TN 0025: Richmond's Analysis of Tom Droege's Feb 1997 images

Author: Michael Richmond
Date: 970228
Revision: #2 970303:  clarified local sky subtraction procedure, 
                      added note on astrometric accuracy
Key Words: CCD, astrometry, observation, techniques

Table of Contents

List of Tables


Introduction

Tom Droege acquired a set of images with a triplet at his home near Chicago (longitude approx 88:20, latitude approx 41:50) during the month of February, 1997. He posted a message describing his dataset to the TASS mailing list, and urged members to try their hand at reducing the images. The images are largely I-band, with a few V-band; unfortunately, there are no fields covered in both passbands.

I spent about a week, on and off, trying to reduce the data. My goal was to produce lists of detected stars for each image, calibrated both astrometrically and photometrically. As I struggled to meet this goal, I found many bugs and inadequacies in the XVista software package. This project certainly has improved XVista, and I hope to release an improved version sometime soon.

This document describes my first attempt to reduce the data. I discovered that my choice of a five-sigma threshold for object detection was conservative: there are clearly many more stars visible in the images than are included in my analysis here. There are other improvements that could be made in the reduction process, too, but I have decided that it's more important to describe my work than to spend more time trying to make it marginally better.

The images

Tom distributed a set of 20 images, consisting of

There are, therefore, a total of 15 images of the night sky, from two different nights. Note that the "dark" images were taken one week after any of the night-sky images.

Note also that there are no twilight-sky or "dome" flatfield images, which might yield better flatfield vectors than night-sky images. There was no choice but to use images of the night sky to try to flatten the target frames.

Below is a table showing the date, time, and central position for each frame.

                  Table 1.  Image names, positions, times
#     H = high galactic latitude
#     L = low galactic latitude
#
# sky values for images g0493946,g0493955 re-calculated with bin=2,
#     to avoid even/odd errors in my reduction procedures
#
# the "RA" and "Dec" values are equinox 2000 coordinates
#     at the rough center of each image
#                                                                         pixel
# name           date       time    camera    RA      Dec    sky   skysig FWHM
g0483967.fts  03-02-1997  11:13:48   0 I  H 202.34  -1.90   -27918    55   3.4
g0483977.fts  03-02-1997  11:27:04   0 I  H 205.67  -1.90   -28511    50   3.5
g0493669.fts  13-02-1997  04:03:37   0 I  L 104.16  -1.88   -27710   104   3.6
g0493678.fts  13-02-1997  04:16:53   0 I  L 107.49  -1.87   -27665   112   3.5
g0493946.fts  13-02-1997  10:42:37   0 I  H 204.18  -1.89   -28030    48   3.3
g0493955.fts  13-02-1997  10:55:55   0 I  H 207.52  -1.89   -27982    49   3.2
  
g1493669.fts  13-02-1997  04:03:37   1 V  L 119.10  -1.04   -25080   185   2.3
g1493678.fts  13-02-1997  04:16:53   1 V  L 122.42  -1.04   -25075   191   2.4
g1493946.fts  13-02-1997  10:42:37   1 V  H 219.13  -1.04   -26547    46   2.6
g1493955.fts  13-02-1997  10:55:55   1 V  H 222.46  -1.04   -26539    46   2.5

g2483921.fts  03-02-1997  10:07:23   2 I  H 199.54  -0.79   -28884    49   3.5
g2483931.fts  03-02-1997  10:20:42   2 I  H 202.90  -0.79   -28869    45   3.8
g2493623.fts  13-02-1997  02:57:18   2 I  L 101.40  -0.76   -26748    82   4.4
g2493900.fts  13-02-1997  09:36:02   2 I  H 201.36  -0.77   -28082    48   3.7
g2493909.fts  13-02-1997  09:49:22   2 I  H 204.71  -0.77   -28010    56   3.8

Tom used Norman Molhant's data acquisition program TM3GET11 to read the data from his triplet. Norman's program quite properly converts the 16-bit unsigned camera output into 16-bit signed integers before storing them in FITS files; it therefore places a keyword BZERO = 32768 into the FITS header. My XVista software has inherited an unfortunate feature from the distant past (when computers didn't have enough memory to fit an entire image into RAM at once): it performs all calculations in 16-bit signed integers, and so will lost the top bit if it converts the FITS values back into their original range of 0 to 65535. I therefore modified the FITS header of each image file so that it contained BZERO = 0. Readers will notice that I quote many "dark" and "sky" values which are large, negative integers; you may find the value more familiar if you add 32768.

