arX

iv:1

108.

4196

v2 [

astr

o-ph

.IM

] 2

3 A

ug 2

011

PIRATE: A Remotely-Operable Telescope Facility for Research

and Education

Holmes1, S., Kolb1, U., Haswell1, C. A., Burwitz2, V., Lucas1, R. J., Rodriguez3, J., Rolfe4,

S. M., Rostron5, J., Barker1, J.

ABSTRACT

We introduce PIRATE, a new remotely-operable telescope facility for use in research andeducation, constructed from ‘off-the-shelf’ hardware, operated by The Open University. Wefocus on the PIRATE Mark 1 operational phase where PIRATE was equipped with a widely-used 0.35m Schmidt-Cassegrain system (now replaced with a 0.425m corrected Dall Kirkhamastrograph). Situated at the Observatori Astronomic de Mallorca, PIRATE is currently used tofollow up potential transiting extrasolar planet candidates produced by the SuperWASP Northexperiment, as well as to hunt for novae in M31 and other nearby galaxies. It is operated bya mixture of commercially available software and proprietary software developed at the OpenUniversity. We discuss problems associated with performing precision time series photometrywhen using a German Equatorial Mount, investigating the overall performance of such ‘off-the-shelf’ solutions in both research and teaching applications. We conclude that PIRATE is acost-effective research facility, and also provides exciting prospects for undergraduate astronomy.PIRATE has broken new ground in offering practical astronomy education to distance-learningstudents in their own homes.

Subject headings: Astronomical Instrumentation, Data Analysis and Techniques, Extrasolar Planets

1. Introduction

PIRATE (Physics Innovations Robotic Astro-nomical Telescope Explorer) is a remote telescopefacility situated on the Balearic island of Mallorca(2◦57′03.34′′E, 39◦38′34.31′′ N), at the Observa-tori Astronomic de Mallorca, ∼162m above sealevel. It is a remotely-operable facility used forresearch and undergraduate teaching, allowing fora practical astrophysics component to be imple-mented within The Open University’s (OU, Mil-ton Keynes, UK) distance-learning modules. PI-

1Department of Physics and Astronomy, The Open Uni-versity, Milton Keynes, MK7 6AA, UK

2Max-Planck-Institut fur extraterrestrische Physik,Giessenbachstrasse, 85748, Garching, Germany

3Observatori Astronomic de Mallorca, Cam del’Observatori, 07144 Costitx, Mallorca, Spain

4Planetary and Space Sciences Research Institute, TheOpen University, Milton Keynes, MK7 6AA, UK

5Department of Physics and Astronomy, University ofWarwick, Coventry, CV4 7AL, UK

RATE is constructed entirely from ‘off-the-shelf’hardware, and is operated primarily with commer-cial observatory control software, with some addi-tional proprietary software developed to supportsimultaneous use by groups of students. The totalcost of the facility, combining all purchased hard-ware and software is of order $150k.

The funding for PIRATE was primarily madeavailable by a teaching innovation initiative(piCETL 1) to explore the integration of aremotely-controlled observatory into university-level distance teaching modules. The aim of theproject was to reproduce the hands-on experienceof a traditional lab course with real-time accessto a telescope from any computer linked to theinternet.

In this paper we refer to the PIRATE Mark1 facility (unless mentioned otherwise) which fea-

1http://www8.open.ac.uk/opencetl/physics-innovations-centre-excellence-teaching-and-learning

1

tured a Celestron2 C14 0.35m Schmidt-CassegrainOptical Tube Assembly (OTA). This OTA is cur-rently no longer in use within the PIRATE facil-ity, having been superseded in August 2010 by a0.425m PlaneWave Systems3 CDK17 f/6.8 Cor-rected Dall-Kirkham Astrograph telescope, whichmakes use of a new custom micro-focuser (pro-vided by PlaneWave), which we designate PI-RATE Mark 2. The Mark 1 hardware remainsan obvious choice for cost-effective astronomy, sowe report in detail on its capabilities, features, anddrawbacks.

1.1. Use in Education

PIRATE needs to cater for a relatively largestudent body with little or no prior experience inobservational astronomy, and with a very limitedamount of remote supervision by a tutor. Grant-ing a sufficiently high level of access to the facility,such as the ability to initiate the opening and clos-ing of the dome, is considered essential for the stu-dent learning experience. The adopted softwaresolution, a combination of the commercial prod-uct ACP4 (Denny 2011) and the proprietary soft-ware described in Lucas & Kolb (2010), achievesthe desired student access privileges without com-promising the safe operation of PIRATE. Afterone tutor-led induction evening conducted via au-dio communication and web interfaces, the stu-dent groups could operate PIRATE successfullyon their own. A (remote) night-duty astronomerwas on call but only occasionally contacted.

PIRATE was deployed for OU undergraduatestudents for the first time in spring 2010 in a 10week project which is part of the third level (thirdyear) OU module S382 Astrophysics. A total ofabout 30 students formed three groups with alter-nating access to PIRATE, for a total of 40 observ-ing nights. Each group selected a suitable targetsource from the catalogue by Norton et al. (2007)of periodic variables found by SuperWASP and co-incident with a ROSAT source. The groups thenbuilt up a long-term light curve of their target,such as the example shown in Fig. 1. Individualobserving sessions were staffed by small observerteams of 2-4 students who kept in audio and text

2http://www.celestron.com3http://www.planewave.com4http://www.dc3.com

contact throughout the night. The collaborationof different observer teams in the larger groups onthe same target source ensured the emergence ofa useable data base to which all group memberscould develop a sense of ownership, even whenoccasional nights were clouded out and the ob-server teams on duty may not have succeeded inobtaining data themselves. A fuller account of thechallenges and solutions for the PIRATE teachingproject can be found in Kolb et al. (2010).

