path: root/noao/twodspec/longslit/doc/lslit.ms

.nr PS 9
.nr VS 10
.ps 9
.vs 10
.po 0.50i
.nr PO 0.50i
.ll 7.0i
.nr LL 7.0i
.nr PD 1v
.EQ
delim	$$
.EN
.TL
Reduction of long slit spectra with IRAF
.AU
Francisco Valdes
.AI
IRAF Group, Central Computer Services, National Optical Astronomy Observatories
P.O. Box 26732, Tucson, Arizona, 85726
March 1986
.AB
Tools for the reduction of long slit spectra within the Interactive
Data Reduction and Analysis Facility (IRAF) at the National Optical
Astronomy Observatory (NOAO) are described.  The user interface
(commands and special features) and the algorithms are discussed.
Application of the reduction package to multi-slit images is briefly
outlined.  The author developed and supports the package at NOAO.
.AE
.LP

.ce
\fB1. Introduction\fR
.PP
This paper describes the tools currently available within the Interactive Data
Reduction and Analysis Facility (IRAF) at the National Optical
Astronomy Observatories (NOAO) for the reduction of long slit spectra.
The reduction tools, called tasks, are organized as an IRAF package
called \fBlongslit\fR.  The tasks in the package are summarized below.

.TS
center;
n n.
apdefine \&- Define apertures for 1D aperture extraction	identify \&- Identify features
apextract \&- Extract 1D aperture spectra	illumination \&- Determine illumination calibration
background \&- Fit and subtract a line or column background	reidentify \&- Reidentify features
extinction \&- Apply atmospheric extinction corrections to images	response \&- Determine response calibration
fitcoords \&- Fit user coordinates to image coordinates	setimhdr \&- Set longslit image header parameters
fluxcalib \&- Apply flux calibration to images	transform \&- Transform longslit images to user coordinates
.TE

.PP
Since there are many types of long slit spectra, detectors, and
astronomical goals we do not describe a reduction procedure or path.
Reduction manuals giving cookbook instructions for the reduction of
certain types of data at NOAO are available from the Central Computer
Services Division.  Instead, each task is discussed separately.  The
primary emphasis is on the algorithms.
.PP
The following terminology is used in this paper.  A \fIlong slit
spectrum\fR is a two dimensional image.  The two image axes are
called \fIaxis 1\fR and \fIaxis 2\fR and the pixel coordinates are
given in terms of \fIcolumns\fR and \fIlines\fR.  The long slit
axes are called the \fIdispersion axis\fR and the \fIslit
axis\fR.  The reduction tasks do not require a particular orientation
of the dispersion and slit axes, however, these axes should be
fairly closely aligned with the image axes.  \fBIn the remainder of
this paper the slit axis will correspond to image axis 1 and
the dispersion axis with image axis 2\fR.
.PP
There are five types of operations performed by the tasks in the
\fBlongslit\fR package: (1) detector response calibration, (2) geometric
distortion and coordinate rectification, (3) background sky subtraction,
(4) flux calibration, and (5) aperture extraction of one dimensional spectra.
These are listed in the order in which they are usually performed and in
which they are discussed in this paper.  There is also an initialization
task, \fBsetimhdr\fR, and a general routine, \fBicfit\fR, used in may of the
long slit tasks.  These are described first.
.SH
SETIMHDR - Set long slit image header parameters
.PP
The tasks in the \fBlongslit\fR package use information contained in the IRAF
image header.  The task \fBsetimhdr\fR sets a required parameter in the image
header advising the long slit tasks which image axis corresponds to the
dispersion axis; the tasks work equally well with the dispersion axis
aligned with the image lines or the image columns.  This is generally
the first task executed when reducing long slit spectra.
.SH
ICFIT - The IRAF Interactive Curve Fitting routine
.PP
Many of the tasks in the IRAF which fit a one dimensional function
utilize the same powerful interactive curve fitting routine called
\fBicfit\fR.  This routine allows the user to perform sophisticated
function fitting interactively and graphically or to specify the
function fitting parameters in advance and run the task
non-interactively.  That this routine is used in many tasks also has
the advantage that the user need not learn a new set of commands and
features for each task requiring function fitting.
.PP
The features of the this curve fitting tool include:
.IP (1)
A choice of four fitting functions; Chebyshev polynomial, Legendre polynomial,
a linear spline, and a cubic spline.
.nr PD 0v
.IP (2)
A choice of the polynomial order or the number of spline pieces.
.IP (3)
Deletion of individual points from the fit.
.IP (4)
Selection of a sample or subset of points to be fit (excluding the rest).
.IP (5)
Iterative deletion of points with large residuals from the fitted function.
.IP (6)
Binning sets of neighboring points into averages or medians which are then
fit instead of the individual points.
.nr PD 1v
.LP
In addition to the above features the interactive graphics mode allows
the user to:
.IP (1)
Iterate any number of times on the fitting parameters.
.nr PD 0v
.IP (2)
Display the fit in several different ways; residuals, ratios, and the fit
overplotted on the data points.
.IP (3)
Manipulate the graphs using a large set of commands for formating and
expanding any part of a graph for detailed examination.
.IP (4)
Produce copies of the graphs with a snap-shot command.
.nr PD 1v
.PP
For the applications described in this paper the most important features
are the ability to adjust the function order, exclude bad points, and
select subsets of points to be fit.  Other useful features are taking the
median or average of a set of points before fitting and iteratively
rejecting deviant points.  When used non-interactively the user
selects the function and the order.  The \fBlongslit\fR tasks using the
interactive curve fitting routine are \fBbackground\fR, \fBidentify\fR,
\fBillumination\fR, and \fBresponse\fR.