Looking for bias/dark columns

Some CCD chips contain "extra" columns at the edge which are shielded from photons. These can be very useful in removing bias/dark current from a raw image, because they provide an example of the value a pixel would have if no light touched it -- in other words, the zero point of the intensity scale. Based on Kodak's technical specs for the KAF-0400 chip, we believe that there are such columns in TASS images. Norman's program produces two sets of such pixels: if we start counting columns with zero (i.e. 0, 1, 2, ...), they are

For each of the dark images (one for each CCD), I created a "row vector" by finding the median value along every column. Thus, from an original image which was 895 rows by 896 columns, I created a "dark vector" which was 1 row by 896 columns.

Clearly, these dark images have a consistent pattern: the pixel values on the left-hand side (low columns) are higher than the rest, decaying quickly to the typical value. Therefore, I concentrated on the "dark columns" on the right-hand side of the chip (high columns), hoping to find values which would be representative of the columns covering the "data area". I discovered that pixels in columns 783-791 seem to satisfy my requirements:

Note that the Kodak spec sheet claims that columns 781 and 782 should also be "dark"; however, when I examined some of the target images, I noticed that these columns had slightly higher values than the others in a row which featured a bright star. Apparently, some of the charge "leaks" through to these columns.

Therefore, I decided to use columns 783-791 as the "bias/dark columns" in all the images.

Here's the idea behind their use: it's quite possible for the "bias level" of a CCD camera to vary slowly, over the course of several hours/days/weeks. One source of such change is very subtle variations in the input voltage, temperature of the electronics, etc. However, one can hope that the shape of the "dark vector" will remain the same, and that changes in the overall bias level merely shift the "dark vector" up or down by a constant amount. So, if one can measure the "dark vector" at any one time, one can apply it to an image taken at some other time merely by shifting it up or down so that the pixels in the "bias/dark columns" of the "dark vector" match the values in the image.

Therefore, I created "dark vectors" for each CCD chip, using the dark images Tom provided; the values of each dark vector are shown in the plot above. For each one, I calculated the mean level inside the "bias/dark area" in columns 784-791, and stuffed this into the FITS header of the "dark vector". Given some other image, I planned to subtract the dark current as follows:

  1. read Md, the mean of medians in the "bias/dark area" of the "dark vector", from its FITS header
  2. calculate the median along columns of values in the image's "bias/dark area", just as we did when creating the "dark vector"
  3. calculate Mi, the mean of all these medians for the image
  4. set Delta = Mi - Md
  5. subtract the "dark vector" value from each column of the image
  6. subtract the constant Delta from the image
This procedure should yield a corrected image which has average value 0 in pixels not touched by light, and positive values where pixels were touched by light; the pixel value should be linearly related to the amount of light, with a proper zero point at 0.

In practice, I discovered that the values for Delta ranged from -125 to +162 ADU. Recall that the dark images were acquired at least one week after the other images.

Unfortunately, after all data reduction and analysis, as I was writing this summary, I discovered that I had made a mistake the final step of this process during the preparation of flatfield vectors: instead of subtracting Delta, I added it instead. This means that my flatfield vectors must be in error (probably by a few percent), and consequently all my "final" images also must be in error. Rats. I can re-process all the data in the future, but I will continue writing this report; the reader should remember that there is an error derived from the bias/dark subtraction which affects all the results reported here.

Is there a relationship between dark current and chip temperature?

We expect that the warmer a CCD chip is, the higher its dark current. Since Norman Molhant's data acquisition program TM3GET11 records the CCD temperature for each row of data, we can check to see if this relationship exists.

For each image, I measured the mean value inside the "bias/dark area", columns 784-791. I also measured the mean value in column 794, which is said to contain "CCD Temperature". Here's a graph of the "bias/dark" average versus the "CCD Temperture:"

There appears to be a correlation between the value of column 794 and the mean of the "bias/dark area." This would make sense if column 794 was proportional to the temperature. However, Norman's documentation states:

  pixel 794: resi is the thermistor's resistance in kiloohms
  pixel 795: resi is the thermistor's resistance in kiloohms
    resi = (196592400 / (ADUs - 32768)) - 1000;
    if (resi < 1300) resi = 1300;
    temperature = 136.55 - 77.246 * sqrt (log (resi) - 7.1671);
which implies that the pixel values are inversely proportional to temperture. So, if I'm interpreting all this correctly, the figure implies that the "bias/dark area" has larger pixel values when the CCD temperature decreases. Does this make sense?

Making flatfield vectors

Remember that I made quite an error in generating these flats.

Ideally, one should create flatfield vectors from some source other than images of the night sky, for several reasons. First, there are many stars in the night sky, which tend to contaminate measurements of the background value. Second, the night sky is dark (surprise!), and so the signal-to-noise ratio is lower than it would be in images of a bright source. One might try using images of the twilight sky to create flatfield vectors (although there can be large gradients in the illumination of the night sky over the 3-degree TASS field of view); or one might try placing a diffuse screen over the camera during the day. However, since neither option was available to me for this dataset, I used the night-sky images themselves to create flatfield vectors.