Fig. 1.— The light curve created from 206 I band im-ages of 1SWASPJ163420.90+424433.4. The data weretaken to fulfill project objectives for the OU’s under-graduate course, S382, by OU undergraduate students.

1.2. Use in Research

In its research role, PIRATE performs follow-upphotometry of SuperWASP Transiting ExtrasolarPlanet (TEP) candidates from the SuperWASP(Wide Angle Search for Planets) (Pollacco et al.2006) survey, and discovers extragalactic novae.

Wide-field transit surveys produce light curvesof millions of objects. Transit search algorithms(Collier Cameron et al. 2006) are applied totrend-filtered (e.g. Tamuz et al. (2005)) lightcurves to select TEP candidates from the millionsof surveyed objects. Most of these candidateswill not be TEPs, but will be ‘astrophysical falsepositives’ (AFPs), which are estimated to out-

2

number TEPs in a transit search survey by atleast an order of magnitude (O’ Donovan et al.2006). High priority candidates are followed upwith radial velocity measurements to confirm themass and therefore planetary nature of the transit-ing object, but this requires scarce large-telescopespectroscopy.

To winnow TEP candidates from wide-field sur-veys like SuperWASP, small-to-medium sized tele-scopes provide higher precision, higher spacial res-olution follow-up photometry (Haswell 2010). Inthis role, PIRATE can act as a link in the planet-finding chain, reducing the amount of large tele-scope time spent on false-positives.

PIRATE has also detected novae in M31 (e.g.M31N 2010-01a) (Burwitz et al. 2010), as partof an XMM-Newton large program for studyingthe source population in the Andromeda Galaxy,M31, in which a large number of supersoft X-ray sources were identified as counterparts of op-tical novae (Pietsch 2008). This requires the longterm monitoring of M31 in X-rays with XMM-Newton, Chandra, and occasionally SWIFT. PI-RATE forms part of a wider network of opticaltelescopes that routinely monitor M31 to discoveroptical novae and track their light curves. Op-tical spectra of the freshly discovered novae arealso obtained with larger telescopes, such as theHobby-Eberly Telescope, USA, and the BTA, Rus-sia, to classify the novae. With the help of the in-formation extracted from these X-ray and opticaldata we are beginning to obtain a better picture ofthe physical processes that take place in nova out-bursts. This optical observing campaign grew outof the XMM-Newton/Chandra M31 nova monitor-ing collaboration (Pietsch 2010).

In the following sections we describe the hard-ware and software used in the PIRATE facility, aswell as detailing commissioning issues, data reduc-tion techniques, and future improvements.

2. Hardware and Software

PIRATE Mark 1 is a remote telescope facilityconsisting of a 14′′ Schmidt-Cassegrain reflectormounted on a Paramount ME5 German Equato-rial Mount (GEM) (Fig. 2). Its main imager is anSBIG STL-1001E, which houses the Kodak KAF-

5http://www.bisque.com

1001E CCD, a 1024× 1024 pixel front-illuminateddetector with a pixel size of 24µ, and a quan-tum efficiency of 0.4, 0.55, and 0.65 for wave-lengths 450, 550, 650nm respectively6. In combi-nation with the 3.91m focal length of the CelestronC14, this produces a plate scale of 1.21′′ pixel−1,and a field of view of 22′ × 22′. An Optec TCFmicro-focuser is located between the photometerenclosure and optical assembly, along with an 8position filter wheel, containing Johnson-Cousinsstandard BVRI broadband filters and 3 narrow-band filters (Hα, SII, OII). PIRATE has a Cele-stron 80mm refractor with SBIG ST402 ME cam-era for auto-guiding. The telescope is housedwithin a Baader Planetarium7 3.5m diameter All-Sky clamshell dome (Fig. 3) with built in weathersystems for automatic shut-down in adverse condi-tions. On-site weather is monitored by a weatherstation and Boltwood cloud sensor (to the side ofthe dome), and both internal and external dome-mounted weather systems (rain, internal humid-ity, internal temperature). The dome has itsown firmware which interfaces with the propri-etary dome driver, as well as with the Boltwoodcloud sensor, allowing for shutdown conditions tobe communicated directly to the dome, bypassingthe control PCs. Four D-Link webcams (3 inter-nal, 1 external) provide live video and audio feeds,as well as IR beams for night-viewing.

Fig. 2.— The C14 as seen through the D-LINK DCS-5230 IR webcam

6KAF-1001E spec. sheet: http://www.kodak.com7http://www.baader-planetarium.com/

3

Fig. 3.— The 3.5m Baader Planetarium All Skydome. The control room is directly below.

2.1. Camera Characteristics

To characterise the CCD we determined thegain, linearity, read noise and the dark current(which was measured at a range of temperatures).The gain and the linearity measurements weremade using the same set of dome flats for each.A set of dome flats of increasing exposure timewere taken in order to measure the median countsof a 100 × 100 pixel subframe in the middle of eachimage. For the short exposure times (of order afew seconds), we attempted to generate a shuttercorrection map using the methodology of Zissell(2000), anticipating that we might see position-dependent corrections to the exposure time (dueto the shutter travel) of order 10−3s. However,this produced a null result, and further investi-gation into the shutter mechanism of the STL-1001E confirms a rotating ‘shutter wheel’ thatshould be devoid of shutter travel effects. Wetherefore do not apply any shutter correction tothe short exposure flat fields. We measured thebias level from contemporaneous bias frames andsubsequently subtracted the mean pedestal levelof 107 ADU from each flat. The subframe waschosen to be the centre of the vignetting function,where the image is at its flattest. To assess thelinearity of the CCD, we plotted median subframecounts against exposure time (Fig. 4 (a)), andfit a linear trend to the same ADU range. Theresiduals of this fit can be seen in Fig. 4 (b). Wenote a deviation from linearity at the top end of

the dynamic range of <1%, and <2% at the bot-tom end. We measured the gain using the fol-lowing relationship: σ2

