SOUSA: the Swift Optical/Ultraviolet Supernova Archive
Peter J. Brown, Alice A. Breeveld, Stephen Holland, Paul Kuin, Tyler Pritchard
aa r X i v : . [ a s t r o - ph . H E ] J u l SOUSA: the Swift Optical/Ultraviolet Supernova Archive
Peter J. Brown • Alice A. Breeveld • Stephen Holland • Paul Kuin • Tyler PritchardAbstract
The Ultra-Violet Optical Telescope on the Swiftspacecraft has observed hundreds of supernovae, cover-ing all major types and most subtypes. Here we intro-duce the Swift Optical/Ultraviolet Supernova Archive(SOUSA), which will contain all of the supernova im-ages and photometry. We describe the observation andreduction procedures and how they impact the finaldata. We show photometry from well-observed exam-ples of most supernova classes, whose absolute magni-tudes and colors may be used to infer supernova typesin the absence of a spectrum. A full understanding ofthe variety within classes and a robust photometric sep-aration of the groups requires a larger sample, whichwill be provided by the final archive. The data fromthe existing Swift supernovae are also useful for plan-ning future observations with Swift as well as futureUV observatories.
Keywords supernovae; ultraviolet
Peter J. BrownGeorge P. and Cynthia Woods Mitchell Institute for FundamentalPhysics & Astronomy, Texas A. & M. University, Department ofPhysics and Astronomy, 4242 TAMU, College Station, TX 77843,USA; [email protected] A. BreeveldMullard Space Science Laboratory, University College London,Holmbury St. Mary, Dorking Surrey, RH5 6NT, UKStephen HollandSpace Telescope Science Center, 3700 San Martin Dr., Baltimore,MD 21218, USAPaul KuinMullard Space Science Laboratory, University College London,Holmbury St. Mary, Dorking Surrey, RH5 6NT, UKTyler PritchardDepartment of Astronomy and Astrophysics, The PennsylvaniaState University, 525 Davey Laboratory, University Park, PA16802, USA
Supernova (SN) explosions have been observed in theultraviolet (UV) since 1972 with the Orbiting As-tronomical Observatory (OAO-2; Holm et al. 1974).In the decades since, UV observations have beenmade by the International Ultraviolet Explorer (IUE;Cappellaro, Turatto, & Fernley 1995), the Hubble SpaceTelescope (HST; e.g. Kirshner et al. 1993; Millard et al.1999; Baron et al. 2000; Foley et al. 2012), the As-tron Station (Lyubimkov 1990), the Galaxy EvolutionExplorer (GALEX; Gal-Yam et al. 2008), and XMM-Newton’s Optical Monitor (OM; Immler et al. 2005).Atmospheric absorption requires observing from space,so the number of SNe observed in the UV is much lowerthan in the optical (see Panagia 2003; Foley et al. 2008;Brown 2009 for reviews).
Fig. 1
Histogram of the number of SNe observed in the UVeach year. Since 2005 the Swift UVOT has observed nearlyten times more SNe than the other missions combined.