The method was relatively simple:

  1. subtract the "dark vector" from a raw image
  2. calculate the median value in each column This yields a "flatfield vector", which, like the "dark vector", has 1 row and 800 columns.

    Below are plots of the vectors derived in this way from each of Tom's raw images: for CCD 0, CCD 1, and CCD 2.

    All the vectors have the same large-scale appearance: pixel values are small on the left, increase to the right, reach a peak near the middle of the chip, then decrease slightly to the right edge. However, there are small-scale differences between vectors from different cameras, and vectors taken at different times. Note that the vectors taken from images at low galactic latitude, where the stellar density is high, show many little bumps and wiggles; these are caused by contamination of the true background level by many stars.

    I decided to make master flats for each camera, rather than use the median of each image to flatten itself (mostly because I wanted to avoid the little wiggles in the crowded fields). For each camera, I picked two consecutive images taken at high galactic latitude, and calculated the median vector, using all pixels in both images together.

    I then used the master flat for each camera to reduce all the night-sky images taken with that camera.

    Residual variations in the images

    When I finished with dark-subtraction and flatfielding, I discovered that there were still some large-scale features left in the images. Some of these are due to my error in creating the master flats (see above). Some of the variation, however, occurs in the other direction: variation from top-to-bottom within in image, which means variation with time. The images g0483967 and g0483977 clearly show large variations (around 400-500 ADU) over the course of the exposure, and there are smaller variations (around 50 ADU) in images g2493900 and g2493909.

    In order to make a more uniform background for the next step in reduction (finding stars), I decided to try to remove the large-scale variations in those images which showed large effects. I created a box-shaped filter with diameter 57 pixels, and moved it over these images, replacing each pixel value with the median value of pixels near the filter's edge. The result was a very smooth version of the original image, from which stars and other small-scale features had been removed. I then subtracted this smooth version from the original, yielding an image with a relatively flat sky background and unmodified small-scale features.

    These images were used only in the star-finding step, and not for photometry; I always used the "original" images to measure stellar magnitudes, even those with non-uniform backgrounds.

    I made no such modification to the images which seemed to have a flat sky background.

    For each frame, I determined an overall sky value by fitting a gaussian to the histogram of pixel values near its peak. In Table 1, you can see the values for the sky (the location of the peak of the gaussian) and the sky-sigma (the width of the gaussian).

    For a couple of the images (g0493946, g0493955), I discovered that my reduction procedures had introduced a weird pattern in the final pixel values. The sky histogram appeared to be two interleaved distributions, one much higher than the other: somehow, the reduction preferentially yielded either even or odd integer values. In order to measure the sky value properly, I binned the pixel values into bins of width 2 ADU, so that the even/odd differences would disappear.

    Finding stars in the images

    Once I had a clean, reduced frame for each field, I measured the FWHM of several bright, isolated stars interactively. My program calculated a local sky value in an annulus around the star I chose, subtracted the sky from all pixels within a small (5-pixel) radius around the cursor position, and calculated a central position by fitting a gaussian to the marginal sum in both directions. Using this center, the routine created a 1-D radial profile for each pixel within 5 pixels, and then fit a gaussian to that profile. I wrote down the FWHM values for several stars, typically finding a range of 0.4 pixel or so; I chose the center of the range as the characteristic FWHM for the image. You can find that value in Table 1.

    I got the impression in several images that stars near the left-hand side of the frame (the Northern side) were a bit sharper than those in the middle, or near the right-hand (Southern) side. This may just be coincidence.

    I ran a few preliminary tests to find stars. Examining the output, I found that my software measured values for FWHM which varied quite a bit within a single frame; faint stars usually had smaller widths than bright ones. I decided that my software might find widths that varied from the "typical" value by up to +/- 1 pixel.

    I used the stars program within the XVista software package to detect sources in each image. I set it to find

    As mentioned earlier, the threshold I selected, five times the standard deviation of the sky, is a conservative one. I noticed quickly that some faint stars, clearly visible to the eye, were not being detected. Nevertheless, I found enough stars to exercise my reduction and analysis procedures. Keep this in mind when you read about the limiting magnitude below.

