Change page style: 

GNIRS Observing Strategies and Guidelines

This page brings together information that might affect decisions on observing strategy in various GNIRS observing modes, and provide guidelines/tips to maximize observing efficiency and avoid some common errors. The other GNIRS web pages (and especially the Known Issues page) also give information about what to expect from the instrument. For information about actually setting up observations in the Observing Tool, please see the Observation Preparation section of these pages. If you discover errors or inconsistencies, please let us know!

Factors the PI may wish to take into account when designing observing sequences:




Read Modes

In the OT the read mode is automatically set based on the length of the individual exposures. However, its choice is not always the best one, because the OT does not take into account the background in the chosen wavelength range, or the brightness of the object to be observed. The choice of read mode can be edited to suit the user's needs. The following example illustrates the issues:

  • The user wishes to observe a L=12.0 mag star from 3.35µm to 3.65µm at R~6000. She selects the 111 l/mm grating and the short camera, the deep well mode (180,000e) and an exposure time of 30 seconds. As soon as she enters 30 seconds the OT selects the faint object mode, which has a readout overhead of 9.0 seconds and a read noise of 12e according to the Detector properties webpage. A check of the GNIRS ITC shows that the signal from the star in the peak row is ~200e. The shot noise associated with that is 14e, so if that were the only criterion, the choice of read noise would be correct. However, the ITC plot also shows that the square root of the background is 100e or higher in this wavelength range. The background fluctuations thus dominate the noise and would do so even if the bright readmode (readnoise = 38e) were chosen. Since the bright read mode has an overhead of only 0.6 seconds, the user should select it in the OT.

  • Maximum Exposure Times

    The maximum exposure time you should use is set by several constraints:

    (1) Saturation on the sky
    (2) Saturation on the object
    (3) Sky variations during a dither sequence

    The GNIRS exposure times page gives safe limits for currently available configurations. You can check background in the ITC for any configuration by squaring the background value shown in the result, and then dividing by the number of pixels in the software aperture used. The background per pixel should be kept at ~50,000 electrons maximum for shallow well (K and shorter wavelengths), and ~100,000 for deep well (L, M). We recommend individual exposures generally not exceed 900 seconds.

    Bright objects may saturate the array before the background does. Check the recommended exposure times and scale the values to your particular case. You can check for object saturation with the ITC by selecting the star brightness and the instrument configuration and, in the "Analysis method" section, selecting "Software aperture of diameter" equal to the pixel scale (i.e., 0.15 or 0.05 arcsec); you will then see the spectrum in the peak row of the array only and will be able to judge if it saturates the array. If your program uses cloud or seeing conditions worse than the median, you should still do the saturation check under median conditions (70% IQ and 50% cloud), as your program may be executed under better than requested conditions. In general, check anything brighter than 12th magnitude in K band and below, 7th magnitude in L, and 1st magnitude at M. As long as the integration times are only a few seconds, you might as well do 2-3 co-adds per position as well. This is especially advisable if you have set allowable cloud percentile to 70% or worse.

    If you only need to spend modest amounts of time on your object to reach the desired signal to noise, doing a longer sequence or 2 repetitions is generally better than a couple of offset positions at the maximum allowable time. Use the ITC to see whether you lose predicted S/N by doing more, shorter exposures. If the loss is small (5% or less), you should get better sky removal with the shorter exposure times. This is mainly an issue at the shorter wavelengths, since you are always doing very short exposures at L and M.

    For very faint objects, it can be better to maximize the exposure time at each dither position to minimize readnoise effects (and improve efficiency), even at the expense of slightly better sky subtraction. Often this is an important trade-off; again, use the ITC to experiment with your particular situation.


    Offsetting (Nodding, Dithering)

    Infrared observations generally require background subtraction. This is accomplished by moving the telescope slightly from one integration to the next (a.k.a offsetting, dithering, nodding). The offsets in the GNIRS Library are all of the form ABBA. Ideally, one should take observations at several (not just two) positions on the slit. This reduces the chances that bad pixels will severely affect specific wavelengths, and in ideal circumstances increases the S/N of the combined spectrum. This is not possible, however, when the slit is short or the object is extended, and it may not be advisable if the source is so faint that it might be difficult to identify all of the locations of its spectrum on the array. Here are some guidelines:

    • When nodding along a long slit remember that smaller nods more accurately retain the target in the slit than do large nods. Although the slit is 50 or 99 arcsec long in long-slit mode, depending on the camera/pixel scale, it is not necessary to use all of it on small targets. For a point source, nods of 3-5 arcsec are ample.
    • For objects of moderate extent (<50 arcsecs), large offsets can be used to keep the object in the slit. In this case, guiding should be enabled at all positions. This reduces efficiency somewhat for each nod, but is compensated by collecting source photons at all positions.
    • For larger extended objects, offsets to sky are necessary which nominally do not require guiding - The PWFS should be set to "freeze" at the sky positions for maximum efficiency. It is a good idea to dither the sky positions by a few arc-seconds to facilitate removal of accidental point sources.
    • The cross-dispersed (XD) slit is so short (5 or 7 arcsec, depending on the exact configuration) that ABBA is the only option, with a typical nod of ~3 arcsec. For an object larger than an arcsec or two it is necessary to nod the source out of the slit.
    • The default offsets in the OT library in XD mode are -1",+2" for the science target and +1",-2" for the standard star. This is so that persistence from bright standard stars will not affect spectra of a faint science target. The 2" offset brings the target fairly close to the end of the slit, especially in worse-than-average seeing. Having some pixels at the end of the slit available to use for residual background subtraction when extracting spectra can be useful. If this consideration outweighs that of persistence for your science, then consider adjusting the offsets accordingly.
    • Generally it is recommended to spend the same amount of time on object and on sky for best sky subtraction.


    Slit orientation

    The goal when doing spectroscopy with a narrow slit is to get as much light as possible from the astronomical source into the slit. The orientation of the slit can reduce the amount of light. Sometimes the slit orientation is determined by the science goal (e.g., the slit must be oriented along the long axis of a galaxy). When science is not the driver then the slit orientation should take the following issues into account.

    • There can be flexure between the guider and the instrument. As the tracking of the telescope always has an EW component, and that component tends to be the dominate direction of tracking near the meridian, it is expected and indeed observed that the dominant differential flexure between an external guider (e.g., PWFS2) and the spectrograph is usually mainly in the EW direction. The information we have obtained so far about the GNIRS - PWFS2 combination is consistent with that, although there are anomalies. At present all GNIRS configurations in the GNIRS OT Library have the slit oriented EW (i.e., at 90 degrees), so that differential flexure will tend to cause drift of the spectrum along the slit as opposed to perpendicular to it.
    • However, atmospheric differential refraction is also important to consider, especially when using narrow slits. The atmosphere refracts shorter wavelengths more strongly than longer wavelengths, which means that a white point source is spread out into a little spectrum along the elevation angle. If the slit is not oriented roughly along the elevation angle (the "parallactic angle") and is too narrow, some light will not pass through the slit. The effect is much larger in the optical than in the infrared, and is generally small enough that within a single GNIRS order (i.e., J, H, K, L, M) it is not a problem. However, in the cross-dispersed mode, particularly when airmasses are moderate to high and the long blue camera (0.05 arcsec pixels) and any narrow slit, or the short blue camera (0.15 arcsec pixels) with its narrowest (0.30 arcsec) slit are in use, it is important to consider it. The following table describes the effect.

    • Shift of point source in arcseconds from its location at 1.0 microns on Mauna Kea
      Wavelength Elevation / Airmass
      (microns) 60 deg / 1.15 45 deg / 1.41 30 deg / 2.00
      1.0 0.00 0.00 0.00
      1.5 0.02 0.08 0.15
      2.0 0.03 0.11 0.20
      2.5 0.03 0.12 0.22

      Slit width vs S/N in the JHK bands

      In many cases wider slit width and lower spectral resolution mean both higher throughput and better sensitivity. However, the latter (better sensitivity) is not necessarily the case in the J and H bands, and in the short wavelength half of the K band, where the sky background and the noise are dominated by a host of strong and very narrow OH lines. This is particularly true when using GNIRS in its low resolution mode (e.g., 0.15 arcsec pixels and 32 l/mm grating, or 0.05 arcsec pixels and 10 l/mm grating). Each of these gives R~1,800 when the narrowest slits (0.30 arcsec and 0.10 arcsec, respectively) are used. In the J,H, and K bands the GNIRS Integration Time Calculator shows that the S/N fluctuates by as much as a factor of ten on and off of the unresolved OH lines.

      The effect of a wider slit is to broaden each region of low S/N due to each OH line. The effect is particularly noticeable in the H and K bands, which have the strongest OH lines. In the H band, for example, there are ~60 strong OH lines between 1.5 and 1.8 microns; thus on average they are separated from their nearest neighbors by 0.005 microns and at a resolving power of 330 a spectral resolution element will on average contain one such line. Thus at R=1800 an H-band spectrum will contain mostly data points at the high end of the S/N range. However, at R~540, obtained by using the short blue camera, 32 l/mm grating, and 1.0 arcsec slit, or R=600, obtained with the long camera, 10 l/mm grating, and 0.3 arcsec slit) each sky line will extend over ~0.003 microns in the spectrum and and the sky noise from those lines will be spread over half of the spectrum or more. That would be offset in part by the higher throughput (due to the wider slit) and by binning the data (with the 1" slit and the short blue camera one spectral resolution element corresponds to 6.7 spectral pixels).

      Thus depending on the specific science the PI may wish to use a wide slit with its resultant higher throughput and more pixels per resolution element, or select a narrower slit width in these bands and accept some loss in throughput in order to obtain high spectral resolution and higher S/N in specific clean spectral intervals in the J, H, and K bands.

      Note that currently the GNIRS Integration Time Calculator is not treating this effect properly and that there are instructions at the top of the ITC web page for "correcting" for this bug.