.ce
\fB2. Detector Response Calibrations\fR
.PP
The relative response of the pixels in the detector and the transmission
of the spectrograph along the slit are generally not uniform.  Outside
of the \fBlongslit\fR package are IRAF tasks for creating \fIflat fields\fR
from quartz lamp calibration images which correct for small scale response
variations.  Flat fields, however, do not correct for spectrograph
transmission variations or any large scale response patterns.  The tasks
\fBresponse\fR and \fBillumination\fR are specially designed for long slit
spectra to correct both the small scale variations as well as
larger scale response patterns and slit illumination and transmission effects.
.PP
These algorithms make the assumption that the wavelength and slit axis
are very nearly aligned with the image lines and columns.  If this is
not true then the images must be aligned first or alternate response
calibration methods used.
.SH
RESPONSE - Determine response calibration
.PP
The task \fBresponse\fR is used with calibration images which (1)
do not have any intrinsic structure along the slit dimension and (2)
have a smooth spectrum without emission or absorption features.
Typically the calibration images consist of quartz lamp exposures.
The idea is to determine a response correction that turns an observed
calibration image into one which is identical at all points along the
slit.
.PP
From (1) a one dimensional spectrum is obtained by averaging along the
slit; i.e. averaging the columns.  Based on (2) a smoothing function is
fit to the one dimensional spectrum to reduce noise and eliminate
response effects which are coherent in wavelength such as fringing.
The response correction for each pixel is then obtained by dividing
each point along the slit (the columns) by the smoothed one dimensional
spectrum.
.PP
The purpose of fitting a function to the one dimensional spectrum is to
reduce noise and to remove coherent response effects which are not part
of the true quartz spectrum.  Examples of coherent response effects are
fringing and regions of low or high response running along the slit
dimension which are, therefore, not averaged out in the one dimensional
spectrum.  The choice of smoothing function is dictated by the behavior
of the particular detector.  Difficult cases are treated with the
interactive graphical function fitting routine \fBicfit\fR.  For the
automated case the user specifies the smoothing function and order.
.PP
This calibration algorithm has the advantage of removing spatial
frequencies at almost all scales; in particular, there is no modeling
of the response pattern along the slit dimension.  The only modeling is
the fit to the \fBaverage\fR spectrum of the calibration source.  In
tests at NOAO this algorithm was able to reduce the response variations
to less 0.2%, to correct for a broad diagonal region of low response in
one of the CCD detectors (the CRYOCAM), and to remove strong fringing
in spectra taken in the red portion of the spectrum where the detector
is particularly subject to fringing.
.PP
One feature common to \fBresponse\fR and \fBillumination\fR is that
the algorithm can be restricted to a section of the calibration image.
The response corrections are then determined only within that section.
If a response image does not exist initially then the response values outside
the section are set to unity.  If the response image does exist then
the points outside the section are not changed.  This feature is used
with data containing several slits on one image such as produced by
the multi-slit masks at Kitt Peak National Observatory.
.PP
When there are many calibration images this algorithm may be applied to
each image separately or to an average of the images.  If applied
separately the response images may be averaged or applied to the
appropriate long slit spectra; typically the one nearest the object
exposure in time or telescope position.  The task allows a list of
calibration images from which a set of response corrections is
determined.
.PP
Figure 1 shows a portion of an average quartz spectrum ratioed with the
smooth fit to the spectrum.  It is one of the graphs which can be
produced with the \fBicfit\fR routine and, with the other figures in
this paper, illustrates the formating,
zooming, and snap-shot capabilities in IRAF.  The figure shows considerable
structure of periodic high response lines and fringing which, because
they are primarily aligned with the image lines, are still present in
the average quartz spectrum.  Note that this is not the response
since it is the average of all the columns; an actual response column
would have much larger variations including pixel-to-pixel response
differences as well as large scale response patterns such as the diagonal
structure mentioned previously.
.SH
ILLUMINATION - Determine illumination calibration
.PP
The task \fBillumination\fR corrects for large scale variations along
the slit and dispersion dimensions due to illumination or spectrograph
transmission variations (often called the \fIslit profile\fR).  When
the detector response function is determined from quartz calibration
images, using \fBresponse\fR, an illumination error may be introduced
due to differences in the way the spectrograph is illuminated by the
quartz lamp compared to that of an astronomical exposure.  This
