Digitized First Byurakan Survey - DFBS







First Byurakan Survey

                            Technical data

                            FBS zones statistics

                            Markarian galaxies

                            FBS based projects




                            About DFBS






                            Sample spectra

                            Data visualization

                            Science with DFBS


                            DFBS at Roma

Virtual Observatories

                            Armenian VO





Other related links

                            Armenian Astronomy

                            Byurakan Observatory





Digitization of the FBS

Main steps of the digitization


          archiving on DVDs and HDDs

          astrometric solution

          extraction of spectra

          wavelength calibration

          density and flux calibration

          multiband (UBVR and O/E) photometry

          making up template spectra

          numerical classification

          DFBS catalog and database

          web page and user interface

Digitization process (scanning)

The plates have been carefully cleaned and then digitized with an EPSON 1680 Pro (A4 size) scanner at 1600 dpi in transparency (positive) mode controlled by a Personal computer running under Windows(TM) operating system. An "ad hoc" program written by Stefano Mottola allowed to write directly the resulting image in FITS format, 16 bit. The data number actually span the range 0 (dark) to 16383 (transparent).

The plates are located on the scanner with the emulsion in contact with the scanner glass plate. A black paper sheet is used to cover one of the unexposed plate corners to allow a measure of the effective zero of the data numbers.

The automatic normalization of the scanner does not work properly for astronomical plates. The scanner software however allows to set manually the data number for the darkest and brightest areas of the image. After several trials we decided to scan each plate setting the lower limit on the black corner and the higher limit on one of the unexposed corners, caring that in no case the data go outside the numerical range. In practice the black corner counts are around 600 DN and the plate veil counts around 14000 DN.

Scanning one plate takes about 8 minutes. Since one plate is 9600x9600 pixels, three plates can be stored on a CD-ROM. One copy of the archive is kept in Byurakan and a second one in Roma.

The scanning work started in Roma in June 2002 with a set of 20 plates, allowing the definition of the instrumental settings. Then the scanner was moved to Byurakan where all the plates are stored, and routine work started. We plan to finish the scanning of all the available plates within 2003. The list of the plates with their identification, date, central coordinates and processing status is updated periodically. Interested researchers may have copies of the CD-ROMs on request.


The scanner Epson Expression 1680 Pro

Resolution 1600 dpi

Size of the pixel 15.875m or 1.542

Mode transparency (positive) mode

Dynamic range 16 bit

Software scanfits by Stefano Mottola, FITS images

Dimensions 9601´9601 pixels each plate

Size of a plate 180 MB

Length of spectra 107 pixels

Width of spectra 5 pixels

Scanning direction East-West (along RA)

Scanning period June 2002 – December 2003

Storing on DVDs whole DFBS on 85 DVDs

Results 1874 plates (1139 FBS fields = 17,056 deg2)

Astrometric solution

The determination of the coordinate system of a plate is made with a manual process in a two-step procedure. The first one requires the identification of a small number (about 20) of bright stars from the Tycho catalogue (brighter than 9th mag). We assume the intensity peak in the red part of the spectrum as the star position. A plate solution is then computed and written on the FITS header using two IRAF tasks of the images.imcoords package: ccmap and ccsetwcs. The second step refines this first solution, using all the Tycho catalogue stars present on the plate. The IRAF ccfind task find all these stars using the first plate solution, and then a second solution is computed using again ccmap and written on the FITS header using ccsetwcs. The star identification and preparation of the input files requires about one hour.

The plate scale is 1.55 arcsec/pixel in the scanning direction and 1.54 arcse/pixel along the CCD. The positional accuracy obtained is 1 pixel rms, quite sufficient for a safe object identification (a spectrum is typically 5 pixels wide).

The actual astrometric (plate) solution has been done using bright stars accurate positions from GSC-2. The software written by Hans Hagen (Hamburger Sternwarte) has been used. At present all plates have astrometric solution. The typical rms accuracy is 1 arcsec.

Extraction of spectra

To work with individual spectra from the DFBS plates and finally create the DFBS catalog, we need to apply quick automatic extraction software. We have performed two opposite approaches to extract spectra. The first uses the program SExtractor, an automatic extraction of all objects and then making the databases. This method is good for finding all objects, however there are still a number of problems: the central positions of images are taken, not the real star positions (red head of the spectrum); defects and artifacts are taken as objects; faint objects are being missed; there are problems with superposed (blended) images (being extracted as one). We have tested this software and concluded that it can be used for relatively low-density fields and brighter objects.