    The number of stars detected in each image varied greatly with galactic latitude, but was about the same in the V and I passbands:

                 Table 2: number of stars detected
    
    image       filter   subtract smoothed?    latitude      number of stars
    --------------------------------------------------------------------------
    g0483967      I              yes             high             476
    g0483977      I              yes             high             467
    g0493669      I               no              low            1401
    g0493678      I               no              low            1333
    g0493946      I               no             high             494
    g0493955      I               no             high             543
    
    g1493669      V              yes              low            1400
    g1493678      V              yes              low            1272
    g1493946      V               no             high             456
    g1493955      V               no             high             547
    
    g2483921      I               no             high             481
    g2483931      I               no             high             545
    g2493931      I              yes              low            1315
    g2493900      I              yes             high             559
    g2493909      I              yes             high             526
    

    The stars program measures the position of each star as follows:

    I suspect that one might identify an algorithm which does a better job on TASS images (which might have asymmetric point-spread functions). I also suspect that one might implement this particular routine in a better fashion than I have done: for example, I weight each element in the marginal sums equally.

    Measuring instrumental magnitudes

    I used the phot program in the XVista software package to measure aperture magnitudes for each star detected in the previous step. There are a number of choices I made here:

    Here are the aperture sizes I used for Tom's images:

                          Table 3.  Aperture sizes
    
         image         image         aperture radii (pixels)
    -----------------------------------------------------------------
      g0483967       g0483977        3.5, 3.0, 4.5
      g0493669       g0493678        3.5, 3.0, 4.5
      g0493946       g0493955        3.2, 2.7, 4.2
      
      g1493669       g1493678        2.4, 1.9, 3.4
      g1493946       g1493955        2.6, 2.1, 3.6
    
      g2483921       g2483931        3.7, 3.2, 4.7
      g2493623                       4.5, 4.0, 5.5
      g2493900       g2493909        3.8, 3.3, 4.8
    

    After I had processed all the star lists to extract instrumental magnitudes, I realized that the estimated uncertainty in magnitude was always listed as "99.00", a special value which means that the uncertainty could not be calculated. I must have introduced an error somewhere in the process of estimating this uncertainty. Certainly, should I choose to re-reduce these images, I will fix the problem; but, as results below show, the extracted magnitudes themselves are properly calculated.

    One quick test of the accuracy of the procedure which extracts magnitudes is to compare the instrumental magnitudes of stars which appear in the "overlap area" of two consecutive images. Since these are not only the same stars, but the same raw pixel values, one expects them to yield nearly identical magnitudes; there may be small differences in the dark subtraction, flatfielding and perhaps sky subtraction between the two. It was possible to find large samples of stars in the overlap area only for the images near the galactic plane.

                   Table 4. Check on instrumental magnitudes
    
                                    all stars         bright stars
       image      image    filt     N delta sig        N delta sig
     ---------------------------------------------------------------
       g0493669  g0493678    I     27 -0.05 0.15      14 -0.01 0.02
       g1493669  g1493678    V     16  0.01 0.04       8  0.00 0.01
    
    Here, delta is the average difference in instrumental magnitudes for a set of N stars in the overlap area, and sig is the standard deviation from the mean. I report the differences for all stars in common, and for the brighest half of all stars in common. Not surprisingly, the bright stars yield much smaller difference and scatter than bright+faint stars.

    Astrometric calibration

    Once I had a list of stars, with measured (row, col) position and an instrumental magnitude, I could try to match the stars with those from some astrometric catalog. My approach is a very simple one: I assume that there is a transformation of the following sort:

            x = A + B*row + C*col
            y = D + E*row + F*col
    
    where (x, y) are the spherical coordinates of a star projected onto a plane tangent to the center of the field. This approximation is commonly used, and is certainly appropriate for the small fields of view (5-15 arcmin) found in most CCD cameras. Since the TASS images span more than 3 degrees, however, it is likely that this transformation may yield very high accuracy over the entire field.

    I had in the past written a program that attempted to find the position on the celestial equator from which a star list was created; this was the automatic astrometric calibration by E-mail facility. Over the past few months, I've discovered that this works only about half the time; for some fields, there just isn't enough overlap between the brightest 50 or so stars detected in an image, and the small catalog of stars (a subset of the Guide Star Catalog) for which I have pre-calculated some matching information.

    This set of images was no exception: some of them were successfully identified by the automatic procedure, some weren't. Therefore, I created a somewhat simpler matching procedure that relies upon the user to supply a decent position for the frame. Since Norman Molhant's data acquisition program TM3GET11 places an approximate RA and Dec into the header of each FITS image, I think this should not be hard to extract in some automated way.

    Using this simple procedure, I was able to match (row, col) positions for stars in each image to (RA, Dec) values for stars in the Hubble Guide Star Catalog. A rough position for the center of each image, equinox 2000, is listed above in Table 1. Below I list properties of the astrometric solution for each field.