violates the the assumption that the calibration spectrum has no
intrinsic structure along the slit.  \fBIllumination\fR is also used
when only the small scale response variations have been removed using a
flat field correction.
.PP
The approach to determining the response correction is similar to that
described for \fBresponse\fR.  Namely, the response correction is the
ratio of a calibration image to the expected calibration image.  Again,
the expected calibration image is that which has no structure along the
slit.  Calibration images may be quartz lamp exposures, assuming there
is no illumination problem, and blank sky exposures.  In the worst
case, object exposures also may be used if the extent of the object in
the slit is small.
.PP
There are several important differences between this algorithm and that
of \fBresponse\fR:
.IP (1)
The spectra are not required to be smooth in wavelength and may contain
strong emission and absorption lines.
.nr PD 0v
.IP (2)
The response correction is a smooth, large scale function only.
.IP (3)
Since the signal-to-noise of spectra from blank sky and object images is
lower than quartz calibration images, steps must be taken to minimize noise.
.IP (4)
Care must be taken that the spectral features do not affect the
response determination.
.nr PD 1v
.PP
The algorithm which satisfies these requirements is as follows.  First the
calibration spectrum is binned in wavelength.  This addresses the
signal-to-noise consideration (3) and is permitted because only large
scale response variations are being determined (2).  Next a smoothing
function is fit along the slit dimension in each bin; i.e. each
wavelength bin is smoothed to reduce noise and determine the large
scale slit profile.  Then each bin is normalized to the central point
in the slit to remove the spectral signature of the calibration image.
Finally, the binned response is interpolated back to the
original image size.
.PP
The normalization to the central point in the slit is an assumption
which limits the ability of the illumination algorithm to correct
for all wavelength dependent response effects.  There is a wavelength
dependence, however, in that the slit profile is a function of the
wavelength though normalized to unity at the central point of the
slit.
.PP
The wavelength bins and bin widths need not be constant.  The bins are
chosen to sample the large scale variations in the slit profile as a
function of wavelength, to obtain good signal statistics, and to avoid
effects due to variations in the positions and widths of strong
emission lines.  This last point means that bin boundaries should not
intersect strong emission lines though the bin itself may and should
contain strong lines.  Another way to put this criterion is that
changes in the data in the wavelength bins should be small when the
bin boundaries are changed slightly.
.PP
The bins may be set interactively using a graph of the average
spectrum or automatically by dividing the dispersion axis into a
specified number of equal width bins.  When the number of bins is small
(and the number of wavelength points in each bin is large) bin
boundary effects are likely to be insignificant.
A single bin consisting of all wavelengths, i.e. the sum of all the image
lines, may be used if no wavelength dependence is expected in the
response.  Illumination effects introduced with \fBresponse\fR,
however, appear as wavelength dependent variations in the slit
profile.
.PP
Smoothing of each bin along the slit dimension is done with the
interactive curve fitting routine.  The curve fitting may be done
graphically and interactively on any set of bins or automatically by
specifying the function and order initially.  The fitting should be
done interactively (at least on the first bin) in order to exclude
objects when the sky is not truly blank and contains faint objects or
when object exposures must be used to determine the slit profile.
.PP
As with \fBresponse\fR, several blank sky images may be available
(though this is less often true in practice).  An illumination
correction may be determined for each calibration image or one
illumination correction may be computed from the average of the
calibration images.  Also the illumination response correction may be
determined for only a section of the calibration image so as to be
applicable to multi-slit data.
.PP
Figure 2 shows the fit to one of the wavelength bins; lines 1 to 150 have been
summed and the sum is plotted as a function of slit position (column).
The data is from a response image produced by \fBresponse\fR.  This
figure illustrates a number of things.  \fBIllumination\fR may be run
on a response image to remove the large scale illumination and slit
transmission effects.  This creates a flat field in a manner different than
normal surface fitting.  The figure shows that response effects occur
at all scales (keeping in mind that the pixel-to-pixel response has
been largely averaged out by summing 150 columns).  It also illustrates
how the illumination algorithm works for a typical slit profile.  In
this example about half the large scale variation in the slit profile
is due to illumination effects and half is real slit transmission
variations.  For a blank sky or object image the main differences
would be larger data values (hundreds to thousands) and possibly
objects present in the slit to be excluded from the fit.