The second approach is to extract objects knowing their positions from an available catalog, e.g USNO. The 2 databases will be matched to reveal the real objects (to avoid artifacts, etc.) as well as variable objects, which may not be present in the USNO catalog. Finally, it will be possible to have the 2D and then 1D spectra. This catalogue-driven procedure has already been created and tested. The list of all objects present in USNO-A2 down to the plate limit and included in the sky area of an FBS plate is converted into pixel coordinates. Then an image section of 21´150 pix including one well-exposed star is selected and the spectrum is extracted in interactive mode. This process allows derivation of the orientation of the spectra on the plate and defines the template for subsequent automatic extractions. Finally, all spectra of the list are extracted automatically, assuming as sky value the median of an area 21´150 pixels centered on each spectrum.


A disadvantage for the second approach is that we can lose a number of variable objects present in DFBS but absent among the bright objects of USNO (or completely absent there). In this case the first method complements the second one.

Wavelength calibration

The red cutoff of the FBS spectra is rather sharp, so that it can be used as a reference point, but it is mildly sensitive to the brightness and spectral type of the object. For calibration, we use stars of intermediate brightness (optimally exposed) and types (having intermediate colors) to have a definite red edge, but not overexposed. The sensitivity gap near 5300 A is used as well, as it is also more or less independent of object type and brightness. However, all wavelength ranges given for the Kodak emulsions are rather crude and approximate: we need to define clearly the start and end wavelengths obtained and the sensitivity gap position in l to obtain a good dispersion curve.


We use WDs, subdwarfs, CVs, and QSOs from the available catalogs, which have broad Balmer, He, and other lines. We will use 10 main references points: l-start (~3400A), HV, He, Hd, Hg, HeII l4686A, Hb, sensitivity ”gap” (~5300A), Ha, and l-end (~6900A). The calibration based on these points is sufficient for a coarse spectral classification. However, we will proceed to produce a good dispersion curve (and linearization), and after the extraction, transform all spectra into wavelengths. To obtain and then refine the dispersion curve and hence make the wavelength calibration, we need to use a few hundred stars with known spectral lines to average their data. The dispersion is strongly non-linear. A preliminary study shows that we obtain 22 A/pix wavelength scale at the blue edge, and 60 A/pix at the red edge, mean dispersion being 32.7 A/pix. At Hg, it is about 28.5 A/pixel. However, the spectral resolution is 1.5-2 times worse, as the photographic grains occupy 1.5-2 pixels.

Flux calibration and Multiband photometry

This process includes density(DN)-to-intensity and intensity-to-flux conversion. The density calibration is made from the original data numbers (DN) according to the following approximate formula:


where D is the (linear) density (units of transparency given by the scanner), V is the average DN value for the unexposed plate, B is the same for the black corner, and T is for a given pixel. The FBS plates do not have photometric calibration, i.e. we cannot build easily a characteristic curve for each plate. However, we plan to use a typical curve for this type of emulsion to minimize the uncertainties and obtain the intensity (I) values. Furthermore, we have made a number of trials to obtain an accurate sensitivity (response) curve for Kodak F emulsion. Finally, we obtain the real spectral energy distribution (SED) for objects and make a transformation for all spectra extracted. This will help the classification significantly.


We plan also to make some rough photometric calibration using photometric standards in each field. It is estimated that up to 0.3m accuracy may be reached. However, we should remember that the photometry is not the main purpose of the DFBS. It should be done using the DSS database (rough photometry of the MAPS and USNO catalogs) (Cabanela et al. 2003; Monet et al. 2003). To complete this task, we need to create corresponding software and apply on each plate to obtain intensity values in the output. We will finish with flux calibrated spectra for about 20,000,000 objects, having appropriate known SEDs easy to use for different purposes.


Plans for making multiband (UBVR) photometry are active. The estimated accuracy that can be achieved is 0.3m, however, it will be 1.5-2 times worse for V band, which falls near the sensitivity gap, as well as for U (3660A) for faint red objects, and R (6930A) for faint blue objects, when the values will be near the background level (the R values will be systematically underestimated as our spectra include only half of its bandwidth, but a correction will be applied). To link our data to the POSS O (4050A) and E (6450A) magnitudes, we try to measure these values as well. In fact, these bands are better suited for our spectra, and they are being measured with higher accuracy. The photometry will be useful to find cases of variability, when data from the FBS plates are matched with other data available. We plan to give the estimated photometric values for each object in the DFBS catalogue.

