SMTN-002: Calculating LSST limiting magnitudes and SNR

  • Lynne Jones

Latest Revision: 2016-02-25

Calculating SNR

Calculating either signal to noise ratios for various sources, or 5-sigma point source limiting magnitudes for LSST can be accomplished using standard SNR equations together with available information on the expected LSST camera and telescope components.

The appropriate methodology to calculate SNR values for PSF-optimized photometry is outlined in the LSST Change Controlled Document LSE-40, and partially summarized below. Note that LSE-40 is awaiting updates to match new throughput curves and updated information on how we’re handling the PSF profile which means the actual values calculated in that document are outdated.

The SNR calculation can be summarized as follows:

\[ \begin{align}\begin{aligned}SNR = \frac{C } {\sqrt{C/g + ( B/g + \sigma^2_{instr}) \, n_{eff}}}\\n_{eff} = 2.266 \, (FWHM_{eff} / pixelScale)^2\end{aligned}\end{align} \]

where C = total source counts, B = sky background counts per pixel, \(\sigma_{instr}\) is the instrumental noise per pixel (all in ADU) and g = gain. The LSST expected gain is 2.3 electron/ADU, but for purposes of calculating SNR or m5, it can safely be assumed to be 1, which has the nice property that then all quantities are equivalent in ADU or photo-electrons.

Source Counts

The total counts (in ADU) in the focal plane from any source can be calculated by multiplying the source spectrum, \(F_\nu(\lambda)\) at the top of the atmosphere in Janskys, by the fractional probability of reaching the focal plane and being converted into electrons and integrating over wavelength (\(S(\lambda)\)):

\[C = \frac {expTime \, effArea} {g \, h} \int { F_\nu(\lambda) \, \frac{S(\lambda)}{\lambda} d\lambda }\]

where expTime = exposure time in seconds (typically 30 seconds for LSST), effArea = effective collecting area in cm^2 (effective area-weighted clear aperture diameter for the LSST primary, when occultation from the secondary and tertiary mirrors and vignetting effects are included, is 6.423 m), and h = Planck constant. The fractional throughput curves, \(S(\lambda)\), for each component in the LSST hardware system plus a standard atmosphere can be found in the LSST syseng_throughputs github repository.

Instrumental Zeropoints

We can also use the above formula to calculate the ‘instrumental zeropoint’ in each bandpass, the AB magnitude which would produce one count per second (note this value depends on the gain used; here we use gain=1, so the counts in ADU = counts in photo-electrons).

Filter Instrumental Zeropoint (exptime=1s, gain=1)
u 26.50
g 28.30
r 28.13
i 27.79
z 27.40
y 26.58

Sky Counts

When calculating sky background counts per pixel, instead of using the entire hardware system plus atmosphere, the \(F_\nu(\lambda)\) value for the sky spectrum should be multiplied by only the hardware.[1] The skybrightness in magnitudes per sq arcsecond then is used to calculate counts per sq arcsecond, and converted to counts per pixel using the pixelScale, 0.2”/pixel.

The expected sky brightness at zenith, in dark sky, has been calculated in each LSST bandpass by generating a dark sky spectrum, using data from UVES and Gemini near-IR combined with an ESO sky spectrum, with a slight normalization in the u and y bands to match the median dark sky values reported by SDSS. The resulting zenith, dark sky brightness values are in good agreement with other measurements from CTIO and ESO.

Filter Sky brightness (mag/arcsecond^2)
u 22.95
g 22.24
r 21.20
i 20.47
z 19.60
y 18.63

The instrumental zeropoints above could be used to calculate approximate background sky counts per arcsecond sq or exact values could be calculated using the calibrated spectrum available at darksky.dat.

Instrumental Noise

The instrumental noise per pixel, \(\sigma_{instr}\), can be calculated as

\[\sigma_{instr}^2 = (readNoise^2 + (darkCurrent * expTime)) * n_{exp}\]

where the LSST requirements place upper limits of 0.2 photo-electrons/second/pixel on the dark current and 8.8 photo-electrons/pixel/exposure on the total readnoise from the camera (sensors plus electronics). Tests of vendor prototypes sensors are consistent with these requirements.

The current LSST observing plan is to take back-to-back exposures of the same field, each exposure 15 seconds long, for a total of \(n_{exp}\) =2 exposures per 30 second long “visit”. The total instrumental noise per exposure is 9 photo-electrons. The combined total instrumental noise per visit is then 12.7 photo-electrons.

