Guide to Interpreting the Extracted Spectra

• 1.0 Introduction
• 2.0 Good, but sometimes peculiar, spectra
  • 2.1 K star
  • 2.2 A star
  • 2.3 Iodine cell
  • 2.4 Atmospheric absorption bands
  • 2.5 Deep absorption lines or Lyman limit
  • 2.6 Ends of orders
  • 2.7 Waves
• 3.0 Reduction artifacts
  • 3.1 Individual problems
• 4.0 Diagnostics
  • 4.1 Spatial profile
  • 4.2 Trace

1.0 Introduction

The extracted spectra provided as a browse product represent the result of automatic processing through one particular data reduction tool, MAKEE. In general, human operators of such reduction tools can do a better job of spectral extraction than an automated pipeline, and hence, we strongly recommend that archive users who wish to make serious scientific use of the data should download the raw images and reduce the data themselves, taking particular care that the reduction parameters do not compromise the science that they hope to glean from the data.

Nevertheless, extractions by the pipeline can be quite good. In order to interpret the quality of the extractions, KOA also has an automated "quality assessment" (QA) grade. Again, the automated nature of this grading process is imperfect, so the provided grade is not correct 100% of the time. However, as described below, it was shown to be reliable at greater than 95% confidence level during our extensive testing. Grading is done order-by-order, and the large number of orders in the archive discourages manual grading of each order.

The automatic QA grade is either "Pass," "Fail," or "Unknown." A "Pass" means that reduction artifacts, while they may be present, do not obscure the underlying astronomical data. A "Fail" means that no sign of the underlying data is visible in the extraction; it is dominated by reduction artifacts. When the automatic QA fails for any reason, a grade of "Unknown" is assigned. The QA grade is provided in the EXQUAL (extraction quality) header keyword in the binary FITS files of the extracted spectra. Note that these grades represent the quality of the spectral extraction only, not necessarily the quality of the original data themselves.

In testing, nearly 3000 orders were chosen, representing a wide range of HIRES observing programs. These were visually inspected, and given one of three grades: 1, 3, or 5. Grade 1 implies that no sign of reduction artifacts is visible; grade 3s show real data from the science target, but also show some reduction artifacts; grade 5s show no sign of the underlying data. Grades 1 and 3 are the “pass” grades, and grade 5 is the “fail” grade. Of the orders visually inspected, 88% were grade 1, 7% grade 3, and only 5% grade 5.

From this visually graded sample, an algorithm was developed to use in the automated grading system. The algorithm was designed to differentiate the grade 5s from the grades 1 and 3; it was not designed to differentiate between grades 1 and 3. The algorithm design was then tested on a random sample of orders in the archive. The random sample was expected to differ from the original sample to include more data from observing programs that have more extensive telescope time.

Visual inspection of the random sample, and comparison of the visual grades with the automated grades, indicate that the automated grading system does a good job on the "Pass" grades; only 1.7% of the visually graded 1s and 3s were erroneously graded "Fail." Of the visually graded 5s, 64.3% were erroneously graded "Pass," indicating that the automated system does not provide an accurate indication of poor reduction. This emphasizes the need to inspect the browse products, and perhaps the raw images themselves, to determine the quality of the browse product in KOA.

The automated grades for each order are shown in a table on the "Extracted Spectra" web page, accessible by clicking on the Extracted link under the Quicklook Previews column on the "Search results" page. Each order is shown, together with the automated QA grade.

To aid interpretation of the extracted spectra, the following sections present examples of both good and poor extractions.

2.0 Good, but sometimes peculiar, spectra

In this section we present spectra of different types, including parts of the spectra that may at first seem to show reduction artifacts. In Section 3.0 we show reduction artifacts.

2.1 K star

Cool stars, such as the K giant shown above in the blue, show many absorption features. While to an untrained eye the reduction above may look very noisy, those dips in the spectrum are actually real absorption features from various ions. Such a plethora of absorption features is known as “line-blanketing,” and tends to be more prominent at bluer wavelengths.

2.2 A star spectrum (Stark broadening)

A stars often show very broad absorption hydrogen (Balmer) absorption lines. The H-alpha line from HI.20041005.29813_Flux_1 is shown above. These broad lines might be mistaken for a data reduction artifact, but in fact they are due to Stark broadening of the lines within the star’s atmosphere. Note that they will generally occur at or near the rest wavelengths of the Balmer lines. The spectrum above shows H-alpha near 6563 Å.

Otherwise the continua of A stars (along with O and B stars) are generally very smooth, as shown above.

2.3 Iodine cell

The iodine cell superposes its own set of absorption features. Since most science with the iodine cell involve stars that already have many absorption features intrinsic to their spectra, an example of one with the extra iodine cells is not shown here.

2.4 Atmospheric absorption bands

There are two very prominent, but somewhat peculiar-looking, atmospheric absorption features. The A band, shown above, is located around 7600 Å. The B band, shown below, is located near 7000 Å. These two order, taken from HI.20041005.29813_Flux_2, shows the bands superposed on a high signal-to-noise spectrum of a relatively featureless hot white dwarf.

The shape and wavelength of the A and B bands are distinctive. The depths of the bands varies with the path length through the atmosphere; the A band is always deeper than the B band. Correction for this “telluric” absorption feature can be challenging.

Near 6280 Angstroms a weaker water absorption feature sometimes appears, as shown above.

