Reduction of CCD observations obtained with the Fabry-Perot scanning interferometer. II. Additional Procedures
aa r X i v : . [ a s t r o - ph ] M a y Astrophysical Bulletin, vol. 63, No.2, 2008, pp.181-192
October 31, 2018Translated from Astrofizicheskii Bulluten, vol.63, No.2, 2008, pp. 193-204
Reduction of CCD Observations Made with the Fabry-PerotScanning Interferometer. II. Additional Procedures
A.V. Moiseev and O.V.Egorov Special Astrophysical Observatory Russian Academy of Sciences, N. Arkhyz, KChR, 369167, Russia Sternberg Astronomical Institute, Universitetskii pr. 13, Moscow, 119992 RussiaNovember 12, 2007/Revised: November 20, 2007
Abstract.
We describe a software package used at the Special Astrophysical Observatory of the Russian Academyof Sciences to reduce and analyze the data obtained with the Fabry-Perot scanning interferometer. We alreadydescribed most of the algorithms employed in our earlier Paper I (Moiseev, 2002). In this paper we focus on extraprocedures required in the case of the use of a high-resolution Fabry-Perot interferometer: removal of ghosts andmeasurement of the velocity dispersion of ionized gas in galactic and extragalactic objects.
1. Introduction
Scanning Fabry-Perot interferometer (FPI) can be usedto perform a detailed analysis of the structure andkinematics of galaxies, nebulae, and other extended ob-jects. Detailed descriptions of the main idea of thisobservational method and references to earlier paperscan found in Moiseev (2002) (hereafter Paper I) andGordon et al. (2000). The result of observations has theform of a set of two-dimensional interferograms—the con-volution of the monochromatic image of the object withthe transmission curve of the FPI at each step of scanning.After special reduction (phase-shift correction) these in-terferograms can be assembled into a “data cube”. In sucha cube, each spatial element in the detector plane is asso-ciated with its individual spectrum. The spectral intervalis usually not very wide and amounts to only 5–50 ˚A, mak-ing it possible to study, e.g., the kinematics of ionized gas,based on the data for one or two emission lines.To reduce observational data obtained with the FPI,one needs appropriate software, which differs from thesoftware employed to analyze the data obtained with slitspectrographs. Paper I gives a brief review of the systemscommonly used to reduce such data. Of the recently pub-lished papers on the subject we point out the paper byDaigle et al. (2006) who suggested a number of new pro-cedures: adaptive spatial binning of data cubes prior toconstructing the velocity fields, algorithms for detectingemission lines and subtraction of night-sky lines.At the Special Astrophysical Observatory of theRussian Academy of Sciences (SAO RAS) the scanningFPI is a part of the CCD-based SCORPIO multimode
Send offprint requests to : A.V. Moiseev, e-mail: [email protected] focal reducer (Afanasiev & Moiseev, 2005). Unlike two-dimensional photon counters, which made a return re-cently and are now used in combination with the FPI(Gach et al., 2002), “slower” CCDs require a different kindof algorithms to reduce the data obtained. This concerns,first and foremost, photometric correction of channels andnight-sky spectrum subtraction. We addressed this issuein detail in our earlier Paper I, where we described thealgorithms used in the software to reduce SCORPIO ob-servations performed in the scanning FPI mode.According to ASPID database, a total of about 180FPI data cubes have been obtained with the 6-m telescopeof SAO RAS in 2000–2007, with at least 30 publicationsbased on these data. The experience in the reduction ofthis observational data allowed us to improve our softwarepackage and develop a number of new useful proceduresfor data analysis. In this paper we focus on describingthese procedures. In section 2 we address the problem ofremoving the ghost light, which arises in high-resolutionFPIs, in section 3 we consider the problem of velocity dis-persion measurements and in section 4 briefly describethe software package currently employed to reduce obser-vations made with the 6-m telescope using the scanningFPI incorporated into SCORPIO instrument.