ADU = 1g 〈NADU 〉 (Howell

2000), where σADU is the standard deviation ofthe sub-frame counts, and 〈NADU 〉 is the meanrecorded counts in the subframe. To derive thisexpression, one assumes the statistical relationshipσe− =

√Ne− holds, where σe− and Ne− are the

uncertainty in and the number of recorded photo-electrons respectively, which only holds when theprocess is governed by Poisson statistics. To de-termine g we therefore plot σ2

ADU against 〈NADU 〉(see Fig. 4 (c)). We needed sufficient photoelec-trons for photon-counting (Poisson) noise to bedominant, and so we fitted to the range 20k ADU< 〈NADU 〉 < 40k ADU. We chose an upper limitof 40k ADU to be well clear of the digital count-ing limit of 65535 ADU (16-bit). We measured avalue for the gain of 1.62 ± 0.01 e−ADU−1. Themanufacturers quote a full-well capacity of 150ke−, which, given the determined gain, would oc-cur at a theoretical ADU value of ∼ 92.5k ADU,an unobtainable value due to the digital countinglimit. We note that falling short of the electronfull well capacity is probably responsible for ourexcellent linearity of response at the upper end ofour digital counting range.

To measure the read-out noise, we took 30bias frames at a temperature of −20◦, and me-dian combined them. The median-combined biasframe (which we assume to have negligible readnoise) was then subtracted from each of the 30bias frames, and the standard deviation of each ofthese difference images measured, we averaged thisfigure to achieve the read noise in ADU (Howell2000). We measure then (using the previously de-termined gain value) a read-out noise of 10.9± 1.3e−. To check for consistency, we also split thebias frames into pairs, subtracting one from theother within each pair, so that d = b1 − b2, whereb1 & b2 are the bias frames in each pair, andσread = σd/

√2, yielding a read-noise measure-

ment for each pair. These values were consistentwith the previous read-noise measurement.

Dark current was assessed for a range of tem-peratures (from 15◦C through to -25◦C in in-crements of 5◦C) by median-combining 10 darksat each temperature, 10 biases at each tempera-ture, subtracting the bias and dividing the resid-ual counts by the exposure time (30s for each

4

dark frame). The measured values are listed be-low in Table 1. We measure a dark current of0.02± 0.02 e−s−1 and 0.08± 0.02 e−s−1 at -20◦Cand -10◦C, the typical operating temperatures forwinter and summer respectively. We also find avalue of 0.51± 0.03 e−s−1 at 0◦C, c.f. the 1 e−s−1

at 0◦C manufacturer’s specification.

2.2. The Mount

The Paramount ME GEM is able to deliver anall-sky pointing accuracy of around 2′. We em-ploy the software ‘TPoint’8 to model system flex-ure to achieve this value, in which we use 30 modelpoints that capture the discrepancy between in-tended and actual pointing for a variety of tele-scope positions. We had previously employed a∼150 point model, which was abandoned due tothe addition of new hardware. We noted that thesubsequent 30 point model replacement providessufficient pointing accuracy to enable the controlprogram to plate-solve and correct for pointing in-accuracy with only a small contribution to inter-frame overhead, and so have settled at 30 pointsfor convenience. We also have the ability to iden-tify and remove periodic error in the sidereal track-ing from within the mount control software, bymeasuring the correcting impulses provided by theautoguider. The periodic error is caused by ir-regularities in the rotation rate of the worm gearthrough one cycle, and so presents itself entirelyin RA. We fitted a 5th order polynomial with apeak-to-peak amplitude of 4.8′′, and subsequentlysaw improvements in the circularity of each star’s

8http://www.tpsoft.demon.co.uk/

T (◦C) D. C. (e− s−1) σD.C. (e− ADU−1)

-25 0.01 0.02-20 0.02 0.02-15 0.05 0.02-10 0.08 0.02-5 0.27 0.030 0.51 0.035 1.13 0.0310 2.25 0.0315 5.20 0.04

Table 1: Measured dark current for a range of chiptemperatures

Fig. 4.— (a) Top left: Counts as a function ofexposure time, with a linear fit through the expectedlinear region 20k - 40k ADU. (b) Bottom left: theresiduals of the above fit. (c) Top right: the photontransfer curve, with linear fit to the same 20k - 40kADU region. The deviation from linearity at ∼0 -10k ADU is due to the transition from the shot noiseto the read noise regime. (d) Bottom right: darkcurrent as a function of chip temperature

5

point spread function. As this is a German Equa-torial Mount, the OTA must ‘flip’ from one sideof the pier to the other when the tracked targetcrosses the meridian. This results in the observedfield being rotated 180◦ with respect to the opticsand camera.

2.3. Software

The hardware is controlled by a collectionof Windows-based software (MaxIm DL9 forimaging, TheSky610 for mount control, Focus-Max11(Weber & Brady 2001) for control of themicro-focuser) which are each controlled by ACP12,an observatory control suite that automates thecommand of each of the component software pack-ages through the ASCOM-standard driver layer.These programs are run on a 2.13GHz CPU Win-dows XP PC with 2Gb RAM. The site receivesits internet connection via satellite; connectionspeeds are good, with typical upload speeds fromthe OAM to the UK ∼ 250kBs−1, and operatinglatencies of around 60-100ms. We use Dimension4 from Thinking Man Software 13 to synchronisethe control PC’s clock to UTC via an NTP serverevery 10 seconds.