Source footprint (\(n_{eff}\))

Optimal source count extraction means matching the photometry footprint to the PSF of the source. Raytrace experiments using models of the LSST mirors and focal plane and atmosphere, as well as observations from existing telescopes, indicate that the PSF for point sources should be similar to a von Karman profile. The details of the profile depend independently on the size of the atmospheric IQ and the hardware IQ. The conversion factors will be described in a planned update of LSE-40 and the LSST Overview Paper.

Because the SNR calculation only depends on the number of pixels contained in the footprint on the focal plane (to determine the sky noise and instrumental noise contributions), we calculate \(FWHM_{eff}\): the FWHM of a single gaussian which contains the same number of pixels as the von Karman profile. This must be calculated for the appropriate atmosphere and hardware contributions in a given observation.

\[ \begin{align}\begin{aligned}FWHM_{sys}(X) = X^{0.6} \, \sqrt{telSeeing^2 + opticalDesign^2 + cameraSeeing^2}\\FWHM_{eff}(X) = 1.16 \sqrt{FWHM_{sys}^2 + 1.04 \, FWHM_{atm}^2}\end{aligned}\end{align} \]

where requirements place the system contributions at telSeeing = 0.25”, opticalDesign = 0.08”, and cameraSeeing = 0.30”. We can then just calculate \(n_{eff}\) using a single gaussian profile,

\[n_{eff} = 2.266 \, (FWHM_{eff} / pixelScale)^2.\]

For purposes where the physical size of the PSF is important, such as modeling moving object trailing losses or galaxy shape measurements, we can also calculate \(FWHM_{geom}\),

\[FWHM_{geom} = 0.822\,FWHM_{eff} + 0.052\]

\(FWHM_{geom}\) is typically slightly smaller than \(FWHM_{eff}\).

The expected median \(FWHM_{eff}\) at zenith in the various LSST bandpasses is

Filter \(FWHM_{eff}\)
u 0.92”
g 0.87”
r 0.83”
i 0.80”
z 0.78”
y 0.76”

where this includes the expected (and modeled) telescope contribution as well as the distribution of IQ measurements from an on-site DIMM.

Calculating m5

With all of these values, we can calculate the \(5\sigma\) limiting magnitude for point sources (m5) in each bandpass, in the dark sky, zenith case. The resulting values are

Filter m5
u 23.42
g 24.77
r 24.34
i 23.89
z 23.33
y 22.42

Useful github repositories

The algorithms described in LSE-40 are implemented in the LSST sims_photUtils package, available on github. In particular, the SignalToNoise module calculates signal to noise ratios and limiting magnitudes (m5) values. Here is an ipython notebook example using this code to calculate SNR in a variety of situations.

The throughput curves used for this analysis are based on the throughput components in the syseng_throughputs repository. There is more information on the origin of these throughput curves and other key number data in the section ‘Data Sources’ below.

[1]The atmosphere should not be included in the calculation of the expected counts in the focal plane, as the sky emission comes from various layers in the atmosphere - a completely proper treatment would involve a radiative transfer model that includes emission and absorption over the entire atmosphere. Instead the standard treatment is to generate a sky brightness and sky spectrum that correspond to the skybrightness at the pupil of the telescope, and then just multiply this by \(S_{hardware}(\lambda)\) to generate the focal plane counts

Calculating m5 values in the LSST Operations Simulator

To rapidly calculate the m5 values reported with each visit in the outputs from the Operations Simulator, the SNR formulas above are used to calculate two values, \(C_m\) and \(dC_m^{inf}\). These values can then be used to calculate m5 under a wide range of sky brightness, seeing, airmass, and exposure times.

\[ \begin{align}\begin{aligned}\begin{split}m5 = C_m + dC_m + 0.50\,(m_{sky} - 21.0) + 2.5 log_{10}(0.7 / FWHM_{eff}) \\ + 1.25 log_{10}(expTime / 30.0) - k_{atm}\,(X-1.0)\end{split}\\dC_m = dC_m^{inf} - 1.25 log_{10}(1 + (10^{(0.8\, dC_m^{inf} - 1)}/Tscale)\\Tscale = expTime / 30.0 * 10.0^{-0.4*(m_{sky} - m_{darksky})}\end{aligned}\end{align} \]