2. Ghosts subtraction
SCORPIO is now equipped with two scanning interferom-eters (hereafter referred to as IFP235 and IFP501) witha gap between the plates corresponding to 235 and 501orders of interference, respectively, at the wavelength of λ . http://alcor.sao.ru/db/aspid/ Moiseev, Egorov: Reduction of CCD observations obtained with the scanning FPI interferometers. Because of the FPI location in the out-put pupil of the system between the collimator and thecamera of the focal reducer ghost lights appear on theimage. Ghost lights are due to backlight reflecting insidethe interferometer plates and between these plates and thenearest lenses of the focal reducer, surfaces of the narrow-band filter, etc. This is a well known problem, and var-ious ghost “families” were described by Bland-Hawthorn(1995), and a somewhat more detailed description can befound in the paper by Jones et al. (2002). Putting IFP235into the beam produces ghosts so that the main imageand these ghosts are aligned symmetrically with respectto the optical axis of the FPI. The intensity of these ghostreflections (which we refer to as “diametral ghosts”, “D”,in accordance with the ghost classification suggested byBland-Hawthorn (1995) and Jones et al., 2002) amountsto about 10% of the intensity of the main image. In suchcases it is usually recommended to tilt the FPI with re-spect to the optical axis of the system so as to move theghost outside the detector field. However, certain designfeatures of SCORPIO focal reducer prevent the use of thisoption. Therefore during observations the image of the ob-ject should be placed off the optical axis, and care mustbe taken to ensure that ghosts from field stars do not fallon the object of interest. All these problems decrease theworking field of view of SCORPIO focal reducer by a fac-tor of two.In observations made with IFP501 another—“exponential” (“E”, according to the terminology ofBland-Hawthorn (1995) and Jones et al., 2002)—familyof ghost images appears. These ghost images form asa result of back reflections inside the plates of the FPIand pose a more serious problem than diametral ghosts,because the nearest exponential ghost is located withinonly 16 ′′ of the object image, and its intensity is ratherhigh (about 12% of the brightness of the object). Suchintense ghost images are due to the degradation of theantireflective coating of the outer surfaces of the IFP501plates. Unfortunately, the cost of plate replacement (orrecoating) in this case is close to that of assembling anew FPI, which is rather high. We nevertheless weretempted to use IFP501 before a new interferometer wasacquired, because the former performs well in terms ofother characteristics—first and foremost, because of its R ≈ α line wavelength,which is rather high for extragalactic astronomy. Suchan instrument proved to be very popular among the ob-servers at the 6-m telescope for tasks that are impossibleto perform with IFP235 because of its lower spectralresolution. These were primarily observations of nebulaewith relatively small (within 100–200 km/s) range ofradial velocities: Herbig–Haro objects, star-formingregions in the Milky Way and other galaxies, etc.In a number of cases ghost traces can be removed atthe stage of data reduction by applying the appropriatealgorithms. Figure 1 (whose idea we borrowed from Bland-Hawthorn (1995) and Jones et al., 2002) clearly illustratesthe complex pattern of ghost images. Namely, the image of the galaxy (G) produces closely-spaced ghost imagesE G and E G whose brightness is equal to 4% and 12%of the object brightness, respectively. Now, E G producesE G, which, in turn, produces E G—and the brightness ofthese ghosts is equal to 1.4% and 0.15% of the brightnessof G, i.e., the brightness of ghosts decreases exponentiallyas it must be in the case of multiple reflections. Theseghosts appear at the front plate of the FPI before lightpasses through the interferometer, and therefore are re-ferred to as monochromatic—their relative intensity in thedata cube does not change with wavelength λ . At the sametime, ghost E G appears at the back plate of the FPI afterinterference has taken place. The image of such a ghost onthe interferogram is located in a domain corresponding toother wavelengths. Hence the data cube should exhibit anoffset in λ between the spectra of G and E G, and it mustdepend on the position of the image on the detector.In Fig. 1 we also indicate diametral ghosts, i.e., ghostslocated symmetrically with respect to the optical axis ofthe interferometer. The image of galaxy G and the corre-sponding ghost DG are located symmetrically with respectto the optical axis, ghost E G produces diametral ghostDE G, and ghost E G, respectively, produces DE G, etc.Ghost DG also generates a family of secondary exponen-tial ghosts: E DG can be seen in the figure. The latter issuperposed by diametral ghost DE G. Diametral ghostshave rather complex structure, as is immediately appar-ent in the case of ghost DS produced by the brightestforeground star. Namely, this ghost consists of two imagesof the star—a normal image (with the intensity equal to ∼
5% of that of star S) and a strongly defocused image,which has the shape of a ring, whose total intensity isequal to about 10% of intensity S.Note that it is by no means always possible to findso many ghost images as in Fig. 1. The figure illustratesa limit case corresponding to observations of a relativelybright object, whereas usually only the brightest ghosts(DG, E G, and E G) can be seen. We describe the pat-tern of ghost images in such detail, because we believethat our analysis will help the users better understandthe data obtained and distinguish real emission featuresfrom ghosts.If we denote the ideal data cube (i.e., the cube withoutghosts) as I real , then the wavelength-calibrated observedcube has the following form: I obs ( x, y, λ ) = I real ( x, y, λ ) + I ghost ( x, y, λ ) , (1)where the family of exponential ghosts can be representedin the form: I ghost ( x, y, λ ) ≈ f ˆ P I − real ( x − ∆ x , y − ∆ y , z )++ P i max i =1 f i I real ( x − ∆ x i , y − ∆ y i , λ ) . (2)Relative brightness f i of ghosts decreases exponen-tially with the number i of reflections and therefore sum-mation is performed not to infinity, but up to i max = 3 − oiseev, Egorov: Reduction of CCD observations obtained with the scanning FPI 3 Fig. 1.