For a given observation or group of observationsin a night, ACP will first typically initiate an aut-ofocus via FocusMax, which searches for mag 4-7stars within a 2◦×2◦ region to focus with. Once fo-cused, ACP starts a 4s ‘pointing exposure’, whichit subsequently plate-solves via PinPoint14 usingthe GSC (Lasker et al. 1996) to determine thepointing error. A corrective slew is applied, andthe image sequence commences. Each full expo-sure, once read-out, is plate-solved once more tocheck the pointing. We apply a maximum pointingerror of 3′′, forcing a pointing update if a given im-age is determined to be misaligned by more than3′′. The benefits of doing so can be seen in Fig-ure 5, where the centroid deviation is significantlyreduced. This reduces the contribution to photo-metric noise from flat fielding errors; more on thisin section 3.3.

9http://www.cyanogen.com10http://www.bisque.com11http://users.bsdwebsolutions.com/ larryweber/12http://acp.dc3.com13http://www.thinkman.com/dimension414http://pinpoint.dc3.com/

The software tasks performed between expo-sures in a time-series run are:

1. The previous image is read out and plate-solved, from which the astrometric residual,pointing error (magnitude and direction),mean FWHM, true focal length, true imagecentre, and camera sky position angle arecalculated.

2. The centering is tested: if within the max-imum pointing tolerance, no slew update isperformed. If outside the maximum errortolerance, the auto-guiding is temporarilystopped, and a corrective slew is performed.

3. A guide star is selected from the guider im-age, with the star providing a signal-to-noise≥ 3 for the shortest exposure interval chosen.The exposure is started.

We typically see astrometric residuals from theplate-solve of ∼0.2′′. The total inter-frame over-head between exposure end and exposure start is∼ 15s. Each frame takes ∼ 2.5s to read-out anddownload, meaning the overhead is dominated bythe time taken to find the plate solution. The15s quoted refers to the case where no subsequentpointing update is required after the plate-solve.When a pointing update is required (pointing er-ror >3′′), 1-2s is added to the overhead time.

ACP also provides a web control interface,which made it an obvious choice for implementa-tion in a distance-learning, group-based educationscenario. When used by researchers, direct accessto the PIRATE control PC is achieved by remoteaccess of the control PC’s desktop. Students,however, are not given direct access to the maincontrol PC, and instead interact with the hard-ware through a proprietary server/client set-upthat limits their control. The server-side appli-cation (‘Switch Server’) provides access to relayswitches which are used to turn components onindividually, and also has a suite of extra ca-pabilities. These include: preventing the domefrom being opened in daylight hours; closing thedome in case of excessively high humidity or strongwinds; automatically cooling the two cameras ina monitored, staggered sequence; and checkingthe availability of the local internet connection,forcing a dome shut-down in case of lost internet

6

Fig. 5.— Position variations during a night with nospecified maximum pointing error (black circles) and anight with maximum pointing error set to 3′′ (crosses).

connection. Further information on the opera-tion of PIRATE’s Switch Server can be found inLucas & Kolb (2010).

3. Data Acquisition, Reduction & Com-

missioning Issues

3.1. Observing Strategy

On a given night, the user cools the main imag-ing camera down automatically using the SwitchServer to a target temperature of ∼ 40◦C belowthe current ambient temperature. Consequentlythe operating temperature of the CCD varies sea-sonally, usually between -10◦C in the Summer and-20◦C in the Winter. This ensures the camera isalways operating at the lowest available temper-ature. Once the target temperature is achieved,ACP is started and the dome opened around sun-set. ACP’s automatic sky flat procedure is theninitiated. This procedure automatically scales ex-posure times to achieve a target peak count of ∼33k ADU in each filter’s flat fields. During the flatprocedure, sidereal tracking of the mount does notoccur. The flat fields are dithered so that any stel-lar sources present in the sky flats can be removed

upon combination of each flat into the master flatfield. Once the flat fields are obtained, the domeis then shut whilst waiting for the end of Nauticaltwilight (ACP allows observations through Astro-nomical twilight). During this time, the user takescalibration frames. 50 bias frames are taken first,to assist in completely flushing the chip of anyresidual charge from the flat fields. Then darkframes are taken in batches of 10 for each antic-ipated exposure time to be used for the scienceframes throughout the night. Typically these ex-posure times are 30, 45, 60, 90, and 120s. Thedome is then reopened in time for the end of nau-tical twilight, and the observing schedule for thenight commences.

3.2. Photometric Reduction, Performance,

and Light Curve Generation

The data is available in real-time via directFTP to the control PC, and users with direct ac-cess to the control PC may also inspect each image‘live’ after being read-out from the camera. Eachmorning the previous night’s science and calibra-tion frames are then transferred from the OAM toa local archive at the OU; the user may obtain theframes the following day. Reduction (in the re-search case) is performed by a custom IDL scriptthat operates interactively, performing file renam-ing and sorting tasks and determining the pres-ence of calibration frames, subsequently compilingIRAF 15 scripts that make use of the DAOPHOTpackage (Stetson 1987). The DAOPHOT out-put is then fed back into an IDL script that per-forms differential photometry and generates the fi-nal light curves for each object in the frame. Whenused for teaching, the students do not make use ofsuch scripts, and must process their frames ‘man-ually’ using MaxIm DL.