Numerical classification

Here also we apply 2 opposite approaches. Both methods are efficient, as they are useful for different purposes. The first is based on modeling template spectra for different types of objects from available catalogs averaging their FBS spectra for each magnitude separately. To model each template, we will need a few dozen typical spectra corresponding to known objects from the catalogs. A search for these objects will be made in the DFBS, and their low-dispersion spectra will be extracted. Templates of these spectra for different types of objects will be created for different magnitudes. At the end, we will have a database of different types of spectra with uncertainty limits to allow searches with different degrees of confidence. Then we search for similar new objects (QSO, BLL, Sy, CV, WD, sd, M, C, etc.) among the FBS low-dispersion spectra. The success rate depends on the limits of given parameters: we can select either a small number of good candidates missing a fraction of objects, or a large number of candidates with a contamination of other types, but having all objects of interest. Thus a compromise should be made depending on the given task. This method is good for a quick search for objects of interest.


The second approach is based on making a numerical classification scheme for all FBS spectra. The classification principles are based on the relation of magnitudes and widths of the spectra (for separation of stellar and diffuse objects), spectral energy distribution (SED), ratio of the red and blue parts (color), length of the spectrum, presence or absence of broad spectral lines, etc. The classification is based on criteria worked out during the selection of blue stellar objects, red stars, and identification of IRAS sources. Our classification scheme will be linked to the general classification using standard objects of known types. This approach is good for working with all objects in the field. After having all objects classified, a cross-correlation with known objects will be made to derive principles of how to use the numerical classification for further scientific purposes.


Automatic search for new objects


After the preliminary reduction (plate solution, extraction, and wavelength and density/flux calibration) and working out the classification principles, it will be possible to conduct searches for definite types of objects. As we have a number of science goals, we are going to perform a search for new candidate QSOs, UVX galaxies, etc. Searches will be available by several methods, including searches for optical counterparts for X-ray, IR and radio sources (optical identifications) from corresponding catalogs, and using low-dispersion templates of QSOs, BL Lacs, Seyferts, BCDGs, starbursts, and other objects to find missed objects and fainter ones unavailable to previous searches just by eye, thus extending the limiting magnitude to 18m, as well as searches by colorimetric methods using the DSS1/DSS2. Corresponding software for quick search giving coordinates, comparison with the corresponding DSS fields, making 1D cuts for each spectrum will be created. Our numerical classification will be applied to these spectra.


The automatic search for objects of interest is the main tool for working with the DFBS and is in fact its main research goal (together with the possibility to check any spectrum of an individual object). All DFBS fields will be surveyed for new active galaxies both by searching for optical counterparts of X-ray (ROSAT, Chandra and XMM), IR (IRAS, 2MASS and SST) and radio (NVSS and FIRST) sources, as well as searching on the basis of template spectra. Based on our previous work, we can estimate that some 10 candidate objects are expected from each field; in all, more than 10,000 from the whole survey. However, an inspection will be made of each spectrum, and the objects will be checked in all other databases to eliminate by-products as much as possible; thus we expect to be left with some 5,000 objects, new bright AGN candidates. A number of objects will be observed with the Byurakan Observatory 2.6m telescope to confirm their nature.


The DFBS catalog and database


A catalog of all DFBS objects with positional, photometric and spectral information (some 40,000,000 spectra corresponding to 20,000,000 objects) will be created after extraction of all objects from the plates. This will allow quick access to DFBS without extraction of large 2D images. It will be linked to most common databases (SIMBAD, NED, MAPS, USNO, etc.) to make the work with objects easier. However, the complete DFBS database will contain all data on objects and their spectra: low-dispersion spectral fields at high galactic latitudes, both 2D and 1D spectra of any object calibrated in wavelengths and flux, low-dispersion numerical classification, photometric UBVR estimates, links to other databases, etc. Many objects have 2 or more spectra from different FBS plates, so users can extract all of them to study, e.g., the variability. Each 1D spectrum will be presented as a small table of 107 rows corresponding to recorded pixel data. The DFBS catalog will be set appropriately for a search for definite types of objects in it by their magnitudes, colors, spectral features, etc.


The DFBS catalog and the spectra, and corresponding software will be written on DVDs (20 plates data on each). The whole DFBS database will occupy 100 DVDs. It will be kept both in Byurakan Observatory and Cornell University, as well as distributed to the main astronomical centers, and will be available through the Internet. This will allow an integration of the DFBS in the international databases. A user interface will give an access to database of 2D and 1D spectra, classification, using the DSS, MAPS, USNO, and other data, links to other databases, etc.




For further info and use of the DFBS contact Dr. Areg Mickaelian

2002-2004 Digitized First Byurakan Survey