.ce
\fB3. Distortion Corrections and Coordinate Transformations\fR
.PP
The removal of geometric distortions and the application of coordinate
transformations are closely related.  Both involve applying a
transformation to the observed image to form the desired final image.
Generally, both steps are combined into a single image transformation
producing distortion corrected images with linear wavelength
coordinates (though the pixel interval may be logarithmic).
This differs from other systems (for example, the Kitt Peak IPPS) which
perform distortion corrections on each axis independently and then
apply a dispersion correction on the distortion corrected image.
While this approach is modular it requires several transformations of
the images and does not couple the distortions in each dimension into
a single two dimensional distortion.
.PP
To transform long slit images requires (1) identifying spectral
features and measuring their positions in arc lamp or sky
exposures at a number of points in the image, (2) determining the
distortions in the slit positions at a number of points along the
dispersion axis using either calibration images taken with special
masks or narrow objects such as stars,
(3) determining a transformation function between the image
coordinates and the user coordinates for the measured wavelength and
slit positions, (4) and interpolating the images to a uniform grid in
the user coordinates according to the transformation function.  The
coordinate feature information and the transformation functions are
stored in a database.  If needed, the database may be examined and
edited.
.PP
An important part of this task is the feature center determination.  This
algorithm is described in a separate section below.
.SH
IDENTIFY - Identify features
.PP
The tasks \fBidentify\fR and \fBreidentify\fR are general tools used
for one dimensional, multi-aperture, multi-slit, echelle, and long slit
spectra.  The tasks are also general in the sense that they are used to
identify features in any one dimensional vector.  For long slit
reductions they are used to identify and trace objects in the slit and
to identify, trace, and determine wavelength solutions for spectral
features from arc calibration images and from sky and object
exposures.
.PP
\fBIdentify\fR is used to identify emission or absorption features in a
one dimensional projection of an image.  This projection consists of an
image line or column or the
average of many lines or columns.  Averaging is used to increase the
signal in weak features and provide better accuracy in determining the
one dimensional positions of the features.  The identified features are
assigned user coordinates.  The user coordinates will ultimately define
the final coordinates of the rectified images.
.PP
For determining the distortions along the slit, the positions of object
profiles or profiles obtained with multi-aperture masks in the slit
are measured at a reference line.  The user coordinates are then taken to be
the positions at this reference line.  The
coordinate rectification will then correct for the distortion to bring the
object positions at the other lines to the same position.
(Note that it is feasible to make an actual coordinate transformation of
the spatial axis to arc seconds or some other units).
.PP
For wavelength features arc calibration images are generally used,
though sky and object exposures can also be used if necessary.  After
marking a number of spectral features and assigning them wavelength
coordinates a \fIdispersion solution\fR can be computed relating the
image coordinate to the wavelength; $lambda~=~f(l)$, where $lambda$ is
wavelength and $l$ is the image line.  The dispersion
solution is determined using the \fBicfit\fR routines described
earlier.  This dispersion solution is used in the long slit package
only as an aid in finding misidentified lines and to automatically add
new features from a wavelength list.  The dispersion solution actually
used in transforming the images is a two dimensional function
determined with the task \fBfitcoords\fR.
.PP
Figure 3 shows a graph from \fBidentify\fR used on a Helium-Neon-Argon
arc calibration image.  Only three lines were identified interactively
and the reminder were added automatically from a standard line list.
Note that the abscissa is in wavelength units and the ordinate is
displayed logarithmically.  The latter again illustrates the flexibility
the user has to modify the graph formats.  Each marked feature is
stored in a database and is automatically reidentified at other columns
in the image with \fBreidentify\fR.
.SH
REIDENTIFY - Reidentify features
.PP
The task \fBreidentify\fR automatically reidentifies the spectral and
object features and measures their positions at a number of other
columns and lines starting from those identified interactively with
\fBidentify\fR.  The algorithms and the feature information produced is
the same as that of \fBidentify\fR including averaging a number of
lines or columns to enhance weak features.  The automatic tracing can
be set to stop or continue when a feature fails to be found in a new
column or line; failure is defined by the position either becoming
indeterminate or shifting by more than a specified amount
(\fIcradius\fR defined in the next section).
.SH
CENTER1D - One dimensional feature centering
.PP
The one dimensional position of a feature is determined by solving the equation