Fig. 2
A timeline of Swift SN observations in the mid-UV ( ∼ =72 km/s/Mpc:Freedman et al. 2001). The y-axis on the right side shows the distance at which that brightness is observable for a limitingmagnitude of 20. The Swift Ultraviolet/Optical Telescope (UVOT;Gehrels et al. 2004; Roming et al. 2005) began observ-ing SNe in 2005 (Brown et al. 2005). Since then it hasobserved over 300 SNe. This dramatic increase in thenumber of SNe observed in the UV is shown in Fig-ure 1. The individual SNe are listed on the Swift SNwebsite . Many have been published already, includ-ing samples in Brown et al. (2009), Milne et al. (2010),Brown et al. (2012a), and Pritchard et al. (2013). Ofthese, only the latter uses the latest zeropoint cali-bration and time-dependent flux sensitivity correctionof Breeveld et al. (2011). For a better comparison ofthe growing sample, we are analyzing or reanalyzingall of the UVOT SN data and creating the Swift Op-tical/Ultraviolet Supernova Archive (SOUSA). The fi-nal archive will include the imaging data as well asthe photometry. The Swift/UVOT also has opticaland ultraviolet grisms to perform low resolution spec-troscopy (Roming et al. 2005; Kuin et al. 2009, Kuin etal. 2014, in preparation). The grisms have been usedfor SNe as well (Bufano et al. 2009; Foley et al. 2012;Bayless et al. 2013; Margutti et al. 2014), but we focusfor now on the photometry. This article is intendedto introduce the archive and describe the photometryproducts that will be released via the Swift SN web-site. Data for most SNe previously published by us,notably Brown et al. (2009) and Brown et al. (2012b),are already available and more will be added over thecoming months. In addition, we provide some of thescripts used to reduce the data and parse the outputfiles. In Section 2 we describe the Swift observationsand in Section 3 we detail the photometric reduction.In Section 4 we use some of the photometry to showhow the different SN classes differ in UV colors and ab-solute magnitudes and how it can be used to plan futureobservations with Swift and future UV observatories. The Swift spacecraft (Gehrels et al. 2004) was designedfor the detection and rapid observation of gamma raybursts (GRBs). It has a special capability whereby atarget position can be uploaded to the spacecraft forimmediate observation whenever viewable, supersedingthe previously planned targets. This allows newly dis-covered SNe to be observed within hours of discovery.The data is regularly sent down from the spacecraft andusually available from the Swift website several hours http://swift.gsfc.nasa.gov/docs/swift/sne/swift sn.html http://swift.gsfc.nasa.gov/cgi-bin/sdc/ql? Fig. 3
Top panel: the effective areas of the UVOT filtercurves. Bottom panel: UV/optical spectra of SNe fromHST, normalized to have the same optical flux to show thediversity in the UV flux levels. later. This allows for rapid feedback on the UV bright-ness of a new target to inform the planning of futureobservations, which is usually created just one or twodays in advance. Swift observes several targets dur-ing its 90-minute orbit, so the overhead on individualtargets is low compared to other space observatories.This allows many relatively short observations to bescheduled to obtain better time coverage than usuallypossible in the UV. These unique features make Swiftan excellent observatory for transients such as SNe.With a few notable exceptions (the GRB-SN 2006aj:Campana et al. 2006 and the shock breakout of SN 2008D:Soderberg et al. 2008) Swift does not discover SNe.SNe discovered elsewhere are proposed as targets ofopportunity (ToOs) and, if approved, subsequently ob-served. Because most are proposed one by one (with theexception of some guest investigator programs), thereis not a uniform selection criteria. We have not triedto obtain an unbiased sample of all SNe but to obtainobservations of SNe across all types and host galaxyenvironments as much as possible. Because of the UVfaintness of many SN types and the relatively smallaperture of UVOT, most are very nearby SNe, with aredshift of z less than 0.02 (but we will discuss limitson this later). SNe without significant extinction or galaxy contamination are usually preferred. SNe havetypically been observed with a two-day cadence dur-ing the early phases and spreading out as the SN agesand changes less with time. After the SN has faded,an observation of the host galaxy is requested as a lowpriority target that can be filled in as the schedule al-lows. The excellent temporal coverage is reflected inFigure 2 which shows preliminary data on most of theSNe observed over Swift’s lifetime.For clarity we will now define a few of the termswe use in the observing and data analysis. An “ob-servation” typically refers to one