The initial IDL script first compiles a list ofavailable light frames from the previous night, andthen iteratively displays each one in turn, askingthe user to reject or accept each frame. This allowsfor the rejection of frames with defects, such as alack of useable point sources in the frame, trails,or excessive pointing error. The user then accepts

15IRAF is distributed by the National Optical AstronomyObservatory, which is operated by the Association of Uni-versities for Research in Astronomy (AURA) under coop-erative agreement with the National Science Foundation.

7

or rejects a final list of rejected frames, producinga master file list on acceptance. Lists are gener-ated for each calibration frame type; discriminat-ing flat fields by filter. If flat fields matching thechosen filter of the light frames do not exist, noflat fielding is performed. An IRAF script is gen-erated for the task CCDPROC, which performsstandard CCD calibrations and then executes therequisite post-calibration rotation commands forpre-flip light frames.

The IDL routine first proposes a master frameaccording to a simple statistical test performedon all frames. The master frame proposed willbe the frame that exhibits the highest standarddeviation from the quartile of frames that havethe lowest mean values. The mean test ensuresa low sky background (high sky backgrounds areoften indicative of Cirrus scattering), whilst select-ing a frame with high standard deviation ensuresmany strong point source fluxes well above thebackground. The user can over-ride the proposedframe selection and select their own frame as themaster frame. An alternative method would beto determine the frame with the minimum aver-age FWHM, and designate this the master framedue to the superior seeing. This is not currentlyimplemented.

Once defined, we use DAOFIND on the masterframe (making use of a user-given average frameFWHM) to find all sources above a 6σ threshold.An initial PHOT task run on the master framedetermines the master coordinate list, with onlysources that do not suffer a PHOT task failureflag going into the master coordinate list. A lin-ear transformation is then deduced between themaster frame coordinates and those of all otherframes, and this is used to generate the final co-ordinate lists for each frame. Final photometry isstandardly performed with an aperture 3× the in-put (master frame) FWHM, though the aperturesize can be changed if the data warrants (e.g. inthe case of close companions to the target thatlie within or partially contaminate the measuringaperture).

Once the instrumental photometry is obtained,it is fed into another IDL script which places thewhole night’s photometry in memory as a datacube. Mid-exposure time in UTC then is con-verted to HJD for all frames. We note the prefer-ence of using BJD over HJD, which can differ by

as much as 4s (Eastman et al. 2010), as well ascalculating BJD from TT instead of UTC, whichcurrently differ by 66.184s as of July 2011. We willbe incorporating these changes into IDL scripts inthe near future. The night’s data is then investi-gated for PHOT task failures. Light curves thatsuffer a photometry failure in one or more framesare rejected from the cube. In case of bad framesstill present in the data (which would lead to anexcessive number of light curves being cut), we in-troduce a parameter (to be set by the user) definedas: Nstar/a, where a frame is removed if it con-tains a fraction of photometry failures > Nstar/a.Here Nstar is the number of stars in the frame, anda is a free parameter. We find a value of a = 10 tousually be sufficient. The cube is then sorted bymagnitude, and an iterative ensemble compilationprocedure begins. We employ a similar method-ology to that of Burke et al. (2006), where foreach star, all the corresponding comparison lightcurves are sorted according to a light curve ‘fig-ure of merit’. Each comparison light curve is nor-malised to unity, and these light curves’ RMS (i.e.the standard deviation about the mean) is used asthe figure of merit. Starting with the best ratedlight curve (i.e. lowest RMS) for a given star,each subsequent light curve in the list is iterativelyincluded in the ensemble, using inverse varianceweights, where the weights derive from the IRAFphotometric uncertainties. Note that the ensem-ble never contains the flux from the target star,as the target star is excluded from the list of lightcurves. The following refers to the production ofthe final light curve for one star only. The en-semble flux, Ej , for a given frame j, containingcomparison stars i = 1...N is compiled as thus:

Ej =

∑Ni=1 ωijFij∑N

i=1 ωij

Where Fji are the stars’ individual fluxes, and theindividual weights are given by:

ωij =1

σ2ij

The uncertainty in the ensemble is given by thestandard deviation of the weighted average:

σEj=

[

N∑

i=1

wij

]−1/2

8

With each new star added to the ensemble, thedifferential target star light curve d∗ = (d1 . . . dN ),

where d∗,j =F∗,j

Ej, j = 1 . . .N , is calculated from

the recorded flux of the target star, F∗,j , and cal-culated ensemble flux, Ej . The light curve d∗ isthen normalised by dividing through by its me-dian value, and the RMS is computed. The fi-nal light curve, d∗,fin makes use of the ensemblethat produces the lowest RMS. Each star there-fore has its own unique comparison ensemble. InFigure 6, we display the results of this methodfor the stars in the WASP-12 field on two con-secutive nights with approximately 30% & 39%moon illumination, at a separation of ∼ 142◦

and ∼ 132◦ from the moon respectively, compar-ing USNO-B1.0 (Monet, et al. 2003) magnitudeswith the RMS precision of PIRATE R band lightcurves from 60s exposures. The observation runsof 22/11/2009 and 23/11/2009 contain light curvesof 759 and 730 objects, each consisting of 359 and321 exposures respectively. We see from this plotthat sub-percent precision is available for magni-tudes ∼13.5 and less with an operating cadenceof 76s (including the aforementioned overheads).We include for comparison the contribution to theuncertainty from photon noise alone.

Fig. 6.— Light curve RMS deviation as a function ofUSNO-B1.0 R magnitude for the nights of 22/11/2009and 23/11/2009, 60s exposures, WASP-12 field

Fig. 7.— WASP-12b transit observation (R band)from 23/11/2009. Residuals from the model fit areshown offset by 0.05.