.EQ
define I0 'I sub 0'
define XC 'X sub c'
.EN
.EQ (1)
int ( I - I0 ) f( X - XC ) dX~=~0
.EN

where $I$ is the intensity at position $X$, $I0$ is the continuum
intensity, $X$ is the vector coordinate, and $XC$ is the desired
feature position.  The convolution function $f(X- XC )$ is a
sawtooth as shown in figure 4.  For absorption features the negative of this
function is used.  The figure defines the parameter \fIfwidth\fR which
is set to be approximately the width of the feature.  If it is too
large the centering may be affected by neighboring features and if it
is too small the accuracy is worse.
.PP
For emission features the continuum, $I0$, is assumed to be zero.
For absorption features the continuum
is the maximum value in the region around the initial guess
for $XC$.  The size of the region on each side of the initial guess is
the sum of \fIfwidth\fR/2, to allow for the feature itself, \fIcradius\fR,
to allow for the uncertainty in the feature position, and \fIfwidth\fR, for a
buffer.  Admittedly this is
not the best continuum but it contains the fewest assumptions and is
tolerant of nearby contaminating features.
.PP
Equation (1) is solved iteratively starting with the initial position.
When successive positions agree within 0.1% of a pixel the position is
returned.  If the position wanders further than the user defined
distance \fIcradius\fR from the initial guess or outside of the data
vector then the position is considered to be indefinite.
.SH
FITCOORDS - Fit user coordinates to image coordinates
.PP
Let us denote the image coordinates of a point in the two dimensional
image as $(c,~l)$ where $c$ is the column coordinate
and $l$ is the line coordinate.  Similarly, denote the
long slit coordinates as $(s,~lambda )$ where $s$ is
the slit position and $lambda$ is the wavelength.
The results of \fBidentify\fR and \fBreidentify\fR is a set of points
$(c,~l,~s)$ and $(c,~l,~lambda )$ recorded in the database.
.PP
Two dimensional functions of the image coordinates are fit to the user
coordinates for each set of slit and wavelength features,
$s~=~t sub s (c, l)$ and $lambda~=~t sub lambda (c, l)$, which are
stored in the database.
Note that the second function is a two dimensional dispersion solution.
It is this function which is used to transform the long slit images to
linear wavelength coordinates.  Many images may be used to create a
single transformation or each calibration images may be used separately
to create a set of transformations.
.PP
This task has both an interactive and non-interactive mode.  For the
non-interactive mode the user specifies the transformation function,
either a two dimensional Chebyshev or Legendre polynomial, and separate
orders for the column and line axes.  When run interactively the
user can try different functions and orders, delete bad points, and
examine the data and the transformation in a variety of graphical formats.
The interactive option is quite useful in initially setting the
transformation function parameters and deleting bad points.
The two dimensional function fitting routine is similar in spirit to the
\fBicfit\fR one dimensional function fitting routine.  It is possible
that this routine may find uses in other IRAF tasks.
.PP
Figure 5 shows a graph from \fBfitcoords\fR.  The feature image coordinates
of four objects in the slit (the first of which is very weak)
from \fBidentify\fR and \fBreidentify\fR are plotted.  This information
is used to measure the distortion of the spectrograph in the slit axis.
This example shows particularly gross distortions; often the distortions
would not be visible in such a graph, though expanding it would make
the distortion visible.  The transformation surface fit to this data
removes this distortion almost entirely as seen in the residual plot
of figure 6.  Figure 7 shows the equivalent residual plot for the
wavelength coordinates; a two dimensional dispersion solution.
.SH
TRANSFORM - Transform long slit images to user coordinates
.PP
The coordinate transformations determined with the task \fBfitcoords\fR are
read from the database.  The transformations are evaluated on a grid of
columns and lines, $s sub i~=~t sub s (c sub i , l sub i )$ and
$lambda sub i~=~t sub lambda (c sub i , l sub i )$.
If no transformation is defined for a particular dimension then a unit
transformation is used.  If more than one transformation for a dimension
is given then a set of points is computed for each transformation.
The inverse transformations are obtained by fitting transformation
functions of the same type and orders to the set of slit position and
wavelength points.  Note how this allows combining separate
transformations into one inverse transformation.
.PP
The inverse transformations, $c~=~t sub c (s, lambda )$ and
$l~=~t sub l (s, lambda )$, are used to rectify a set of input images.
The user specifies a linear grid for the transformed images by defining some
subset of the starting and ending coordinates, the pixel interval, and the
number of points.  In addition the pixel interval can be specified to be
logarithmic; used primarily on the wavelength axis for radial
velocity studies.  The inverse transformations define the image column
and line to be interpolated in the input image.  The user has the choice
of several types of image interpolation; bilinear, bicubic, and biquintic
polynomials and bicubic spline.  In addition the interpolation
can be specified to conserve flux by multiplying the interpolated value
by the Jacobian of the transformation.
.PP
The wavelength of the first pixel and the pixel wavelength interval are
recorded in image headers for later use in making plots and in the
\fBonedspec\fR package.  In addition a flag is set in the header indicating
that the image has been dispersion corrected.