or more exposuresscheduled as a set in the pre-planned science timeline(PPST) or executed by the spacecraft as the result of anuploaded command to immediately observe an “auto-mated target” (AT). An observation may include mul-tiple exposures in different filters and may span severalorbits. As Swift is in a low Earth orbit, locations inthe sky are not observable for large continuous chunksof time (typically not more than 40 minutes). Long ob-servations are broken up into “snapshots”, continuousviewing periods during which exposures are taken in apredetermined sequence of UVOT filters.The Swift UVOT observations usually utilize thesix main broadband filters (see Roming et al. 2005 andPoole et al. 2008 for details). The UVOT filters arecompared to SN spectra in Figure 1. The white (clear)filter is not used due to its broad passband which is hardto flux calibrate for objects of different or varying spec-tral shape. The UVOT filter mode determines whichfilters will be used and the exposure times in each. Fora scaled mode, the exposure times in each filter arecalculated based on the exposure time ratios given bythe mode and the calculated length of the snapshot.For planned targets the snapshot length is calculatedby the spacecraft based on the planned time on target.For ATs not in the planned timeline the snapshot lengthis calculated as the time until an observing constraint isreached. Exposures in some filters may not be observedif the full, planned snapshot is not observed due to ahigher merit AT becoming visible during the snapshot.If a higher merit AT causes the snapshot to begin late,the exposure times in the filters will be calculated basedon the time remaining and all requested filters will beobserved (albeit for a shorter than planned amount oftime). ATs can also be superseded by higher merit tar-gets in the PPST, resulting in a truncated snapshotwhich is shorter than what the spacecraft would cal-culate based on the observing constraints. To get allfilters for prompt SN observations “unscaled” modescan be used. In these modes the filters are observed fora set amount of time so that all filters can be completedwithin the snapshot (whose length can be determinedbeforehand). Here we highlight the recommended modes for SNphotometry: the preferred PPST mode for red objects(like most SNe) is the scaled mode 0x223f, which hasthe six UVOT filters (all of the broadband filters exceptfor white) with the following approximate time frac-tions (uvw1,u,b,uvw2,v,uvm2) (17,8,8,25,8,33). Thepreferred AT mode is the unscaled mode 0x0270 withtimes of (uvw1,u,b,uvw2,v,uvm2) (160 s, 80 s, 80 s, 320s, 80 s, 280 s) for a 1000 s snapshot. If the snapshot islonger than 1000 s the remaining time is spent in theuvm2 filter, usually valuable for UV-faint targets.UVOT uses a photon-counting detector. As such,the count rates from sources brighter than ∼
13 magcannot be accurately measured in the normal modesand photometric procedures. Special “hardware-window”modes read out a smaller portion of the detector. Thefaster readout means higher temporal resolution so thatthe count rate can be determined for sources as brightas ∼ ∼ . The archive is searched using the SN positionso that all images of the field are obtained regardlessof the target identification number (TID). Sometimesmultiple TID numbers are used to differentiate differ-ent programs or observations made of the galaxy ratherthan targeting the SN. We use the sky images whichare shifted and rotated into the World Coordinate Sys-tem. Each fits file (suffix .img) contains all exposuresin a given filter for that observation corresponding toa unique observation identification number (OBSID). For bright SNe like 2011fe (Brown et al. 2012b), in-dividual exposures are used. Otherwise, all full-fieldexposures within a single OBSID are coadded into asingle image for that epoch. Exposures using differentframe rates are not coadded because the coincidencelosses (and corrections) are different. Images are ex-amined so that individual exposures that show imageartifacts (such as streaking stars due to the spacecraftmoving during the exposure) can be excluded. We do http://heasarc.gsfc.nasa.gov/cgi-bin/W3Browse/swift.pl See http://archive.stsci.edu/swiftuvot/file formats.html for amore detailed description. not require the aspect correction to have been success-fully performed but manually correct or exclude anyimages that are offset from the rest. A new fits fileis created for each filter with extensions including thesummed image from each epoch. A separate fits file iscreated for the images designated as templates whichwere observed before the SN exploded or after it hadsignificantly faded.3.2 PhotometryThe photometric reduction follows the same basic out-line as Brown et al. (2009).