2.5 Deep absorption lines or Lyman limit

Some science targets display deep absorption features that might seem at first glance unusual. Above is an example from HI.20041005.19901_Flux_1 which shows saturated absorption due to Lyman lines of hydrogen. Note that these absorption lines are normally found at ultraviolet wavelengths that do not penetrate the Earth’s atmosphere. When they are seen in HIRES spectra at longer wavelengths it is because they are highly redshifted; usually the continuum comes from a background QSO (quasistellar object).

A diagnostic of such deep absorption features is that they should not extend below zero flux. As you can see in the leftmost line above, however, imperfect data reduction can sometimes cause the extracted spectrum to be below zero. The target names for these spectra are generally associated with QSOs or other distant objects.

The more complete spectrum below shows the number of absorption features in the spectrum that can occur in the so-called Lyman forest.

2.6 Ends of orders

Often the first and last orders of an image have a range which shows constant flux. This is because part of the curved order falls outside the CCD, and was not recorded. MAKEE replaces the missing part of the data with a constant. Shown is the first order of HI.20041005.48600_Flux_3, showing missing data at the short wavelength end.

2.7 Waves

In some stars, such as HR 6534 (??) shown above, there are real low frequency ripples. Note that the wavelength range of this CCD was very blue.

3.0 Reduction artifacts

3.1 Individual problems

Individual problems are described below. We tell you what to look for to identify the problem, the range of characteristics (strength of the effect, frequency of ripples, range of orders or wavelengths at which it is most common, etc.) We describe (if known) the cause, and suggest ways of mitigating the problem.

3.1.1 No wavelength scale

This is quite easily diagnosed, as the x-axis is labeled from 0 to just over 4000, on all orders. (Four orders from HI.20040819.26675_Flux_2 are shown.)

The cause of this problem is likely that proper wavelength calibration images are not available for this setting of the instrument. It is not uncommon for engineering data. Engineering data often have different goals than science data, and wavelength calibration of the engineering data might not be necessary. There is also a chance that the instrument setting was outside the range that has been programmed into MAKEE. In this case, appropriate wavelength calibration files may exist, but MAKEE does not recognize the spectra.

You could search the archive for wavelength calibration spectra taken at similar settings. Make sure that any wavelength calibration images have enough signal-to-noise and enough emission lines for a reasonable wavelength fit. If you do have appropriate wavelength calibration files, you might need to do the wavelength calibration manually.

The grade of the order does not depend on whether a wavelength scale was successfully calculated or not.

3.1.2 Low-level ripples

The figure above, used to demonstrate the lack of a wavelength scale, also shows another problem with the reduction; low-level ripples in the flux. The general shape of the spectrum, including strong, broad absorption features, is still easily visible, but very weak features may be obscured or hard to measure.

3.1.3 Spikes

Some reduced spectra show a series of spikes—sometimes extending mostly upward from the true spectrum level, sometimes downward, and sometimes in both directions, as the example shown above (from HI.20041001.17455_Flux_1). Often the spikes show different characters in different orders, or even different parts of the same order. This reduction problem tends to recur throughout a given dataset.

The visual grade of the spectrum above is “F”.

3.1.4 “Steps”

Some orders may show “steps,” as the image above (from HI.20041001.18712_Flux_1) demonstrates. There are large flat areas of the extracted spectrum that are often separated by spikes. Sometimes there are more steps than the order above shows, and sometimes there are fewer. The lack of significant noise in the flat areas is another diagnostic.

The grade of the spectrum above is "F".

3.1.5 Others

The full set of orders from HI.20041001.17455_Flux_2 show a range of problems, including spikes and (in order 3, third from the left in the top row) a large gradient in the extracted flux. Order 9 (leftmost of the third row) also shows a rapid rise in flux from a data reduction anomaly. In this order it is combined with “steps.” Note, too, that some orders (5, 6, and 18 above) sometimes show no data at all. The diagnostic here is that the y-axis scale ranges from zero to 1.

All of the orders shown above would be visually graded “F”.

4.0 Diagnostics

4.1 Spatial profile

Poor extractions can arise from a number of causes. Refer to the documentation for the spectral extraction tool you use in order to acquaint yourself with known reduction problems, their diagnosis and repair. What we have seen in the automated KOA pipeline has often shown up as peculiarities in the spatial profile for each order. These profiles are found in the extracted/binaryfits/ccd[1-3]/profile/ subdiredctory in downloaded datasets. The “[1-3]” in the directory name indicates that you should use the number corresponding to the CCD number of interest to you.

One of the most common problems is seen as a truncated or wide profile, as shown below. The first two panels in the figure show the object half-widths and centroids. Starting with the next panel, the spatial profile of the star is shown for each of the orders on the CCD.

You can see that in this reduction the low-numbered orders show only part of a normal PSF (point-spread function). (The higher orders at the bottom of the plot show a more normal profile.) The extractions for these low orders are poor; they were visually ranked as a “5” or “fail.” The automated QA algorithm has difficulty properly grading these types of reduction problems.

4.2 Trace

The trace spectra from the reduction is another diagnostic. The trace below, from HI.20061213.46327.fits, shows a trace that seems to jump (probably from one order to another), and has a high fit RMS (6.071 pixels). The dots are the measured centroids on this plot, and the solid line is a low order polynomial fit to the data.

The trace of order 2 of HI.20041001.17455_Flux_2 (above) shows a centroid of the spectrum with high frequency ripples. Normally such a high signal-to-noise spectrum would show a much smaller dispersion around the fit, and would not show such high frequency changes.However, ripples in the trace do not necessarily imply that the reduction is poor. Note that while the ripples are quite clear in the plot above, the RMS is still only 1/50 of a pixel.