                         Table 5. Astrometric calibration
    
      image          B        C        E        F        scale    Nstar   sigma
    -----------------------------------------------------------------------------
      g0483967    13.6932  -0.0551  -0.0035 -13.7403    13.7167    426    4.7 
      g0483977    13.6849  -0.0458   0.0003 -13.7442    13.7145    404    4.1 
      g0493669    13.6847  -0.0588  -0.0098 -13.7454    13.7150   1154    5.8 
      g0493678    13.6904  -0.0478  -0.0055 -13.7430    13.7167   1141    5.6 
      g0493946    13.7205  -0.0509   0.0064 -13.7506    13.7355    455    4.5
      g0493955    13.7257  -0.0417   0.0006 -13.7488    13.7372    475    4.6
    
      g1493669    13.6927   0.0454  -0.0012 -13.6487    13.6707   1009    5.2
      g1493678    13.7036   0.0434  -0.0018 -13.6496    13.6766    915    5.1
      g1493946    13.7360   0.0514   0.0014 -13.6510    13.6934    336    4.4
      g1493955    13.7249   0.0488  -0.0012 -13.6568    13.6908    405    4.9
    
      g2483921    13.7549  -0.0446   0.0034 -13.8192    13.7870    424    4.5
      g2483931    13.7424  -0.0370  -0.0010 -13.8182    13.7803    471    4.3
      g2493623    13.7006  -0.0500  -0.0063 -13.8182    13.7593   1050    7.1
      g2493900    13.7649  -0.0420   0.0012 -13.8234    13.7942    432    4.4
      g2493909    13.7535  -0.0356   0.0025 -13.8189    13.7861    449    4.4
    
    
    Here, the coefficients BCEF of the TRANS are listed, as well as the overall plate scale derived via the formula
             scale = sqrt { abs(B)*abs(F) + abs(C)*abs(E) }
    
    One can derive the angle of rotation between the chip's (row, col) coordinate system and the catalog's (RA, Dec) from these coefficients, but I have not done so. The scale is given in units of arcseconds per pixel. The results for each camera are reasonably consistent, although the slight jump in scale on CCD 0 between early and late times on the same night of Feb 12, 1997, is puzzling.

    I accepted any stars which fell within a distance D=15 arcsec of a corresponding star in the GSC as "true matches." The penultimate column lists the number of "true matches" found in common on the image and in the GSC, from which the solution was derived. The final column lists the standard deviation between the measured position and the catalog position, in arcseconds.

    The values in the final column are disturbingly large. They correspond to a precision of about 0.3 pixels. Now, a general rule of thumb is that one can find the centroid of a well-sampled, symmetric PSF to at least 0.1 pixels, and sometimes much less. There are three reasons that I might have calculated such imprecise positions:

    1. the simple transformation between (row, col) and (RA, Dec) breaks down over the large field (4x3 degrees) of each image
    2. the PSF is not symmetric, and/or varies significantly across the frame
    3. the code I used to calculate the position of each star contains some error, or the algorithm (gaussian fitted to marginal sums) is not appropriate for these images

    Let me test the first hypothesis. If the errors are caused by distortions from one edge to the other of the field of view, I would expect them to be small in a sub-section of the entire image. I choose image g0493669 for testing, and extracted all the stars which fell within a box of size 200 rows and 200 columns, near the center of the image. Matching these stars alone to the GSC, I found

    Since the first result is very similar to the scatter between matches over the whole field (as shown in Table 5 above), I conclude that wide-field distortion is not the dominant source of error.

    Another way to check if the scatter is caused by distortion is to compare the raw (row, col) positions measured from two different images for stars in the same part of the sky. One would expect that the distortion should be similar in each image, and therefore the scatter should be smaller in this comparison. I used star lists from images g2483931 and g2493900, both taken with the same camera.

    I conclude that the unexpectedly large scatter in measured position is not primarily due to distortion across the field of view, but perhaps caused by some (variable?) asymmetry in the PSF, or perhaps due to errors in my software.

    Note added Mar 3, 1997: Arne Henden points out that some of the GSC stars I used were much fainter than any star visible in the images; hence, some of the matches were spurious. He suggested restricting the GSC sample to bright stars, mag < 13. When I tried this on one field, g2493900, I found I could decrease the scatter in position from 4.4 to 2.9 arcsec.

    Photometric calibration

    Once each star list has its positions calibrated into (RA, Dec), one can compare it to a catalog of standard stars to find any standards which appears in the list. I have created a list of 669 stars with UBVRI measurements from papers written by Arlo Landolt:

    I've precessed all coordinates from these papers to equinox 2000, and used values from the 1992 paper in preference to those from the 1983 paper. I would like to make this compilation public sometime, but Arlo Landolt asked me not to do so with an earlier table made from the same papers. Perhaps readers could ask him, very politely, to allow us to put this list (or an equivalent) on the TASS home page. I can supply his address upon request.