HEASOFT (currently version6.13 corresponding to Swift release 4.0) is used to per-form the photometry using uvotsource . It is called by uvotmaghist which operates on a list of images or a sin-gle fits file with multiple image extensions. It createsas output a fits table of the extracted and calculatedvalues. The calibration database (CALDB) version re-leased 2013-01-18 is used, which includes the zeropointsin the UVOT Vega and AB systems (Breeveld et al.2011). The default photometry in SOUSA is on theUVOT-Vega system, while conversion to the AB sys-tem is straightforward using the zeropoint differencesin (Breeveld et al. 2011). Counts in the source regionare measured using 3 ′′ and 5 ′′ apertures. The coinci-dence loss correction for the source is determined us-ing the 5 ′′ aperture. The coincidence loss is computedseparately for the background. The source counts areobtained by subtracting the coincidence-loss correctedbackground (scaled for the size of the aperture) fromthe corrected total counts in the source aperture. Thecount rates are also corrected for the time-dependentloss in sensitivity (Breeveld et al. 2011) which amountsto 1% per year in all filters except v which is now cor-rected by 1.5% per year . Necessary corrections tothe exposure time include subtracting the time dur-ing which the frames are being downloaded and rareanomalies. If individual images are used, a correction is made fordifferences in the large scale sensitivity (Breeveld et al.2010). When photometry is done on coadded images(where the source does not correspond to a unique de-tector position) the correction is not done, and a sys-tematic uncertainty of 2.3% of the count rate is added inquadrature to the photometric error (Poole et al. 2008).The above steps are done for each of the SN imagesas well as the summed template image, giving correctedcount rates in the 3 ′′ and 5 ′′ apertures. These are taken http://heasarc.gsfc.nasa.gov/docs/heasarc/caldb/swift/docs/uvot/uvotcaldb throughput 02b.pdf http://swift.gsfc.nasa.gov/analysis/uvot digest/timing.html from the fits file output of uvotmaghist and then cor-rected for the galaxy using our own scripts as follows.The count rates from the galaxy template in the appro-priate aperture are subtracted from the count rates inthe SN images. This is done before the aperture cor-rection since the SN is a point source and the galaxybackground is likely not. For the 3 ′′ aperture the galaxycount rate is subtracted and then the aperture correc-tion from uvotmaghist (calculated using the averageUVOT PSF in the CALDB) is applied.In choosing the aperture size there is a trade-off be-tween maximizing the signal to noise ratio for faint ob-jects (Li et al. 2006; Poole et al. 2008) and the uncer-tainties of the correction to the full photometric aper-ture. We calculate photometry using 3 ′′ and 5 ′′ aper-tures. A 1.5% uncertainty on the flux is added inquadrature to the error when using the 3 ′′ apertureto account for variations in the point spread function(Breeveld et al. 2010). For each photometric point wechoose the aperture with the smallest magnitude error.Upper limits are calculated from the number ofcounts required to achieve a signal to noise (S/N)of three when accounting for the statistical error onthe source counts and the errors on the backgroundand galaxy counts. Using a Poisson error rather thanthe binomial error appropriate for photon counters(Kuin & Rosen 2008) makes this analytically possibleand is a good approximation in the low count regime.The count limit is corrected for the aperture size, largescale sensitivity, and time dependent sensitivity andconverted to a magnitude. This limiting magnitude isgiven in the data table. Magnitudes falling below thisS/N=3 limit are removed from the data table. Thecount rates and errors are still given for those epochs,as they are more useful for constraining models thanthe upper limits. The upper limits are a function of theexposure time, the background count rates, and thegalaxy count rates. As shown in Figure 4, above an ex-posure time of 1000 s the galaxy count rate dominatesthe upper limit. Because the large scale sensitivity andPSF uncertainties (which scale with the count rate) arepropagated into the error, the underlying galaxy countrate imposes a floor on how faint a source could be sig-nificantly detected. Image subtraction techniques mayalleviate some of these problems but would also haveto deal with coincidence loss issues which may be sig-nificant for the extended galaxy light (Breeveld et al.2010) even if the SN itself is faint. For the standardUV-weighted UVOT mode 0x223f, 1/3 of the time isthe uvm2 filter. One can estimate the time needed inuvm2 and multiply by three to estimate the exposuretime needed for all six filters. Some sources are bright enough to saturate UVOT’sphoton-counting detector, meaning that nearly everyframe is recording a count such that the true number ofincident photons cannot be determined. For each epochwe set the limiting magnitude on the bright side (giventhe background and galaxy counts) corresponding toa measured count rate of 0.98 counts per frame andreport that in the data table. For count rates abovethis limit, both the magnitude and the count rate areexcluded from the data table. The count rate errors forsaturated sources are not useful as constraints as theupper bound is infinite.The accuracy of the photometry has been checkedusing a variety of SN and non-SN sources (Poole et al.2008). We restrict comparisons to ground-based datato B and V due to the shorter wavelength response ofthe Swift u filter compared to ground-based JohnsonU and Sloan u. Where differences have been found,it is usually due to a nearby star (as in the case ofSN 2005am) or a high count rate from the underlyinggalaxy which causes the coincidence loss correction tobe underestimated (SNe 2006dd, 2006mr, 2011iv, oth-ers). Based on comparisons, we exclude data wherethe underlying galaxy is measured to be brighter than8 counts/s with a caution that the photometry mightbe off by 0.05 mag between 6-8 counts/s. The coinci-dence loss from a flat, extended source was studied inBreeveld et al. (2010) but is not known for the case of astructured background like a galaxy. In the absence ofsuch issues, we find the UVOT b,v photometry gener-ally agree within 0.05 mags of published ground-basedB,V photometry even without accounting for small dif-ferences in the filter shapes. uvm2 exposure time (template~1300-3000 s) uv m li m iti ng m a gn it ud e ( s i g m a ) SN2005keSN2005cfSN2006ejSN2007SSN2007bmSN2007cqSN2008hvSN2011bySN2005hkSN2007cvSN2008aePTF09dnpPTF09dnlPTF10icbSN2011dnSN2011fe (a) Title A uvm2 galaxy count rate in 5" aperture uv m li m iti ng m a gn it ud e ( s i g m a ) SN2005keSN2005cfSN2006ejSN2007SSN2007bmSN2007cqSN2008hvSN2011bySN2005hkSN2007cvSN2008aePTF09dnpPTF09dnlPTF10icbSN2011dnSN2011fe (b) Title B
Fig. 4
Upper limits for a sample of SNe as a function ofthe exposure time (top panel) and the count rate of the un-derlying galaxy (bottom panel). The limits level out beyond1000 s and are dominated by the brightness of the galaxy.