Examples of ghost images produced in the IFP501 interferometer. H α observations were performed on January30/31, 2007 at the request of C. Mu˜noz-Tu˜n´on. Here “G” and “S” denote the II Zw 70 galaxy and a bright foregroundstar, respectively. The sum of all channels in the data cube is shown. Diametral and exponential ghost images aredenoted by “D” and “E”, respectively. The cross indicates the center o the field of view.insignificant. For the same reason, we neglect the contri-bution of secondary ghosts, such as non-monochromaticsatellites of the ghosts of the monochromatic family. Thefirst term in (2) describes the wavelength-shifted ghostE G. Here ˆ P denotes “phase correction”, i.e., the transfor-mation from spectra expressed in terms of the FPI chan-nel numbers z to spectra expressed in terms of the wave-lengths: I ( x, y, λ ) = ˆ P I ( x, y, z ) = I ( x, y, k ( z + p ( x, y ))) , where p ( x, y ) is the so-called phase map (see Section 2 inPaper I for details), and k ( z ) = k z + k is a linear func-tion whose coefficients are determined by the particularFPI employed. Correspondingly, I − real is a result of the re-verse transformation from wavelength scale to the scale ofinterferometer channels: I − real ( x, y, z ) = ˆ P − I real ( x, y, λ ) . Our aim is to infer from I obs the best approximationto ghost-cleaned cube I real . We use the following iterative procedure for this. We substitute I obs for I real in formula(2) to obtain the first approximation to the ghost model I ghost ( I real ). In this case, the first approximation to theghost-cleaned cube has, according to formula (1), the fol-lowing form: I real = I obs − I ghost . We now substitute I real into formula (2) to derive the following approximation toghost model I ghost , etc., up to I nghost . Because of the smallrelative brightness of ghost images ( f i ≪ n =3–4) are needed to construct a close-to-reality ghost-image model.The following relations are true for the family ofmonochromatic ghost images ( i ≥ x i = i ∆ x , ∆ y i = i ∆ y , f i = f i , and hence model (2) has only six free parameters, whichare chosen in a way to minimize the differences between I nghost and I obs . The following table gives the mean valuesof these parameters, which vary little from night to night.Our experience in the reduction of the data obtainedwith IFP501 shows that the model based on the above Moiseev, Egorov: Reduction of CCD observations obtained with the scanning FPI
Fig. 2.
Subtraction of ghost images. Here we show the sum of all channels of the data cube for the II Zw 70 galaxy:(a) initial image, (b) model of brightness distribution in ghost images, and (c) the result of subtracting the modelfrom the initial cube.
Table 1.
