 mag) limit than typ-ical “monitoring” images (whose limit is about 11 mag). Theframes acquired to date in this regime cover every point of thenorthern sky hundreds to thousands times (see Figure 1 ), andmay be potentially used for both detection and characterizationof the variability of astrophysical objects on time scales fromhours to years in the magnitude range between 8 and 13.In this article we report the status of our ongoing activ-

ities on processing and analyzing these survey data fromMini-MegaTORTORA.

1Of course, due to significant difference of night duration at different seasonsthe yearly sky coverage is not uniform, with more pointings covering the regionsvisible in winter time.

2 PHOTOMETRY OF SURVEY FRAMES

The imaging data acquired by Mini-MegaTORTORA in sur-vey regime are contaminated with various defects and artifacts,related both to the instrumentation used and observing con-ditions. The detector, Andor Neo sCMOS, sports a numberof problems for photometry, including quite significant over-all non-linearity (Karpov et al., 2018) and excessive noiseof a small (∼1%) subset of pixels, with some of them pre-masked and being filled on the fly with average value of theirsurroundings by a camera firmware, while the others still man-ifesting in the images as either pixels with erratic values orlong after-memory. The point-spread function (PSF) providedby Canon EF85/1.2 objective is characterized by an extendedwings which both complicates the processing in moderatelycrowded fields (as there are effectively no regions with truesky background in the image) and causes significant back-ground level variations around bright objects. Moreover, somefraction of images are contaminated by the artifacts causedby the satellite laser range meter system installed close toMini-MegaTORTORA (see left panel of Figure 3 for anexample image), as well as by occasional trees and horizon linewhen observing at large zenith distances. Therefore, propermasking of unusable image regions is an extremely importantpart of photometric processing. It will be described below inSection 2.2.

2.1 Image reductionFor the reduction of survey images we use the dark framesacquired during normal Mini-MegaTORTORA operation.They are typically acquired in the evening with both closeddome and closed channel cover to reduce the amount of straylight. 1000 frames with 0.1 s exposure are acquired in everydark sequence and averaged on the fly to produce single masterdark frame (which we will refer below just as dark frame). Sev-eral dark sequences are usually acquired consecutively. Due todynamic baseline clamp feature of the detector, the bias level isessentially equal to the one of dark frame, is roughly stable overtime and does not depend on the image exposure. Thereforewe subtract from every survey image the dark frame closest intime.Next, we apply the non-linearity correction for bias-

subtracted image using coefficients measured for every pixel ina set of dedicated laboratory characterization measurements.Details of this procedure are outlined in Karpov et al. (2018).The final step of image reduction is flat-fieldingwhich is per-

formed using “night sky flats”2 – median averages of all surveyframes acquired in a given hardware configuration (channel id,

2Acquisition of proper evening flats is nearly impossible on Mini-MegaTORTORA due to limited dynamic range of Andor Neo sCMOS detectors and

S. Karpov ET AL 3

FIGURE 2 Examples of “night sky flats” – flat frames derived using median averaging of all normalized survey images fordifferent channels. Both large-scale vignetting and small-scale features are successfully captured. These flats allow successfulcorrection of most of vignetting and non-uniform sensitivity features seen in the survey frames, but most probably not time-dependent inter-pixel sensitivity variations, which will manifest as an uncorrected fixed-pattern noise in the reduced images.

coelostat mirror and focus positions, etc) with all stars maskedand mean level normalized to common value. Such flats suc-cessfully capture both large-scale vignetting and small-scaledetector inhomogeneities (see Figure 2 ).