The very nearby SN 2011fe is the best-studied SN Iaat all wavelengths. In the UV, it was observed withintwo days of explosion (Nugent et al. 2011) with a well-sampled light curve from UVOT (Brown et al. 2012b).Pereira et al. (2013) found a 0.2 mag discrepancy inall 6 UVOT bands compared to their HST and Su-pernova Factory spectrophotometry. Most of this dif-ference is the result of the time-dependent sensitiv-ity not being uniformly applied to the photometry inBrown et al. (2012b). To further check the consistency,we have downloaded the HST spectra of SN 2011fe fromthe HST/MAST archive Mazzali et al. (2013) and per-formed our own spectrophotometry in the UVOT sys-tem. The spectra used by Pereira et al. 2013 were inter-polated in time to their optical spectra for the creationof a bolometric light curve. After correction for theUVOT time-dependent sensitivity, the HST UV spec-trophotometry is generally within the scatter of theUVOT photometry. This is shown in Figure 5. In theuvm2 filter the HST-STIS/CCD spectrophotometry isstill 0.1 mag brighter than the UVOT photometry andthe HST-STIS/MAMA spectrophotometry for the oneepoch with observations in both. This is likely due toscattered light in the STIS/CCD. See Figure 6 in http://archive.stsci.edu/pub/hlsp/stisngsl/aaareadme.pdf
Fig. 5
UV light curves of SN 2011fe with UVOT pho-tometry in the uvw1 and uvm2 bands and spectrophotome-try from HST/STIS CCD and MAMA spectra. The bot-tom panels show the residuals from a polynomial model(Brown et al. 2012b) to flatten the curves and allow a vi-sual comparison between the UVOT photometry and theHST spectrophotometry. It is not the y-value that is im-portant but the consistency between the UVOT and HSTpoints.
The well-sampled, multi-filter Swift SN data is excellentfor studying the UV and optical evolution of individ-ual SNe and comparing across different classes. Thesample size is also large enough to compare objectswithin the same class or subclass (Milne et al. 2013;Pritchard et al. 2013). The Swift sample also includessome rare objects and subclasses that can be comparedto the others. One potentially fruitful application is us-ing the UV/optical photometry to distinguish the SNclass or even subclass without requiring spectroscopy.Cappellaro, Turatto, & Fernley (1995) and Panagia(2003) used IUE spectra/spectrophotometry to showthe UV color differences between SNe I and II. InBrown et al. (2009) we used the Swift/UVOT data toadd the temporal dimension to show that SNe IIP areonly bluer than SNe I at early times. SNe IIP becomeredder with time, becoming indistinguishable in colorbeginning about two weeks after explosion. The situ-ation has become more complicated with the increasein subclasses observed by Swift/UVOT. Brown et al.(2009) did not include SNe IIL, IIn, or IIb, the recentlyidentified classes of super-Chandrasekhar mass SNe Iaor superluminous SNe (SLSNe; Gal-Yam 2012). The in-trinsic dispersion of colors within a class of SNe is alsobetter understood with a larger sample and may leadto the identification of differences or subclasses withina class (Milne et al. 2013).In Figure 6, we revisit some of the color-color andcolor evolution plots from Brown et al. (2009) usingwell-observed, local SNe of most classes and subclasses.Figure 6 shows the time evolution of the u-band ab-solute magnitudes and u-v colors, an absolute u bandmagnitude versus uvm2-u color plot, and a u-v versusuvm2-u color-color plot. Given a SN type and cur-rent optical magnitude, one can estimate by the colorof similar objects the current and future UV bright-ness and thus the observability by Swift or a futureUV