.ce
\fB4. Background Subtraction\fR
.SH
BACKGROUND - Fit and subtract a line or column background
.PP
If required, the background sky at each wavelength is subtracted from
the objects using regions of the slit not occupied by the object.
This must be done on coordinate rectified images since the lines or
columns of the image must correspond exactly to the same wavelength.
A set of points along the slit dimension, which are representative of the
background, are chosen interactively.  Generally this will consist of two
strips on either side of the object spectrum.
At each wavelength a low order function is fit to the sky points and then
subtracted from the entire line or column.
.PP
Ideally the response corrections and coordinate rectification will make
the background sky constant at all points on the slit at each
wavelength and the subtracted background is just a constant.  However, if
desired a higher order function may be used to correct for
deficiencies in the data.  A possible problem is focus variations which
cause the width of the sky emission lines to vary along the slit.  One
may partially compensate for the focus variations by using a higher
order background fitting function.
.PP
The background fitting uses the
interactive curve fitting routine \fBicfit\fR described earlier.
Figure 8 shows a graph from \fBbackground\fR illustrating how the user
sets two sample regions defining the sky (indicated a the bottom of
the graph).


.ce
\fB5. Flux Calibration\fR
.SH
EXTINCTION - Apply atmospheric extinction corrections to images
.PP
A set of coordinate rectified images is corrected for atmospheric
extinction with the task \fBextinction\fR.  The extinction correction
is given by the formula

.EQ
    roman {correction~factor}~=~10 sup {0.4~E sub lambda~A}
.EN

where $E sub lambda$ are tabulated extinctions values and $A$ is the air
mass of the observation (determined from information in the image
header).  The tabulated extinctions are interpolated to the wavelength of
each pixel and the correction applied to the input pixel value to form
the output pixel value.  The user may supply the extinction table but
generally a standard extinction table is used.
.PP
The air mass is sought in the image header under the keyword AIRMASS.
If the air mass is not found then it is computed from the zenith
distance, ZD, using the approximation formula from Allen's
"Astrophysical Quantities", 1973, pages 125 and 133