Parameters of ghost-image model in H α line i = 0 i=1 f i x i , ′′ -7.67 9.80∆ y i , ′′ -16.42 13.63 algorithm usually describes real ghost images rather well,with no appreciable bias. The only exception are the dataobtained under unstable atmospheric conditions, whencounts in individual channels had to be averaged in orderto compensate for appreciable (more than 10–20%) varia-tions of the FWHM of stellar images (for a description ofphotometric correction see Paper I.Figure 2 shows how the ghosts around the image ofthe dwarf galaxy II Zw 70 are subtracted. Other exam-ples of observational data processed using the above de-ghostification procedure can be found, e.g., in Lozinskayaet al. (2003,2006) – a study of star-forming regions in thedwarf galaxies IC1613 and VII Zw403, Mart´ınez-Delgadoet al. (2007) – kinematics of ionized gas in blue compactgalaxies, Movsessian et al. (2007) – study of outflows fromyoung stellar objects in the HL/HX Tau region.Yet another problem arises in observations of objectswith strong surface-brightness gradients, where an intenseghost of, e.g., the galaxy nucleus, projects onto regions ofmuch lower brightness. The level of Poisson noise is de-termined by the combined intensity of the ghost and thebase. Therefore after the subtraction of the ghost modela situation may arise where the useful signal in the regionconsidered is comparable to the amplitude of photon noise.To avoid loss of spectral information from low-brightnessregions in observations of such objects, we recommend todivide the planned exposure into two and perform obser-vations successively with two different orientations of theinstrument’s field of view turned by about 90 ◦ in posi-tion angle. The ghosts obtained in the two data cubesshould then project onto different regions of the object.After primary reduction, the corresponding ghost modelis subtracted from each data cube and the regions are masked where the signal-to-noise ratio decreased stronglyafter ghost removal. The two cubes are then combinedinto one and spectra of masked regions in each cube aresubstituted by the corresponding “good” spectra from theother data set. One of the best examples of application ofthe above algorithm is the reduction of observations ofthe nearby dwarf galaxy IC 10 reported by Lozinskaya etal. (2008). Here the image of emission shells of ionizedgas occupies more than half of the entire field of view ofSCORPIO instrument. Figure 3 illustrates the sequence ofoperations in the process of subtracting ghosts from theimages of these galaxy.
3. Velocity dispersion measurement
A number of observational programs performed with theFPI (e.g., mapping the ionized-gas velocity dispersion ingalaxies) require accurate estimates of the halfwidths ofemission lines. These estimates must take into accountthe broadening due to the instrumental profile. It is com-mon practice to use the following formula (which assumesthat both the instrumental profile of the spectrograph andthe initial—unbroadened—line profile can be described byGaussian functions with the dispersions equal to σ real and σ ins , respectively): σ obs = q σ real + σ ins . (3)Here σ obs denotes the dispersion of the Gaussian used todescribe the profile of the line observed at the output ofthe instrument. Hereafter by measuring the velocity dis-persion we mean estimating σ real from the observed spec-tra. One should keep in mind that broadening of lines inthe spectrum of the observed object may be caused notonly by velocity dispersion σ gas (the measure of chaoticmotions along the line of sight), but also by a number ofother factors. Thus, according to Rozas et al. (2000), thefollowing formula can be written for the integrated spec-trum of HII regions: σ real = σ gas + σ N + σ tr , (4) oiseev, Egorov: Reduction of CCD observations obtained with the scanning FPI 5 Fig. 3.
Sum of channels in the data cubes for the IC 10 galaxy. H α -line observations were made on September 8/9,2005 at the request of T. A. Lozinskaya. The figure shows a 72 ′′ × ′′ fragment of the image. (a) Initial image in thefirst cube; (b) the image after ghost subtraction, the arrow indicates regions with unsatisfactory quality of subtraction;(c) the image in the cube obtained by turning the field of view by 90 ◦ ; (d) subtraction of ghosts from the second cube(the arrows indicate low-quality regions), and (e) the combinations of the two data cubes. Moiseev, Egorov: Reduction of CCD observations obtained with the scanning FPI where σ N ≈ σ tr ≈ . K, respectively.Because of its simple and self-explanatory form, for-mula (3) is widely used to analyze spectroscopic data. It isoften generalized by substituting FWHM for σ . However,one must keep in mind that exact equality in (3) isachieved only for Gaussian functions, because a convolu-tion of two Gaussians is also a Gaussian. The assumptionabout the Gaussian form of the instrumental profile is usu-ally true for slit spectrographs. However, the instrumentalprofile of the FPI, which is given by the Airy function, haswide wings and is best approximated by a Lorentz pro-file (see, e.g., Bland-Hawthorn (1995) and Paper I) ratherthan by a Gaussian. Therefore if