As an example, this particular method of lightcurve compilation was used in the productionof light curves of the transit of WASP-12b, forwhich 6 nights of data were captured during themonths of October-November 2009, the details ofwhich can be found in Haswell et al. (in prepara-tion). We show the light curve from the nightof 23/11/2009 in Figure 7, which has a post-fit residual RMS of ∼2.6 mmag. This data wasused to confirm the published ephemeris for anHST visit investigating the exosphere of WASP-12b (Fossati et al. 2010). The displayed modellight curve is the published model light curve (fitto the data from the discovery paper) using the pa-rameters of Hebb et al. (2009), and analytic lightcurve solutions of Mandel & Agol (2002), usingthe limb-darkening coefficients of Claret (2000).

We find that the entire light curve’s RMS isan unsuitable figure of merit for determining howaccurately a light curve has captured an object’sinherent variability in certain cases. Substantialout of transit coverage of WASP-12 (out to phase1.4 (not shown)) meant that the RMS is an ac-ceptable figure of merit for the observations shownin Fig. 7. Obviously, if one has prior knowledgeof the transit ephemeris, as with these WASP-12observations, solely the out-of-transit light curveshould be used to calculate the figure of merit.In the case of the follow-up of candidate plan-ets, or for other purposes (e.g. cluster surveys),the ephemeris might not be well constrained. The

9

method’s success is therefore subject to the ampli-tude, timing, and character of the star’s intrinsicvariability.

When tackling variable stars with ampli-tudes over and above ∼10%, the procedurewill often select very faint, non-varying starsfor inclusion within the ensemble, so long astheir light curve RMS is less than the intrin-sic variability of the target star. We showedin Fig. 1 the light curve of a larger amplitude(∼20%) variable, 1SWASPJ163420.90+424433.4(Norton et al. 2007), taken on 07/06/2010 byOU undergraduate students as part of their prac-tical module. The data were reduced separatelyfor inclusion here, with 4 comparison stars cho-sen by hand for combination through inverse-variance weighting into a comparison ensemble,as the RMS-minimisation method does not workfor this target. We also find the procedure inade-quate in some lower amplitude situations, such asa typical hot jupiter transit with insufficient out-of-transit coverage. For example, a light curvethat captures 1 hour in-transit, egress, and 1 hourout-of-transit approximates a step function; andis bimodal in d∗, a scenario for which the RMS isa poor figure of merit.

To combat this, we are developing an exten-sion of the technique that breaks each comparisonlight curve into multiple segments, and performs alinear regression in each segment, producing a se-lection of relevant statistics: segment RMS, resid-ual segment RMS (after subtracting the fit), seg-ment median value. By careful analysis of eachlight curve’s segment statistics, as well the statis-tics of the segments as a whole (e.g. range orstandard deviation of the segment medians), itwill be possible to identify variable stars for ex-clusion from a comparison ensemble, whilst simul-taneously allowing for any variable star to have anoptimal comparison ensemble automatically deter-mined. This technique will facilitate serendipitousPIRATE discoveries of previously unknown vari-able stars.

3.3. Flat Field Inaccuracy and Optical Re-

sponse Calibration

The systematic errors of flat fields as opticalresponse calibrators are often overlooked in theapplication of differential photometry, as modernauto-guiding systems are sufficiently good at lock-

ing the stellar point-spread functions to fixed po-sitions on the chip for the duration of a time se-ries. As the observer is only interested in time-variability, systematic offsets in the magnitudezero-points of any two stars in the field can beignored. PIRATE’s GEM executes a ‘pier-flip’as the tracked target crosses the Meridian. Thismoves the OTA to the other side of the pier; andinverts the image of the stellar field with respect topre-flip frames. This operation effectively moveseach stellar point spread function to an entirelydifferent part of the focal plane. To maintain con-tinuity in the flux ratio between two stars acrossthe pier flip, the vignetting must be determinedperfectly. In reality, most flat fields suffer frominaccuracies at the 10−2 level (Manfroid 1995).We therefore expect and indeed see varying fluxratios between two objects in the field across thepier-flip. We term this effect a ‘Light Curve Dis-continuity’ (LCD). Several mechanisms contributeto the flat field systematic errors:

• Uneven illumination from the source. Thiscan occur in sky flats, even when they imagethe sky null point (Chromey & Hasselbacher1996) as PIRATE’s automatic proceduredoes. It will also occur in dome flats.

• Physical changes in the hardware betweentaking flat fields and science images thatmodify the response function of the system.In the case of PIRATE Mark 1, it is sus-pected that primary mirror ‘flop’, which oc-curs despite the presence of mirror locks,and/or flexure in the micro-focuser assem-bly changes the vignetting function. It mayalso lead to the misalignment of the focalplane with the CCD surface, meaning con-sistent focus across the field of view is notachievable. In this scenario, the FWHM ofthe stellar profiles will be a function of chipposition, and the measuring apertures usedfor photometry would see different levels offlux leakage due to the changed stellar profileafter a pier-flip.

• Scattered light in the optical tube assembly.Off-axis, unfocused stray light due to insuffi-cient baffling can contribute to the flat fieldexposure. Scattered light is additive. If in-cluded in the flat field, this additive light

10

is incorrectly used in the multiplicative flatfield calibration.

