IFU: Sensitivity

IFU: Sensitivity

IFU versus long-slit – should I, shouldn’t I…?

The IFU is optimised for use in the H and K bands, though it may also be used at longer and shorter wavelengths. To characterise its performance, during commissioning, observations of the same bright standard star were obtained with the 2-pixel-wide long slit and then with the IFU. The same exposure times were used with each set-up; all grisms were checked. The stellar continuum was (optimally) extracted from a differenced pair of raw, long-slit data. A spectrum from a differenced pair of raw, IFU spectral images was extracted in the same way; in this case the strongest continuum spectrum seen on the array was used. The single IFU spectrum should therefore be comparable to the “peaked-up” 2-pixel long-slit spectrum (provided the source is centred on one IFU mirror slice) and the relative signal strengths should be a measure of the losses associated with the IFU optics. These losses are quantified in the table below.

IFU/Long-Slit Throughput Comparisons

GrismThroughput relative to Long-Slit mode
short J53%
long J59%
short H62%
long H63%
HK63% (H)
61% (K)
short K~60%
long K~60%
short L~70%
long L65%
Click on the Grism name for a comparison of the IFU and long-slit spectra. IFU spectra in green; long-slit spectra in red.

The increase in relative transmission at longer wavelengths is due to the increase in reflectivity of the aluminium IFU mirror segments (with a smaller contribution from the lessening effect of scattering). Diffraction will reduce the overall throughput at longer wavelengths (although the relative throughput shouldn’t decrease since this should be the same for the long-slit and the IFU).

Note also that, although the signal through the IFU is attentuated, so is the source of the noise: the IFU spectra in the above table are noticably less noisey than the long-slit data. So even for 50% light loss, the S/N on a given spectrum will only decrease by 0.70.

The IFU and Point Sources

Although the additional optics (and mask) used with the IFU do clearly result in some light loss, these losses may in some cases be regained because the IFU is a 2-dimentional spectrograph. The IFU may even be the best option for point sources if the seeing is bad and moderate spectral resolution is required (the alternative – use of a wider long-slit – would of course reduce the spectral resolution). Averaging adjacent rows and/or spectra from adjacent slitlets will in many cases (poor seeing and/or extended sources) improve S/N. The pixel scale along each 6″-long slitlet is 0.12″; the width of each slitlet is 2 pixels, or 0.24″ (14 slitlets, or slices, gives the 3.3″ width of an IFU image), so averaging adjacent rows in the scrunched spectral image would give square 0.24″ x 0.24″ pixels. Note, however, that averaging rows also introduces more read-noise, so please be wary of this with faint point/compact sources at shorter wavelengths.

IFU Sensitivities

For IFU sensitivity estimates for telescope proposals please use the values listed below:

  • For Extended, continuum sources – Use column 3
  • For Extended, line-emission sources; line spectrally RESOLVED – Use column 4
  • For Extended, line-emission sources; line spectrally UNRESOLVED – Use column 5


Grism*WavelengthExtended Source
Extended Source
Extended Source
short J1.08um15.23e(-15)2e(-18)
long J1.23um15.42e(-15)2e(-18)
short H1.52um15.51e(-15)6e(-19)
long H1.70um15.41e(-15)6e(-19)
short K2.13um14.96e(-16)6e(-19)
long K2.30um13.91e(-15)6e(-19)
short L3.27um10.63e(-15)2e(-17)
long L3.84um10.73e(-15)1e(-17)
*Click on the grism name to get a rough idea of how the Signal-to-Noise ratio is likely to vary across the wavelength range covered by the grism.


  • The surface brightness sensitivities are per resolution element (as opposed to per spectral pixel, used for point/continuum sources on the long-slit pages: note that per resln. element is more appropriate for line emission sources [line in a resln element], while per pixel is more appropriate for continuum flux spread across the whole wavelength coverage).
  • The sensitivities assume nodding to blank sky, i.e. 30 minutes represents 15 minutes on source and 15 on sky.
  • Values were therefore derived from the 4-pixel long-slit spectroscopy sensitivities, with the following assumptions:
    • 4pix to 2pix (x sqrt-2): The long-slit sensitivities were derived from observations obtained with a 4-pixel slit, whereas an individual spectrum on an IFU image is through a 2-pixel slit (with double the spectral resolution). For the sensitivity on an individual IFU spectrum, the figures from the long-slit spectroscopy page have therefore been reduced by a factor of sqrt-2, or by 0.4 mag. However, the quoted 4-pixel long-slit performance could probably be regained by averaging adjacent slices from the IFU spectral image and/or by binning over 2 pixels in the dispersion direction.
    • Nod to sky (x sqrt-2): The long-slit sensitivites assume nodding up-and-down the slit. This will not be possible with the IFU in most cases, so sensitivities were reduced by a further sqrt-2 (or 0.4 mags).
    • Transmission (x 1/0.8): To account for the transmission losses associated with the additional IFU optics (described earlier), we have also reduced sensitivities by ~1/0.8 (or 0.25 mag) E.G. if the IFU throughput in the K-band is 60% of the throughput for ordinary long-slit spectroscopy, then the S/N in an IFU spectrum will be the square root of 0.6 times the S/N expected with long-slit spectroscopy (assuming Poisson noise from the sky and source itself dominates the noise).

IFU “white-light” images

The white-light image produced by the DR essentially represents the collapse (along the dispersion axis) of the individual slices in the scrunched spectral image. These 1-D image strips are then displayed side-by-side to give a 3″x6″ image. Note, however, that this image will not be as deep as a “normal” image of the source, because of the additional read-noise added into the data from the 1024pixels in the dispersion direction. The best spectrum in the scrunched image should, however, be “comparable” to a spectrum from the 2-pixel slit.

Units for PATT proposals…

When estimating exposure times for a patt proposal, please give surface brightnesses in W/m2/arcsec2 (or mags/arcsec2); a flux of, say, 10{-18} W/m2 will only be detectable if it is confined to a reasonably small area. A bright object that is extended over many arcseconds may not be a good target for the IFU!

Previous: IFU grisms | Up: UIST IFU | Next: IFU Readmodes