2.2 MaskingWe implement the masking of detector defects and image arti-facts in a three-step way, applied after image reduction andlinearization. First, we mask all “bad” pixels according to themap derived in a set of dedicated laboratory tests of everydetector, as well as pixels close to saturation in the originalimage. Next, we compute the background using traditionalmesh-based approach and amodal estimator as implemented inSEP (Barbary, 2016) library adaptation of SExtractor (Bertin& Arnouts, 1996) algorithms. We do it on several mesh scales(e.g. with 64 pixels and 256 pixels step) and then select theregions where background estimations are significantly (>5�) different. Such regions correspond to the positions insideimage where background rapidly changes, and therefore can’tbe reliably reconstructed, making them unusable for aperturephotometry with global background model as we use. Then weapply a dilation operator to these regions, extending them by∼100 pixels in every direction, and use the resulting bitmapfor masking. Finally, we perform SExtractor-like object detec-tion on background-subtracted image using the combination

extremely large field of view. The latter also prevents from using artificially con-structed “dome flats” as it is extremely difficult to make an uniform illumination ofa large field of view.

of two masks defined above, perform simple thresholding onthe list of sizes of extracted objects, select the footprints of theobjects significantly larger than the rest, dilate them using 10pixels wide kernel and use the resulting bitmap as one moremask. After that, we mask the image using all three maskingcomponents (see right panel of Figure 3 for an example offinal mask), construct the final background, detect again theobjects, estimate their FWHM and then perform simple aper-ture photometry in circular regions with 1.5 average FWHMradius (optimal aperture) using the routines from SEP library(Barbary, 2016). Then the results of aperture photometry arefiltered, and all erroneous measurements or measurementswith worse than 10% formal accuracy are discarded.

3 PHOTOMETRIC CALIBRATION

The list of objects detected on every frame is then astromet-rically calibrated using local Astrometry.Net (Lang, Hogg,Mierle, Blanton, & Roweis, 2010) installation, and then posi-tionally matched against Tycho-2 catalogue. The model

instrumental = V +Z0(x, y) + CBV(

BT − VT)

(1)

is then built for instrumental magnitudes which includes cata-logue Vmagnitude (derived using official transformation fromBT and VT magnitudes from Høg et al. (2000)), fourth-orderspatial polynomial Z0(x, y) to correct for residual vignettingleft uncorrected after flat-fielding, as well as to account for

4 S. Karpov ET AL

FIGURE 3 Masking procedure implemented for filtering out imaging defects and artifacts in Mini-MegaTORTORA data. Leftpanel – original image with imaging artifact given in negative scale. Right panel – mask bitmap excluding unreliable sCMOSpixels (dots), regions in the immediate vicinity of saturated objects (smaller spots), and the regions filtered out by multi-scalebackground routine (larger spots). The latter correspond to the regions of the image where background estimation will likely bebiased, and therefore aperture photometric measurements will be unreliable. The artifact is successfully masked.

FIGURE 4 An example of photometric residuals for photo-metric model not taking into account the color term. Charac-teristic dependence on stellar color is seen, which reflects boththe difference of catalogue and instrument bandpasses and theinfluence of secondary extinction, which in turn depends onatmospheric conditions.

positionally-dependent aperture correction due to PSF varia-tion over the frame, and a color term proportional to BT −VT . The model is then fitted to the data using only the cat-alogue stars which are not potentially blended3 and have noextraction flags (i.e. not saturated, not truncated and do notcontain masked pixels inside their footprints). The last term– color one – captures both the intrinsic photometric sys-tem ofMini-MegaTORTORA hardware defined by its detectorsensitivity and objective throughput (which in general differsbetween channels of the system), and due to its extremely wide

3To quantify the potential blending of stars we are cross-matching Tycho-2catalogue (Høg et al., 2000) with 2MASS point source one (Skrutskie et al., 2006),and mark every star which has more than one matched objects with J < 12m asunreliable.

white light bandpass – the atmospheric conditions and extinc-tion level. The Figure 4 shows an example of photometricresiduals if this latter term is not taken into account, wherecharacteristic trend on stellar color is seen. The slope of thisdependence suggests also non-zero higher order terms, but inour analysis we approximate it with just a linear one, whoseslope is a free parameter for every frame.As a result of such fitting, we have a set of zero point Z0

coefficients specific for every object in every frame, and CBVcoefficients specific to every frame and common for all objectsin it. Below we will show how these coefficients may be usedfor a statistical reconstruction of stellar colors and reduction ofphotometric measurements to standard system.