the initial profile of theemission line is a Gaussian with the dispersion determinedby formula (4), then the observed profile is the convolutionof the Gaussian and Lorentz profiles and is hence given bythe Voigt function: V ( λ, y ) = 1 √ πσ real yπ Z ∞−∞ e − x dxy + ( a − x ) , (5)where a = λ − λ √ σ real , y = w ins √ σ real . Here λ is the central wavelength and w ins denotes thehalfwidth of the instrumental (Lorentz) profile of the FPIas determined from the spectrum of the lines of the cal-ibration lamp. We then approximate the observed profileby function (5) to obtain the required estimate σ real . Itis evident from Fig. 4 that compared to the Gaussian theVoigt profile fits much better the line wings in the galaxyspectra observed with IFP501. It is clear from general con-siderations that formula (5) is a more correct tool for es-timating the velocity dispersion than formula (3). In thelatter case we have to use Gaussian approximation for theprofiles of the lines that deviate systematically from theadopted approximation (Fig. 4a), and this may introducean additional error in the estimated σ real . However, thisapproach is highly popular in velocity-dispersion measure-ments for extragalactic HII regions (see, e.g., Mart´ınez-Delgado, 2007; Mu˜noz-Tu˜n´on, 1995; Rozas et al. 2000).We believe that this method owes its popularity not onlyto the less complex appearance of formula (3) comparedto that of formula (5), but also to the fact that Gaussianapproximation of spectral-line profiles is incorporated intovirtually all packages of astronomical data reduction. Theuse of the Voigt profile for analyzing extragalactic spec-tra is less common (see, e.g., 1994), despite the fact thatintegration in (5) poses no problem for modern computers.We estimated the errors of measurement of the kine-matical parameters (radial velocity and velocity disper-sion) for both approaches considered. We smoothed theinstrumental profile of the FPI by a Gaussian with thedispersion equal to σ in and then added noise to the result-ing spectrum and estimated the velocity dispersion using Fig. 4.
Example of the H α -line spectral profile in theII Zw 70 galaxy based on observations with IFP501 (dots).The solid line shows Gaussian (a) and Voigt (b) fits.both methods. The difference between the output ( σ out )and input ( σ in ) velocity dispersions allows us to estimatethe error σ err of the velocity dispersion measuring. We ob-tained a total of 1000 independent measurements for eachfixed signal-to-noise ratio ( S/N ). We similarly estimatedthe error of measured radial velocity. We performed ourcomputations for the instrumental profiles with the widthnear the H α line equal to w ins = 35 and w ins = 115 km/sfor IFP501 and IFP235, respectively. Figures 5 (a, b) and6 (a, b) show the results of computations—radial-velocityand velocity-dispersion errors as functions of the signallevel. As expected, the error of radial-velocity measure-ments for symmetric lines does not depend on the algo-rithm employed. At the S/N = 30 it is equal to 2.5 and 8km/s for interferometers IFP501 and IFP235, respectively.The situation is quite different for the error of the mea-sured velocity dispersion. As expected, the error of mea-surements based on Voigt-profile fits smoothly decreaseswith decreasing signal level and amounts to only severalkm/s for
S/N >
20. These errors are due to noise in themeasured spectra and they contain no systematic compo-nent. Contrariwise, if measurements are based on relation(3) then systematic but not random component of mea-surement error starts to dominate at
S/N ≥
10. Morespecifically, velocity dispersion is overestimated, as it is ev-ident from Figs. 5c and 6c. Velocity-dispersion estimatesinferred from Gaussian fits to the profiles exceed the ac-tual values by 7–8 and 20–25 km/s for IFP501 and IFP235,respectively. Such a systematic error is unimportant for es-timating widths of lines with σ real >
100 km/s. However,velocity dispersion may be overestimated by up to 100% inthe case σ real = 10 −
20 km/s. This may be of critical im-portance, e.g., in the studies of ionized gas in star-formingregions, when it is necessary to identify expanding shellsor regions where the velocities of chaotic motions exceedthe speed of sound in the interstellar medium (Mart´ınez-Delgado et al. 2007; Rela˜no & Beckman, 2005). In thesecases velocity dispersion estimates inferred by fitting theVoigt profile are to be used. Or, if for some reasons theauthors prefer Gaussian approximation, it is necessary to oiseev, Egorov: Reduction of CCD observations obtained with the scanning FPI 7
Fig. 5.
Simulation of errors of measurement of kinematic parameters (at the 1 σ level). The computations were madefor IFP501. The diamond signs and filled circles show the Gaussian and Voigt-profile approximations, respectively. (a)Dependence of the error of radial-velocity measurements on the signal-to-noise ratio. The average velocity dispersionwas equal to 50 km/s; (b) error of measured velocity dispersion as a function of signal-to-noise ratio under the sameconditions; (c) comparison of the initial and measured velocity dispersion (for S/N = 20).
Fig. 6.