    Fortunately for us, each consecutive pair of Tom's images falls in one of the Selected Areas containing a photometric sequence of Landolt stars. Since I used the same aperture to extract instrumental magnitudes from consecutive images, I can transfer the photometric zero-point from one image to its neighbor with some hope of avoiding large errors.

    Recall that the night of February 2, 1997, had occasional clouds. The images g0483967, g0483977, g2483921 and g2483931 therefore may be expected to have relatively poor photometry.

    Unfortunately, there is no field for which Tom provided both V and I images. Therefore, the best one can do to calibrate these data photometrically is to assume that there are no color effects (i.e., that the TASS V passband matches the Landolt V passband perfectly, and TASS I matches Landolt I); with this assumption, one calibrates by

    Now, as mentioned above, the threshold I chose for finding stars in these images was pretty high. Therefore, although a typical Selected Area might have 10 Landolt standards, which would all fit entirely within a single TASS image, I found only a few Landolt standards per field. I suspect strongly that lowering the threshold would yield more standards.

    For each pair of consecutive images, I tried to match the star lists with the Landolt catalog. For each aperture (recall that I used 3 different aperture sizes to extract instrumental magnitudes from each image), I calculated the mean difference Dmag between instrumental and Landolt magnitudes, as well as the standard deviation sig of the differences from the mean.

                           Table 6. Photometric calibration
    
      
                                            aper R       aper R-0.5   aper R+1.0
        image    filt    field   Nstar     Dmag  sig     Dmag  sig     Dmag sig
    ------------------------------------------------------------------------------
      g0483967    I                0
      g0483977    I      SA105     2      4.021 0.024   4.125 0.041   3.887 0.004
    
      g0493669    I      SA98      2      3.677 0.003   3.793 0.003   3.523 0.003
      g0493678    I                0
    
      g0493946    I      SA105     3      4.038 0.024   4.175 0.024   3.845 0.027
      g0493955    I                0
    
      g1493669    V      SA99      6      3.695 0.021   4.084 0.028   3.576 0.025
      g1493678    V                0
    
      g1493946    V      SA106     3      3.728 0.057   3.875 0.072   3.594 0.037
      g1493955    V                0
    
      g2483921    I                0
      g2483931    I      SA105     3      3.914 0.011   4.030 0.010   3.762 0.012
     
      g2493623    I      SA98      1      3.569         3.695         3.369
    
      g2493900    I                0
      g2493909    I      SA105     2      3.853 0.059   3.957 0.071   3.707 0.044
    
    It would, of course, be nicer to have more than 2 or 3 (or 1!) standard per field, but this suffices for the current needs. Note that the typical agreement between measured and Landolt magnitudes, after making a single offset, is typically 0.05 magnitudes, or about 5 percent. This is similar to accuracies I've derived for other sets of TASS data (see Technical Note 19, for example), but is probably not the best one can do.

    Note also that there is little difference between the accuracy provided by each of the three apertures. It's nice to have evidence that the exact size of the aperture isn't critical.

    I then applied the zero-points listed above to the instrumental magnitudes for all stars in all images.

                      Table 7. Distribution of measured magnitudes
    
      
     name       filter  lat     6    7    8    9   10   11   12   13   14   15 
    ----------------------------------------------------------------------------
    g0483967      I      H      4    7   18   26   71  131  199   20    0    0 
    g0483977      I      H      2    6   12   38   65  132  183   28    1    0 
    g0493669      I      L      2   16   48  154  401  512  239   23    4    1 
    g0493678      I      L      0   20   63  158  358  469  239   21    3    2 
    g0493946      I      H      5    9   16   30   71  134  175   53    1    0 
    g0493955      I      H      0    4   15   44   79  140  211   50    0    0 
    
    g1493669      V      L      0    0   23   38  118  278  522  386   35    0 
    g1493678      V      L      0    1   18   42  117  263  402  396   33    0 
    g1493946      V      H      0    0    7   17   39   78  144  162    9    0 
    g1493955      V      H      0    3    5   23   45   96  159  194   22    0 
    
    g2483921      I      H      2    5    9   24   65  139  212   24    1    0 
    g2483931      I      H      3   10   15   38   68  165  210   36    0    0 
    g2493623      I      L      5   24   55  169  407  461  186    6    0    2 
    g2493900      I      H      0    9   12   36   70  140  232   60    0    0 
    g2493909      I      H      2    5   13   30   66  158  207   45    0    0 
    
    In rough terms, the number of stars detected rises to mag 11 or 12, then drops pretty sharply. Let me re-iterate that I used a relatively high threshold for detecting stars, so that the true limiting magnitude for these images is somewhat fainter than this table would suggest.