3.1 Statistical reconstruction of object colorsAs we do not generally know the actual color of every objectseen in the frame, we can’t just compute the magnitudesin standard system for them using Eq. 1. However, under areasonable assumption of independence of atmospheric con-ditions from intrinsic object variability we may statisticallydetermine these colors by examining the scatter of their lightcurves. To do so, we first cluster all measurements from indi-vidual frames into the sequences belonging to separate objectsby means of their spatial closeness on the sphere, and then useEq. 1 to fit for BT − VT that minimizes the scatter of V val-ues derived using instrumental magnitudes and correspondingZ0 zero point values. This way we derive both the colors and

S. Karpov ET AL 5

FIGURE 5 Comparison of V magnitudes (upper panel)and BT − VT colors (lower panel) derived from Mini-MegaTORTORAwhite light multi-epoch photometry with thevalues listed in Pickles & Depagne (2010) catalogue. The scat-ter of magnitude residuals in upper panel is primarily due tointrinsic errors in the Tycho-2 catalogue. The diagonal featureon the faint edge is due to magnitude limit of Tycho-2 cat-alogue and intrinsic correlation between the magnitude itselfand magnitude residuals.

mean magnitudes in standard system for every object seen onthe sufficiently large number of frames4.To test the validity of this methodology, we compared the

derived V magnitudes and BT − VT colors with the catalogueof synthetic photometry based on cross-matching Tycho-2with 2MASS and NOMAD catalogues by Pickles & Depagne(2010). Figure 5 shows that the derivedmagnitudes are essen-tially unbiased, and the colors also correlate well with thecatalogue ones. There is, however, a slight systematic shiftin colors there, especially for brighter stars, which requiresadditional analysis.Finally we should note that this methodology is essentially

equivalent to the popular SysRem family of trend filteringalgorithms (Tamuz, Mazeh, & Zucker, 2005), just with anexplicit specification of the source of systematic effects.

8 9 10 11 12 13V magnitude

0.00

0.05

0.10

0.15

0.20

RMS

AAVSO VSX

8 9 10 11 12 13V magnitude

0.00

0.05

0.10

0.15

0.20

RMS

AAVSO VSX

FIGURE 6 The intrinsic scatter of light curves for a singlefield before (left panel) and after (right panel) refinement run,which includes rejection of frames with badly fitted photomet-ric model from Eq. 1. Red circles represent already cataloguedvariable stars from AAVSO Variable Star Index (Watson etal., 2006). The refinement essentially removes the majority ofoutlier points in the light curves leaving only actual stellarvariability. The residual scatter may be used as a criterion fordetecting variable objects.

3.2 Variability detectionAfter determining the mean magnitudes and colors of everyobject in the field, we may re-run calibration of individualframes using these values as a reference catalogue and exclud-ing the ones with large intrinsic scatter. This way the numberof matched objects on every frame is significantly larger whichleads to much better fitting of Z0 spatial polynomial part.Moreover, during this “refinement” run we may explicitlyexclude from the analysis the frames where photometric modelfit (Eq. 1) converges poorly, or even where the number ofdetected stars is much smaller than expected, which may sig-nal extremely bad observing conditions. This filtering, alongwith the refinement procedure, lead to significant improve-ment of light curves stability (see Figure 6 ), even without anyadditional light curve level filtering.There are a lot of possible techniques for the detection of

variability in the ensemble of light curves (see e.g. Sokolovskyet al. (2017) for a review). We are presently applying the

4Note however that it is impossible to apply this methodology to objects seenon just a few frames, and therefore to get an estimation of a standard magnitudesfor an intrinsically transient objects. Such objects are outside the scope of presentarticle, and their search and characterization will be the topic of a separate publica-tion. Moreover, this procedure assumes that the color (SED shape) of the object isconstant in time; it will give biased results if it does change. As the weather changesare uncorrelated with such intrinsic variability, it will result in an increased scatterof the light curve.