Same as Fig. 5, but for IFP235. Figures (a) and (b) are for the case where an average velocity dispersion isequal to 200 km/s.estimate systematic errors like we did it above and correctcorrespondingly the estimates based on formula (3).Note that Rela˜no & Beckman (2005) proposed an al-ternative method to account for the instrumental profileof the FPI using the reconstruction (deconvolution) tech-nique. This procedure is to be used for multicomponentlines. However, the method can be applied only to spectrawith sufficiently high signal-to-noise ratios.
4. Brief description of the software package
To reduce observations made with the FPI on SCORPIO,we wrote IFPWID program package with multi-window user-friendly interface in IDL 6.X language(Fig. 7). The codes are publicly avail-able at . Below inthis section we describe the principal sequence of data-reduction steps to be made when using these programs.We do not describe in detail most of the algorithmsemployed—wavelength-scale calibration, photometric cor-rection, night-sky line subtraction, etc. (see our sufficiently detailed Paper I, and only briefly list the initial data forthe reduction of a set of CCD frames: – OBJECT —inteferograms of the object studied. – NEON —images of interference rings from the emis-sion line selected by the narrow-band filter from thespectrum of the He-Ne-Ar calibration lamp. This cal-ibration is usually performed before and after the ob-serving night. – FLAT —interferograms of uniform “flat-field” illumi-nation produced by the continuous-spectrum lamp andobtained with the same narrow-band filter as was usedfor observations of the object. – TEST —images of the rings from the line-spectrumlamp in some FPI channels obtained along with ob-ject integrations at the same position of the telescope.These images are used to monitor the accuracy of scan-ning and to control the offset of the center of the ringsdue to the instrument flexures.
Moiseev, Egorov: Reduction of CCD observations obtained with the scanning FPI
Fig. 7.
IFPWID interface: the main menu (on the left) and the menu of operations with calibrated cubes (on theright).
The first stage of reduction usually requires no specialsettings. As a rule, it suffices to indicate only the num-ber of the observational night, file-name templates, andthe required frame format (some observations with theFPI involve the readout of only a fragment of the CCD).The program extracts all the remaining necessary informa-tion (parameters of the FPI, numbers of spectral channels,etc.) from the descriptors of the corresponding FITS files,which are filled automatically in the process of observa-tions. Therefore the user only has to check the requiredreduction steps in the menu (see the upper part of themenu in Fig. 7) and press the “run” button. Below webriefly list these steps (their names used in the reductionprogram are italicized): – Search for and averaging of the bias-current framestaken in the required mode of CCD readout (
MeanBiascreation ). The resulting superbias frame is then sub-tracted from all object integration and calibrationframes. The CCD employed contains virtually no “hot”pixels, dark current is small and therefore can be ne-glected for exposures about several minutes. – Creation of data cubes from individual frames (
Cubescreation ). Superbias is subtracted and bad columns aremasked. Further operations are the reduction of three-dimensional cubes (object interferograms and calibra- tion data assembled in order of channels). Cubes arestored in the standard FITS format (NAXIS=3). – Removal of cosmic-ray hits from the cubes ofcalibrating-lamp integration with line (
NEON clean )and continuous (
FLAT clean ) spectra. Simple σ -filteris used here: the counts in the spectra that deviate fromthe mean by more than preset threshold value are sub-stituted by the half-sum of the neighboring channels. – Construction of the phase-shift map based on the re-sults of Lorentz-profile fits to the lines in the NEONcube (
Phase map ). A correction is applied, where nec-essary, to allow for nonuniform scanning of the cali-brating cube (see Paper I). – Testing the accuracy of wavelength scale (
Phase test ).The NEON cube is corrected for the phase shift andthen the position of the emission line of the calibrationlamp is measured for each pixel. – Computation of corrections to the wavelength scaleconstructed from the calibration cube (
Test rings ).The offsets (along both coordinates in the CCD planeand along the wavelength coordinate) of TEST framesare computed relative to the NEON cube (to correctfor the instrument flexures and to monitor the scan-ning accuracy of the FPI). – Correction (if needed) of relative variations of FLATlamp brightness during scanning of the calibrationcube (
Flat norm ). oiseev, Egorov: Reduction of CCD observations obtained with the scanning FPI 9 – Measurement of the offsets between the transmissionmaxima of the narrow-band filter measured for theFLAT cube and for the spectra of selected stars in thefield of the object (