.EQ
	A = ( cos ( roman ZD ) sup 2~+~2 s~+~1) sup half
.EN

where $s$, the atmospheric scale height, is set to be 750.  If the
zenith distance is not found then it must be computed from the
hour angle, the declination, and the observation latitude.   The
hour angle may be computed from the right ascension and the siderial time.
Computed quantities are recorded in the image header.
Flags indicating extinction correction are also set in the image
header.
.SH
FLUXCALIB - Apply flux calibration to images
.PP
The specified images are flux calibrated using a flux calibration file
derived with the \fBonedspec\fR package using standard stars.  The
standard stars are extracted from response corrected, coordinate
rectified, and background subtracted long slit images using the tasks
\fBapdefine\fR and \fBapextract\fR.  The standard stars must not be
extinction corrected because this is done by the \fBonedspec\fR flux
calibration algorithms.  The user may specify flux per unit wavelength,
$roman F sub lambda$, or flux per unit frequency, $roman F sub nu$.
The flux is computed using the exposure time and dispersion from the
image headers and a flux calibration flag is set.


.ce
\fB6. Extraction of One Dimensional Spectra\fR
.PP
The user may wish to extract one dimensional spectra at various points
along the slit.  As mentioned earlier, this is necessary if observations
of standard stars are to be used to calibrate the fluxes.  The flux
calibration values are determined from one dimensional spectra of standard
stars using the \fBonedspec\fR package.  The tools to extract
one dimensional aperture spectra from long slit spectra are \fBapdefine\fR and
\fBapextract\fR.
.SH
APDEFINE - Define apertures for 1D aperture extraction
.PP
Extraction apertures are defined as a list consisting of an
aperture number and lower and upper limits for the aperture.  The aperture
limits are specified as column or line positions which need not be
integers.  The user may create a file containing these
aperture definitions with an editor or use the interactive
graphics task \fBapdefine\fR.
.PP
\fBApdefine\fR graphs the sum of a number of lines or columns (depending
on the dispersion axis) and allows the user to interactively define and
adjust apertures either with the cursor or using explicit commands.
If an aperture definition file exists the apertures are indicated on
the graph initially.  When the user is done a new aperture definition
file is written.
.SH
APEXTRACT - Extract 1D aperture spectra
.PP
One dimensional aperture spectra are extracted from a list of
long slit images using an aperture definition file.  The extraction
consists of the sum of the pixels, including partial pixels, at
each column or line along the dispersion axis between the aperture limits.
.PP
More sophisticated algorithms than simple strip extraction are available
in IRAF and will soon be incorporated in the long slit package.  The
other extraction tasks trace the positions of features, i.e. the aperture
is not fixed at certain columns or lines, and allow weighted extractions
and detecting and removing bad pixels such as cosmic rays.  The
weighted extractions can be chosen to be optimal in a statistical sense.


.ce
\fBConclusion\fR
.PP
The IRAF long slit reduction tasks have been used at NOAO for about six
months and have yielded good results.  The package does not contain specific
analysis tasks.  Some analysis task will be added in time.  The package
is part of the software distributed with release of the IRAF.  The
author of this paper wrote and supports the tasks described here.
Any comments are welcome.
.sp5
.ll 4.2i
.nr LL 4.2i
.LP
\fBCaptions for Figures:\fP
.sp 1
Figure 1.  Ratio of average quartz spectrum to fit of a 20 piece cubic spline
for determination of response correction using \fBresponse\fR.

Figure 2.  Fit of 4 piece cubic spline to the slit profile from the average
of the first 150 lines in a response image using \fBillumination\fR.

Figure 3.  Identification of emission lines from the central column of a
Helium-Neon-Argon spectrum using task \fBidentify\fR.

Figure 4.  Sawtooth convolution function of width \fIfwidth\fR used in the
profile centering algorithm.

Figure 5.  Graph of stellar object positions identified with \fBidentify\fR,
traced with \fBreidentify\fR, and graphed by \fBfitcoords\fR showing the
spectrograph distortions.

Figure 6.  Residuals of the fit of a two dimensional 6th order Chebyshev
polynomial to the data of figure 5 using \fBfitcoords\fR.

Figure 7.  Residuals of the fit of a two dimensional 6th order Chebyshev
polynomial to the image positions of wavelength features using \fBfitcoords\fR.

Figure 8.  Constant background fit to a line of an object spectrum using
\fBbackground\fR.  The marks at the bottom of the graph indicate the
set of points used in the fit.