6 S. Karpov ET AL

simplest criteria, which employs the analysis of scatter-vs-magnitude diagram like the ones shown in Figure 6 . We usethe SuperSmoother algorithm (Vanderplas, 2015b) for estimat-ing the level of noise scatter, and then apply the threshold onthe scatter level to detect the objects with smooth large ampli-tude variability. This simple approach successfully recovers anumber of already known variables listed in AAVSO VariableStar Index (Watson et al., 2006), as well as a lot of uncata-logued ones showing various types of periodic and aperiodicvariability (see Figure 7 for some examples).Note that the lower panel of Figure 6 shows that the light

curve spread is about 0.02 mag even for brighter stars, muchlarger than theoretical accuracy expected from Poisson noise,atmospheric scintillations and read-out noise levels. It sug-gests the presence of some additional noise source that limitsthe actual photometric performance of the system. We tendto believe its source to be the pixel-level instability of thesCMOS detectors, leading to slowly-varying inter-pixel sensi-tivity variations and therefore fixed-pattern noise, which can’tbe completely corrected by a flat-fielding using night sky flatsas derived in Section 2.1.

4 CONCLUSIONS

The amount of data acquired by Mini-MegaTORTORA insurvey regime is huge, and its photometric processing andcalibration will allow for a large scale variability character-ization of the stars in the 8-13 mag range, which is poorlycovered by current variability surveys, across the whole north-ern hemisphere. While its accuracy is not sufficient for thetasks like detection of transiting exoplanets or small amplitudestellar pulsations, it will still allows the detection of signif-icant number of bright uncatalogued variables. We plan toboth perform the complete variability analysis of all the lightcurves in order to detect such variables, and to publish freelyaccessible light curve extraction web service for the wholeMini-MegaTORTORA survey archive on the project web site.

ACKNOWLEDGMENTS

This work was supported by European Structural and Invest-ment Fund and the Czech Ministry of Education, Youthand Sports Project CoGraDS – CZ.02.1.01/0.0/0.0/15003/0000437. Mini-MegaTORTORA belongs to KazanFederal University and the work is performed accord-ing to the Russian Government Program of CompetitiveGrowth of Kazan Federal University. Observations onMini-MegaTORTORA are supported by the Russian Sci-ence Foundation grant No. 14-50-00043 (exoplanets). The

reported study was funded byRFBR according to the researchproject No. 17-52-45048.

Author contributionsS.K. developed the implemented the approach to photomet-ric calibration, created the software for data acquisition andanalysis, and prepared the manuscript. G.B. formulated the ini-tial problem, and participated in discussions on its progressand results, as well as in preparation of manuscript. A.B. con-tributed to method formulation and software preparation. S.B.,E.I., E.K., N.O., A.P. and V.S. participated in construction,maintenance and operation of the instrumentation used for dataacquisition.

Financial disclosureNone reported.

Conflict of interestThe authors declare no potential conflict of interests.

REFERENCES

Barbary, K. 2016, Journal of Open Source Software, 1(6), 58. doi:Bertin, E., & Arnouts, S. 1996, June, A&A, 117, 393-404.Beskin, G., Bondar, S., Karpov, S. et al. 2010, Advances in

Astronomy, 2010. doi:Beskin, G., Karpov, S., Biryukov, A. V. et al. 2017, January,

Astrophysical Bulletin, 72, 81-92. doi:Beskin, G., Karpov, S., Bondar, S. et al. 2014, December, Mini-

MegaTORTORA status update. Revista Mexicana de Astrono-mia y Astrofisica Conference Series Vol. 45, p. 20-.

Beskin, G., Karpov, S., Plokhotnichenko, V. L. et al. 2013, August,Physics Uspekhi, 56, 836-842. doi:

Høg, E., Fabricius, C., Makarov, V. V. et al. 2000, March, A&A, 355,L27-L30.

Karpov, S., Beskin, G., Biryukov, A. et al. 2017a, June, Mini-MegaTORTORA Wide-Field Monitoring System with Subsec-ond Temporal Resolution: Observation of Transient Events.Y. Y. Balega, D. O. Kudryavtsev, I. I. Romanyuk, et al. (Eds.),Stars: From Collapse to Collapse Vol. 510, p. 526.