    There are a number of the I-band images (but not the V-band images) which cover common areas of the sky. One way to assess the accuracy of photometry is to compare the magnitudes of the same star in two different images. I have done that for all pairs of images covering the same area, and list the results in the table below. I give the differences between all stars in common, and between "bright" (I < 11.0) and "faint" (I >= 11.0) subsamples. There are three tables, one for each choice of aperture size (but the results are very similar in each).

          Table 8. Comparison of magnitudes of from different images,
                   aperture radius = FWHM pixels
    
    #                         all stars            (I < 11.0)        (I >= 11.0)
    # image1   image2 filt   N    mu   sig      N    mu   sig      N    mu    sig
    ------------------------------------------------------------------------------
    g0483967 g0493946  I   186  0.028 0.080    92  0.024 0.036    94  0.032 0.107
    g0483967 g2483921  I    22  0.101 0.101     9  0.164 0.026    13  0.058 0.112
    g0483967 g2483931  I   225  0.101 0.106    97  0.100 0.068   128  0.102 0.128
    g0483967 g2493900  I   175  0.081 0.115    74  0.088 0.057   101  0.075 0.144
    g0483967 g2493909  I    83  0.011 0.083    35  0.006 0.042    48  0.015 0.104
    
    g0483977 g0493946  I   201 -0.005 0.082    89  0.004 0.026   112 -0.011 0.107
    g0483977 g0493955  I   173  0.039 0.081    87  0.053 0.062    86  0.026 0.096
    g0483977 g2483931  I    50  0.014 0.090    24  0.011 0.043    26  0.018 0.119
    g0483977 g2493909  I   164 -0.025 0.094    82 -0.015 0.048    82 -0.036 0.123
    
    g0493669 g2493623  I    63  0.080 0.072    48  0.085 0.047    15  0.064 0.124
    
    g0493946 g2483931  I   162  0.036 0.082    77  0.029 0.038    85  0.043 0.107
    g0493946 g2493900  I    38  0.014 0.079    22 -0.002 0.041    16  0.035 0.111
    g0493946 g2493909  I   212 -0.020 0.080    99 -0.011 0.040   113 -0.028 0.103
    
    g0493955 g2493909  I    43 -0.051 0.076    24 -0.027 0.053    19 -0.082 0.089
    
    g2483921 g2493900  I   189 -0.059 0.078    78 -0.055 0.028   111 -0.062 0.100
    
    g2483931 g2493900  I   270 -0.049 0.080   117 -0.043 0.029   153 -0.053 0.103
    g2483931 g2493909  I   190 -0.039 0.075    91 -0.038 0.027    99 -0.040 0.100
    ------------------------------------------------------------------------------
    
    
          Table 9. Comparison of magnitudes of from different images,
                   aperture radius = FWHM-0.5 pixels
    
    #                         all stars            (I < 11.0)        (I >= 11.0)
    # image1   image2 filt   N    mu   sig      N    mu   sig      N    mu    sig
    ------------------------------------------------------------------------------
    g0483967 g0493946  I   186  0.029 0.074    93  0.027 0.036    93  0.030 0.099
    g0483967 g2483921  I    22  0.123 0.096     9  0.178 0.030    13  0.084 0.107
    g0483967 g2483931  I   225  0.113 0.104    97  0.109 0.071   128  0.116 0.123
    g0483967 g2493900  I   175  0.083 0.108    74  0.087 0.059   101  0.081 0.133
    g0483967 g2493909  I    83  0.008 0.085    36 -0.001 0.044    47  0.014 0.106
    
    g0483977 g0493946  I   201  0.002 0.075    90  0.009 0.027   111 -0.004 0.098
    g0483977 g0493955  I   173  0.048 0.075    87  0.060 0.062    86  0.035 0.085
    g0483977 g2483931  I    50  0.021 0.086    25  0.016 0.045    25  0.025 0.115
    g0483977 g2493909  I   164 -0.026 0.091    83 -0.016 0.049    81 -0.037 0.119
    
    g0493669 g2493623  I    63  0.124 0.064    48  0.129 0.045    15  0.109 0.103
    
    g0493946 g2483931  I   162  0.041 0.079    77  0.035 0.040    85  0.047 0.103
    g0493946 g2493900  I    38  0.013 0.076    22 -0.007 0.037    16  0.039 0.105
    g0493946 g2493909  I   212 -0.026 0.080    99 -0.018 0.041   113 -0.033 0.102
    
    g0493955 g2493909  I    43 -0.052 0.071    24 -0.038 0.052    19 -0.070 0.087
    
    g2483921 g2493900  I   189 -0.075 0.073    78 -0.074 0.027   111 -0.075 0.092
    
    g2483931 g2493900  I   270 -0.060 0.076   119 -0.056 0.027   151 -0.064 0.098
    g2483931 g2493909  I   190 -0.050 0.067    92 -0.050 0.026    98 -0.050 0.090
    ------------------------------------------------------------------------------
    