We have been unable to develop a wholly effec-tive procedure for self-consistently calibrating theLCD effect without investing much of the nightin taking calibration observations. We thoroughlyexplored these possible approaches:

We used data from the night of 23/07/2009,during which the moon had 3.2% illumination, sowe might expect sky background gradients to belimited. The data were processed in the usualmanner, and investigated separately as two groupsof ‘pre-flip’ and ‘post-flip’ frames. The frames ineach group were median-combined, and a 6th orderLegendre polynomial was fit to the sky backgroundin each of the two (pre and post) resultant framesusing the IRAF task ‘IMSURFIT’. Note that thepre-flip frames were rotated 180◦. We denotethe background fits by Apre(x, y) and Apost(x, y),where x and y are image co-ordinates correspond-ing to positions on the sky (not pixel co-ordinates,due to the aforementioned rotation of the pre-flipframes). A simple ratio of the pre and post-flipbackground fits reveals any discrepancy in the skybackground for a given star position. We denote

this ratio as Amap(x, y) =Apre(x,y)Apost(x,y)

. We show

Amap in Fig. 8. We note that it has structure pre-dominantly in the x direction, and displays a peak-to-peak variation of ∼ 7%. The (non-differential)light curves of each of the N stars (here N = 517),which we denote by Fi(t), where i = 1...N weremedian-combined to create an approximation (tofirst order) of the sky transparency, F . We assumeeach star remains at a fixed position in the image(xi, yi) for the duration of the observing run. Forall of the light curves (including the transparencyfunction), the fluxes were averaged over time forpre-flip frames and also for post-flip frames, pro-ducing a single pre-flip and post-flip flux valuefor each light curve. We therefore use Fpre,i andFpost,i to refer to these time-averaged values. Wedefine the LCD of a star to be:

∆Fi =Fpre,i/Fpost,i

Fpre/Fpost

We now have the ability to assess the applica-bility of the background ‘map’; if the shape of thesky background (after flat-fielding) provides an es-timation of any residual error in the flat-fielding

process, then it should correlate well with the ob-served LCDs. To compare the map with the LCDs,we fit a 2nd order polynomial through least squaresregression to the irregularly gridded ∆Fi values.The result of this fit can be seen in Fig. 9. As canbe seen, far from correlating well, the pattern ofthe LCDs also exhibits greatest deviation in thex direction, but with opposing orientation. Thestructure of the map in Fig. 8 suggests a fixedand constant light source co-moving with the op-tical tube assembly, inducing the same backgroundstructure in both pre and post-flip frames, whichis then amplified when the pre-flip frames are ro-tated. One would expect any transformation ofthe vignetting function to be recorded in the skybackground, as this gives a continuous indicationof the vignetting function. However, the presenceof any scattered light in either the science framesor flat fields renders the true vignetting functionfor each side the pier flip unrecoverable.

The scattered light structure present in the sci-ence frames could also have been present in theflat field, and thus the LCDs introduced to thedata through the flat fielding process alone, butthe un-flat fielded data shows the same LCD struc-ture seen in Fig. 9. Given the dark night (onenight after a new moon) for this data set, it isapparent that scattered (additive) light is likelyalways present, and greatly reduces the accuracyof standard flat-fielding and sky-flat procedures;the GEM simply highlights this inaccuracy.

The data of 23/07/2009 exhibit strong struc-ture in the LCD map, and is hence a good demon-stration of the problem. A more typical structurecan be seen in Figs. 10 and 11, from the night of23/11/2009, WASP-12 field. The polynomial fitto the LCDs exhibits a smaller peak-to-peak am-plitude of ∼2%, and there is significant scatter inthe residuals of this fit to ∆Fi.

We conclude that background fitting, sky flats,and even flat fielding are all ineffective at negatingthe percent-level LCD effect seen in PIRATE Mk.1. Instead, there are two routes to achieving cor-rect photometric calibration across the pier-flip.

The first of these involves creating a photomet-ric super-flat. This is discussed in detail in Boyle(2007), Grauer et al. (2008), Manfroid (1995),Regnault et al. (2009), and Selman (2004). Atits most simple, this involves observing a set ofstandard stars at different positions in the field of

11

Fig. 8.— Sky background map produced by fitting tothe sky background for all pre-flip frames and all post-flip frames, then taking the ratio of these two fits. Thestructure of this map suggests there may have been afixed scattered light source with respect to the OTA.

Fig. 9.— 2nd order polynomial fit to the flux deficits,∆Fi

Fig. 10.— As for Fig. 8, but for the night of23/11/2009

Fig. 11.— As for Fig. 9, but for the night of23/11/2009

12

view, and determining the position-dependent re-sponse to the standards in order to build the large-scale, low-frequency optical response of the systeminto the final flat field. Sky flats or dome flatsare observed in tandem, with polynomial fits totheir large scale variation made in order to ‘flatten’them. This retains the small scale, high frequencyintra-pixel response, which can then be combinedwith the standard star-determined low frequencyresponse function to create a ‘photometric super-flat’. In constructing the large-scale component,it is preferable to make multiple dithered obser-vations of a cluster containing many standardsthat span the field of view in order to reduce thenumber of frames needed. As we expect the re-sponse to be different from one side of the pier toanother, this involves observing the same clustertwice, once in the eastern sky, and again later inthe western sky.