Karpov, S., Beskin, G., Biryukov, A. et al. 2017b, Untriggered searchfor rapid optical transients withMini-MegaTORTORAwide-fieldmonitoring system. A. Gomboc (Ed.), New Frontiers in BlackHole Astrophysics Vol. 324, p. 85-86. doi:

Karpov, S., Beskin, G., Biryukov, A. et al. 2018, Mini-MegaTORTORA wide-field monitoring system with sub-secondtemporal resolution: observation of transient events. Proceed-ings of The International Conference “SN 1987A, Quark PhaseTransition in Compact Objects and Multimessenger Astronomy”,Russia, Terskol (BNO INR RAS), Nizhnij Arkhyz (SAO RAS),2-8 July 2017 p. 86-95. INR RAS.

S. Karpov ET AL 7

FIGURE 7 Examples of uncatalogued large amplitude variable stars of various types detected in a single sky field. For everystar, the panels display overall light curve reduced to standard Vmagnitude system, Lomb-Scargle periodogram computed usingGATSPy code (Vanderplas, 2015a), and the folded light curve corresponding to the largest peak in periodogram.

Karpov, S., Katkova, E., Beskin, G. et al. 2016, December, Massivephotometry of low-altitude artificial satellites on Mini-Mega-TORTORA. Revista Mexicana de Astronomia y AstrofisicaConference Series Vol. 48, p. 112-113.

Karpov, S., Orekhova, N., Beskin, G. et al. 2016, December, Meteorobservations with Mini-Mega-TORTORA wide-field monitoring

system. Revista Mexicana de Astronomia y Astrofisica Confer-ence Series Vol. 48, p. 97-98.

Lang, D., Hogg, D. W., Mierle, K., Blanton, M., & Roweis, S. 2010,May, AJ, 139, 1782-1800. doi:

Pickles, A., & Depagne, É. 2010, December, PASP, 122, 1437. doi:Skrutskie, M. F., Cutri, R. M., Stiening, R. et al. 2006, February, AJ,

131, 1163-1183. doi:

8 S. Karpov ET AL

Sokolovsky, K. V., Gavras, P., Karampelas, A. et al. 2017, January,MNRAS, 464, 274-292. doi:

Tamuz, O., Mazeh, T., & Zucker, S. 2005, February, MNRAS, 356,1466-1470. doi:

Vanderplas, J. 2015a, February, gatspy: General tools for Astronom-ical Time Series in Python. Retrieved from https://doi.org/10.5281/zenodo.14833 doi:

Vanderplas, J. 2015b, January, supersmoother: Efficient PythonImplementation of Friedman’s SuperSmoother. Retrieved fromhttps://doi.org/10.5281/zenodo.14475 doi:

Watson, C. L., Henden, A. A., & Price, A. 2006, May, Society forAstronomical Sciences Annual Symposium, 25, 47.

How cite this article: S. Karpov, G. Beskin, A. Biryukov, S. Bondar,E. Ivanov, E. Katkova, N. Orekhova, A. Perkov, and V. Sasyuk (2018),Photometric calibration of a wide-field sky survey data from Mini-MegaTORTORA, Astronomische Nachrichten, 2018;00:1–6.

AUTHOR BIOGRAPHY

Karpov, Sergey. Sergey Karpov finishedMoscow State University in 2003, andgot his PhD in astrophysics in Spe-cial Astrophysical Observatory, Russia, in2007. His scientific interests include hightemporal resolution astrophysics, time-domain sky surveys, astronomical data

processing pipelines and transient detection algorighms. He iscurrently working on development of automated data analy-sis pipelines for several time domain sky surveys, as well ason various aspects of testing and characterization of opticaldetectors used in astronomy.

How cite this article: S. Karpov, G. Beskin, A. Biryukov,S. Bondar, E. Ivanov, E. Katkova, N. Orekhova, A. Perkov,and V. Sasyuk (2018), Photometric calibration of a wide-fieldsky survey data from Mini-MegaTORTORA, AstronomischeNachrichten, 2018;00:1–6.