    
          Table 10. Comparison of magnitudes of from different images,
                    aperture radius = FWHM+1.0 pixels
    
    #                         all stars            (I < 11.0)        (I >= 11.0)
    # image1   image2 filt   N    mu   sig      N    mu   sig      N    mu    sig
    ------------------------------------------------------------------------------
    g0483967 g0493946  I   186  0.018 0.092    93  0.007 0.038    93  0.029 0.124
    g0483967 g2483921  I    22  0.058 0.112     9  0.141 0.028    13  0.001 0.113
    g0483967 g2483931  I   225  0.079 0.114    97  0.078 0.067   128  0.080 0.140
    g0483967 g2493900  I   175  0.066 0.133    73  0.082 0.055   102  0.055 0.167
    g0483967 g2493909  I    83  0.003 0.093    36 -0.003 0.045    47  0.008 0.118
    
    g0483977 g0493946  I   201 -0.021 0.097    90 -0.014 0.030   111 -0.027 0.128
    g0483977 g0493955  I   173  0.018 0.094    88  0.031 0.062    85  0.005 0.117
    g0483977 g2483931  I    50 -0.011 0.096    24 -0.004 0.046    26 -0.018 0.127
    g0483977 g2493909  I   164 -0.030 0.104    83 -0.021 0.047    81 -0.039 0.141
    
    g0493669 g2493623  I    63 -0.005 0.088    48 -0.004 0.060    15 -0.006 0.149
    
    g0493946 g2483931  I   162  0.030 0.089    76  0.026 0.035    86  0.033 0.118
    g0493946 g2493900  I    38  0.030 0.089    22  0.001 0.045    16  0.069 0.117
    g0493946 g2493909  I   212 -0.014 0.087    97  0.000 0.043   115 -0.026 0.110
    
    g0493955 g2493909  I    43 -0.044 0.090    24 -0.009 0.069    19 -0.088 0.097
    
    g2483921 g2493900  I   189 -0.040 0.093    78 -0.037 0.027   111 -0.042 0.119
    
    g2483931 g2493900  I   270 -0.036 0.095   116 -0.029 0.038   154 -0.042 0.122
    g2483931 g2493909  I   190 -0.030 0.086    92 -0.029 0.031    98 -0.032 0.116
    ------------------------------------------------------------------------------
    

    To summarize the information in the above tables very briefly, the scatter between independent measurements of the same stars in different images is about 0.05 magnitudes for "bright" stars (with I < 11.0), and about 0.10 magnitudes for "faint" stars (with I > 11.0).

    Looking just a little bit deeper, one can see that, for the "faint" star subsample, large apertures give slightly smaller offsets and larger scatter than small apertures, which is just what one would expect from normal sources of error.

    Conclusions

    Here I list the major points of my analysis for quick review.

    1. The pixels in columns (zero-indexed) 784-791 may be good sources of a bias/dark level
    2. There is a puzzling anti-correlation between CCD temperature (as recorded in column 794) and bias/dark level
    3. I goofed when constructing flatfield vectors for this analysis (improperly removing the bias/dark contribution)
    4. The FWHM of the images range from 2.2 to 4.4 pixels. The V-band images all have smaller FWHM than any I-band image, and there may be a tendency for the focus to be best near the north edge of the frame
    5. I used a 5-sigma threshold to find stars, which does not find all the stars which one can see clearly by eye, but nonetheless produces 400-1200 stars per frame
    6. I measured instrumental magnitudes for each image, using several apertures, the main one having a radius equal to the FWHM
    7. Astrometric calibration against the GSC yields positions with a precision of about 4-5 arcsec. This is about one-third of a pixel, which is larger than one would expect. The scatter is probably not due to distortion across the wide field of view. Using a magnitude cutoff for GSC reference stars causes the precision to improve to less than 3 arcsec.
    8. There are Landolt standards in each pair of images which can be used for simple, zero-point photometric calibration; however, since the set does not include images in both V and I of the same field, one cannot check for color terms
    9. Comparison of instrumental magnitudes against each other, or against the Landolt stars, reveals a precision of about 0.05 magnitudes for bright stars, about 0.10 magnitudes for faint ones

    You can find pointers to the raw images upon which this work was based, and the calibrated star lists I created, on Data Archive for Tom Droege's February 1997 images.


    Back to Index of Technical Notes
    Back to TASS home page