The second method involves lending a degreeof freedom to the normalisation of each lightcurve section (pre & post-flip), shifting the fluxlevels up and down to minimize the χ2 of amodel fit. This method has its limitations in theextent of fore-knowledge required to work suc-cessfully. In the case of SuperWASP follow-up,the Markov-Chain Monte Carlo fitting routine ofCollier Cameron et al. (2007) can take the lightcurves from each side of the pier as separate inputlight curves and include the normalisation factoras a free parameter in the fitting procedure. How-ever, this algorithm works with ‘prior knowledge’of the light curve, in that its initial parametersfrom which the MCMC routine iterates are takenfrom the SuperWASP light curves. This prior ex-pectation of the light curve’s model parameters(including, crucially, the transit ephemeris) as-sists with the cross-flip normalisation. If there isno prior knowledge of the light curve, adjustingflux levels is almost certainly perilous. Take, forexample, the worst case scenario of a pier-flip oc-curring mid-way through ingress or egress. In sucha scenario it would prove difficult to determine ifthe flux deficit across the pier-flip is instrumentalor astrophysical. Great caution is called for ininterpreting such light curves.

4. Summary and discussion

PIRATE is a remotely-operable telescope facil-ity built primarily from readily available ‘off-the-shelf’ components, used in both education and re-search. In its education role, it has successfullysupported small groups of simultaneous studentusers in their efforts to observe and classify vari-able stars in the SuperWASP archive. We showedthat relatively inexpensive equipment can play auseful role in modern astrophysical research, in-cluding high-precision time-series photometry re-quired for transiting Jupiter-size exoplanets. ThePIRATE system employs a GEM which providesexceptional stability and all-sky pointing accuracyfor its cost. The pier-flip introduces LCDs whichare difficult to correct for by calibration proce-dures alone; the best strategy is to treat pre- andpost-flip light curves as separate observing runs.

The OTA and camera combination describedhere awaits re-assembly in a neighbouring domeat the OAM. We note that the combination of aParamount ME and the affordable Celestron C14can be found in many installations world-wide.

The use of a remotely-operable telescope in theOU module S382 16 proved the feasibility of de-ploying such complex hardware in real-time to cre-ate an inspirational teaching tool for distance ed-ucation. The great enthusiasm of the students in-volved, and the quality of the acquired data and ofthe scientific reports generated at the end of theproject demonstrate the success of the PIRATEteaching project. Use by level 2 (year 2) studentsis being implemented, and use by level 1 (year 1)OU students is currently under consideration.

PIRATE has very recently undergone an OTAchange to a PlaneWave Systems CDK17 f/6.8 Cor-rected Dall-Kirkham astrograph, which provides abigger field of view and larger aperture (0.425m).Future work will involve linking ACP to the domecontrol via the ASCOM driver layer, allowing forfull scripting of a night’s events; adding a levelof autonomy to the observing process. Use ofACP Scheduler17 with the current weather sta-tion hardware and switch-server intelligence wouldeffectively allow PIRATE to run in a fully au-tonomous, robotic mode, which will be explored

16http://www3.open.ac.uk/study/undergraduate/course/s382.htm17http://scheduler.dc3.com/

13

as an operation mode for research purposes. Thereal-time remote operation will remain the onlymode of operation for teaching applications, how-ever, in order to make the learning process asinteractive as possible.

We thank the referee for their constructiveand insightful comments, which improved the pa-per. We are grateful for the continued support ofBaader Planetarium towards optimising the domefor the PIRATE facility, and the on-site supportprovided by staff at the OAM. SH acknowledgesthe support of an STFC studentship.

REFERENCES

Boyle, R. P., 2007, BaltA, 16, 459

Burke, C. J. et al., 2006, AJ, 132, 210

Burwitz, V. et al. 2010, CBET, 2124,1

Chromey, F. R. & Hasselbacher, D. A., 1996,PASP, 108, 944

Claret, A., 2000, A&A, 363, 1081

Collier Cameron, A. et al. 2006, MNRAS,373:799

Collier Cameron, A. et al. 2007, MNRAS, 380,1230

Denny, R., 2011, A Web-Remote/Robotic/Scheduled AstronomicalData Acquisition System, Poster presented at:Telescopes From Afar, February 28, Hawaii

Eastman, J., Siverd, R., & Gaudi, B. S., 2010,PASP, 122, 935

Fossati, L. et al., 2010, ApJ, 714, 222

Grauer, A. D. et al., PASP, 120, 992

Haswell, C. A., 2010, Transiting Exoplanets, Cam-bridge: Cambridge University Press, 80

Hebb, L. et al., 2009, ApJ, 693, 1920

Howell, S. B., 2000, Handbook of CCD Astron-omy, Cambridge: Cambridge University Press

Kolb, U. et al., 2010, Electronic Resourcesfor Teaching and Learning, A. Norton (ed.),http://pirate.open.ac.uk/publications.html

Lasker, B. M. et al. 1996 VizieR Online Data Cat-alog, II/143A

Lucas, R., & Kolb, U., 2010, arXiv:1012.3058v1

Mandel, K. & Agol, A., 2002, ApJ, 580, 171

Manfroid, J., 1995, A&AS, 113, 587

Monet, D. G. et al., 2003, AJ, 125,984

Norton, A. J. et al., 2007, A&A, 467, 785

O’ Donovan, F. T. et al., 2006, ApJ, 644, 1237

Pietsch, W., 2008, Astronomische Nachrichten,329, 170

Pietsch, W., 2010, Astronomische Nachrichten,331, 187

Pollacco, D. L. et al., 2006, PASP, 118, 1407

Regnault, N. et al., 2009, A&A, 506, 999

Selman, F. J. 2004, Proc. SPIE, 5493, 453

Stetson P. B., 1987, PASP, 99, 191

Tamuz, O., Mazeh, T., & Zucker, S. 2005, MN-RAS, 356, 1466

Weber, L., Brady, S., 2001, Minor Planet Ama-teur/Professional Workshop Proceedings

Zissell, R. E., 2000, JAAVSO, 28, 149

This 2-column preprint was prepared with the AAS LATEXmacros v5.2.

14