Flat Z-shift ). This procedure is usu-ally needed only for observations made with IFP235,where the half-width of the narrow-band filter is ap-preciably smaller than the wavelength interval betweenthe neighboring orders of interference. The offset isusually close to zero, however, it may reach 2 − . – Division of the cube by the flat-field (to correct forthe transmission curve of the narrow-band filter) andcosmic-ray hit removal (
OBJ flat and clean ). Then comes the turn of the reduction procedures that re-quire the user’s intervention more often than during thestage of assembling the data cubes. The procedures corre-spond to the buttons of the main interface (the left-handpart of Fig. 7a), which are italicized below. Reduction isusually performed in the following order: – Correction for the background gradient (
2D gradient ).In case of observations of objects in some narrow-bandfilters the background brightness distribution in theobject cube exhibits appreciable gradient even afterthe division by the FLAT, especially in the presenceof additional light pollution due to the Moon. Thisresidual gradient must be due to nonuniform illumina-tion from the FLAT lamp combined with specifics ofthe interference coatings of particular filters. The usercan correct this effect by setting the parameters of thetwo-dimensional brightness distribution to which theobject cube is normalized. – Creation of the mask for subtracting the spectrum ofthe night sky (
Mask OH ). The mask is based on theimage of the sum of the channels of the object cube.Regions with the brightness below the given thresholdare considered as “background”. The resulting maskcan be edited if necessary. – Setting the parameters for the sky-background sub-traction (
SKY parameters ). We described our tech-nique of sky-background subtraction in the Paper I.Its main idea is to average sky background emission ineach channel of the OBJ cube over the azimuthal anglewithin narrow rings centered on the optical axis of theFPI with taking into account the mask constructed atthe previous stage of reduction. The average bright-ness profile is then subtracted from the interferogramof the object. The user may choose various modes ofsuch subtraction, vary the width of the rings whereaveraging is performed, fix the center of the rings orset options for an automatic search of the ring centerin each channel. When needed, averaging can be per-formed within individual sectors (such a procedure can be useful for correcting for variations of the instrumen-tal contour over the field of view), etc. – Subtraction of the night-sky spectrum in accordancewith the parameters given above (
Sky remove ). – Photometry of stars in each channel of the object cube(
Channels preparation ). The relative offsets of the im-age centers, variations of FWHM and integral flux aremeasured for stars from a precompiled list and the av-eraged dependences of these parameters on the numberof the channel in the cube are constructed. – Photometric correction of the object cube based onthe results of photometry of stars in each individualchannel (
Photo/Shift corrections ). Corrective channeloffsets are applied and account is taken of variationsof atmospheric transparency (the channel counts aremultiplied by the computed coefficients) and seeing(convolution with two-dimensional Gaussians). – Conversion of the object cube to the wavelength scale(
Linearization ). – Subtraction of ghosts using the procedure described inSection 2 (
Ghosts remove ). The reduction steps described above produce awavelength-calibrated data cube with maximum ac-count taken of all instrumental effects. In this form it canbe analyzed using various software tools depending on theuser preferences or on the task to be accomplished. Thedata can also be converted to the format adopted in thepopular ADHOC reduction system. Below we describethe operating sequence to be performed with our softwaretools in order to map emission-line radial-velocity andvelocity-dispersion distributions.Figure 7 (on the right) shows the menu of operationswith calibrated cubes. The following procedures (theirnames are italicized) are performed: – Rotation of the cube to the “correct” orientation of theimages (with North at the top and East on the left),because observations can be performed at any arbi-trary (or specially selected) position angle (
Rotationto NE ). The turn angle is computed from the data ofthe descriptors of the FITS-file header to within 0.1 ◦ .In the cases where better accuracy is required astro-metric reduction is to be performed using field stars. – Correcting the radial velocities for the motion of theEarth about the Sun (
V-heliocentric ). The necessaryinformation is extracted from the FITS-file header. – If necessary, a cube fragment is cut containing the ob-ject studied (
Sub-cube ). In this form reduced FPI cubesare usually stored in our ASPID database 1. – Smoothing the cubes using a one- and two-dimensional Gaussians of given width along the spec- ADHOC software package was developed byJ. Boulesteix (Observatoire de Marseille) and is availableat
Fig. 8.