mission. Adding absolute magnitude as a dimen-sion breaks some degeneracies of color and extinction.Some regions of some plots are more congested thanothers, but rarely are the same degeneracies present inall the plots. If the colors of an object match more thanone type, usually the addition of multiple epochs willallow the object to be uniquely typed. For real-timeadaptation of observing plans, Figure 6 or somethingsimilar may be sufficient (see also Gal-Yam et al. 2004for an example of optical phototyping). When the clas-sification needs are more rigorous (i.e. for identifica-tion of cosmologically useful SNe Ia or differentiatingcore-collapse and thermonuclear SNe for rate measure-ments), a larger sample needs to be utilized to include the dispersion within classes and a statistical treatmentof the likelihoods. The full version of SOUSA will be anexcellent data set of rest-frame UV photometry againstwhich SNe can be compared. Photometric classificationof SNe will be critical for large surveys such as LSSTwhich will find many more SNe than can be followed upspectroscopically, and many of these will be observed inthe rest-frame UV.From the absolute magnitudes one can also deter-mine the distance out to which one can follow de-sired phases of different SN types with Swift/UVOTor future UV observatories (see also Figure 2 for themid-UV apparent and absolute magnitudes). For mostSN types the limiting distance is farther than thez=0.02 commonly observed in the past with Swift.SNe Ia are now being targeted between z=0.02-0.035in the nearby Hubble flow to improve their distanceand absolute peak magnitude measurements. YoungSNe II could be observed even farther. Several UV-bright SLSNe can be seen rising above the rest ofthe SNe in the bottom panel of Figure 2. These in-clude the hydrogen-rich SNe 2008es (Gezari et al. 2009;Miller et al. 2009) and 2008am Chatzopoulos et al.(2012) and several hydrogen-poor SLSNe (PTF09atu,PTF09cnd, and PTF09cwl; Quimby et al. 2011) whichwere observed at redshifts z ∼ .
2. The extreme bright-ness of SN 2008es suggests similar objects could bedetectable by Swift/UVOT out to redshifts of z ∼ . Acknowledgements
We wish to thank the many members of the Swiftteam who respond to and schedule our many SN re-quests, sometimes at very inconvenient times. We arealso grateful to the many different groups who dis-cover SNe and announce them to the world so we canfollow them up in the UV. SOUSA is supported byNASA’s Astrophysics Data Analysis Program throughgrant NNX13AF35G.
Fig. 6
Top left: u-band absolute magnitudes versus estimated time since explosion. Top Right: u-band absolute magni-tudes versus uvm2-u colors. Bottom left: u-v colors versus time. Bottom right: u-v colors versus uvm2-u colors. References
Baron, E., Branch, D., Hauschildt, P. H., et al. 2000, ApJ,545, 444Bayless, A. J., Pritchard, T. A., Roming, P. W. A., et al.ApJL, 764, 13Breeveld, A. A., et al., 2010, MNRAS, 406, 1687Breeveld, A. A., et al., 2011, in American Institute ofPhysics Conference Series, 1358, 373Brown, P. J., Holland, S. T., James, C., et al. 2005, ApJ,635, 1192Brown, P. J. 2009, “The Ultraviolet Properties of Super-novae” PhD Thesis, The Pennsylvania State UniversityBrown, P. J., Holland, S. T., Immler, S., et al. 2009, ApJ,137, 4517Brown, P. J., Dawson, K. S., Harris, D. W., et al. 2012a,ApJ, 749, 18Brown, P. J., Dawson, K. S., de Pasquale, M., et al. 2012b,ApJ, 753, 22Bufano, F., Immler, S., Turatto, M., et al. 2009, ApJ, 700,1456Campana, S, Mangano, V., Blustin, A. J., et al. 2008, Na-ture, 2006, 442, 1008Cappellaro, E., Turatto, M., & Fernley, J. 1995, in IUE–ULDA Access Guide