Interface of the program of the analysis of emission-line profiles.tral (
Smoothing along Z) and spatial (
Smoothing alongX ) coordinates, respectively.We construct the velocity fields, velocity-dispersionmaps, monochromatic images in the emission line and inthe continuum, using GAUS program whose main inter-face is shown in Fig. 8. This program can be used to ana-lyze individual profiles in selected pixels of the data cubeby fitting the emission-line profiles to Voigt and Gaussfunctions. The integral in (5) is computed via standardVOIGT function of the IDL language. The main task per-formed by the program is an automatic identification ofemission lines in the data cube and their approximationby various functions for the given parameters. This is im-portant because the number of pixels that contain usefulsignal in our data cubes may amount to several hundredthousands, thereby preventing individual approach to eachspectrum. Our procedure yields two-dimensional maps ofprofile parameters (Doppler velocities, line fluxes, etc.).
5. Conclusions
In this paper we briefly describe the software currentlyused to reduce observations made with the scanning FPIoperated as a part of SCORPIO instrument. We believethe software complex described here to fully meet the re-quirements imposed by the research tasks performed atthe 6-m telescope of the SAO RAS using the observa-tional technique considered. In our opinion, further evo-lution of the data reduction software package should beassociated with project of the development of SCORPIO-2 new-generation multimode focal reducer at the SAORAS. We hope, first and foremost, that the new illu-minator of the calibration beam would make it possi-ble to abandon flat-field correction procedures (nonuni-form residual background, FLAT cube wavelength shiftobserved in some narrow-band filters etc.). Higher au-tomation level of SCORPIO-2 (compared to the currentversion of the instrument) will allow further unification ofthe process of acquisition data with the FPI and managewithout “manual” setting of parameters in a number ofprocedures. The latter primarily concerns subtraction ofthe night-sky spectrum. In this case it will suffice to simply oiseev, Egorov: Reduction of CCD observations obtained with the scanning FPI 11 develop a pipeline for the reduction of the data obtainedwith the scanning FPI.
Acknowledgements.
We are grateful to A. A. Smirnova for herassistance in the preparation of the text. This work was sup-ported by the Russian Foundation for Basic Research (projectno. 06-02-16825) and a grant of the President of the RussianFederation (project MK1310.2007.2).
References
Afanasiev V.L., Moiseev A.V., 2005, Astronomy Letters, ,193; astro-ph/0502095Bland-Hawthorn J., 1995, in Proceedings of IAU Colloquium149 on Tridimensional Optical Spectroscopic Methods inAstrophysics, Marseille, France, 1994 , Eds. by G. Compte,and M. Marcelin, (ASP Conference Series, San Francisco,1995), 72Chilingarian I., Afanasiev V., Bonnarel F., et al., 2007,in
Proceeding of the Conference on Astronomical DataAnalysis Software and Systems XVI, Tucson, USA, 2006 ,Eds. by R.A. Shaw, F. Hill, and D.J. Bell, (ASP ConferenceSeries, 2007), 271; arXiv:0711.0341 [astro-ph]Daigle O., Carignan C., Hernandez O., et al., 2006,MNRAS , 1016Gach J.-L., Hernandez O., Boulesteix J., et al., 2002,PASP , 1043Gebhardt K., C. Pryor, T. B. Williams, and J. E. Hesser,AJ , 2067 (1994).Gordon S., Koribalski B., Houghton S., Jones K., 2000,MNRAS , 248Jones D.H., Shopbell P.L., Bland-Hawthorn J., 2002,MNRAS , 759Lozinskaya T.A., Moiseev A.V., Podorvanyuk N.Yu., 2003,Astronomy Letters , 77; astro-ph/0301214Lozinskaya T.A., Moiseev A.V., Avdeev V.Yu., Egorov O.V.,2006, Astronomy Letters , 361; astro-ph/0605189Lozinskaya T.A., Moiseev A.V., Podorvanyuk N.Yu., BurenkovA.N., 2008, Astronomy Letters , 217; arXiv:0803.2435[astro-ph]Mart´ınez-Delgado I., Tenorio-Tagle G., Mu˜noz-Tu˜n´on C., etal., 2007, AJ , 2892Moiseev A.V., 2002, Bull. SAO , 74; astro-ph/0211104(Paper I)Movsessian T.A., Magakian T.Yu., Bally J., et al., 2007,A&A , 605Mu˜noz-Tu˜n´on C., Gavryusev V., Castoneda H.O., 1995, AJ , 1630Rela˜no M., Beckman J.E., 2005, A&A , 911Rozas M., Zurita A., Beckman J.E., P´erez D., 2000,A&AS142