The \(dC_m^{inf}\) term accounts for the transition between instrument noise limited observations and sky background limited observations as the exposure time or sky brightness varies. For most LSST bandpasses, we are sky-noise dominated even in 15 second exposures, but in the u band, the sky background is low enough that the exposures become read noise limited. The \(k_{atm}\) term captures the extinction of the atmosphere and how it varies with airmass. It can be calculated as \(k_{atm} = -2.5 log_{10} (T_b / \Sigma_b)\), where \(T_b\) is the sum of the total system throughput in a particular bandpass and \(\Sigma_b\) is the sum of the hardware throughput in a particular bandpass (without the atmosphere).

Filter Cm dCm_inf k_atm
u 22.74 0.75 0.50
g 24.38 0.19 0.21
r 24.43 0.10 0.13
i 24.30 0.07 0.10
z 24.15 0.05 0.07
y 23.70 0.04 0.18

These values are used within OpSim to calculate m5 values for each pointing in the calc_m5 function in gen_output.py within the sims_operations codebase.

The remaining required inputs to calculate m5 in OpSim are the sky brightness and the seeing, as the airmass and exposure time will come from the scheduling data itself.

The sky brightness is currently calculated using a V-band sky brightness model based on Krisciunas & Schafer (1991) (K&S), which is then adjusted to give sky brightness values in various bandpasses using color terms that depend on the phase of the moon. The V-band sky brightness calculations are implemented in the AstronomicalSky.py module of OpSim, and the per-filter adjustments based on lunar phase are done in Filters.py. The current OpSim model simply sets y band skybrightness to 17.3 and implements a step-function for twilight if the altitude of the sun is above -18 degrees, setting the sky brightness to 17.0 in z and y (and the scheduler is then constrained to observed in z and y during this time, currently). In the near future we will be updating the OpSim sky brightness model, to a new sims_skybrightness model that more closely follows the ESO sky calculator along with an empirical model for twilight. The sims_skybrightness model has been validated with nearly a year of on-site all-sky measurements. The current model has various flaws compared to the upcoming new model, but for the most part these flaws result in a brighter sky brightness value being used currently than the more realistic sims_skybrightness model predicts (see comparison).

The input seeing data used in OpSim are the atmosphere-only FWHM at 500 nm at zenith, based on three years of on-site DIMM measurements. The raw atmospheric FWHM values (\(FWHM_{500}\)) are adjusted to the image quality delivered by the entire system by

\[ \begin{align}\begin{aligned}FWHM_{sys}(X) = \sqrt{(telSeeing \, X^{0.6})^2 + opticalDesign^2 + cameraSeeing^2}\\FWHM_{atm}(X) = FWHM_{500} \, (\frac{500nm}{\lambda_{eff}})^{0.3} \, (X)^{0.6}\\FWHM_{eff}(X) = 1.16 \sqrt{FWHM_{sys}^2 + 1.04 \, FWHM_{atm}^2}\end{aligned}\end{align} \]

where the system contributions are telSeeing = 0.25”, opticalDesign = 0.08”, and cameraSeeing = 0.30”. \(\lambda_{eff}\) is the effective wavelength for each filter: 366, 482, 622, 754, 869 and 971 nm respectively for u, g, r, i, z, y.

Calculating C_m values

The values for \(C_m\) and \(dC_m^{inf}\) can be calculated using the m5 value of a dark sky, zenith visit.

\[C_m = m5 - 0.5\,(m_{darksky} - 21.0) + 2.5 log_{10}(0.7 / FWHM_{eff}) + 1.25 log_{10}(expTime / 30.0)\]

where \(m_{darksky}\) is the dark sky background value in the bandpass, as described in the table above. A related \(C_m^{inf}\) can be calculated using an m5 value generated by assuming that the instrument noise per exposure is 0. The difference between \(C_m^{inf}\) and \(C_m\) is \(dC_m^{inf}\).

Data Sources and References

Change controlled documents:
Official project documents not under change control -
Primary mirror clear aperture [2] 6.423 m LSE-29, LSR-REQ-0003, LSST Key Numbers
Median delivered Image Quality 0.65” Overview Paper, fig. 1 (Site DIMM + telescope model)
Total instrumental noise per exposure 9 e- LSE-59, CAM-REQ-0020 (readnoise and dark current)
Diameter of field of view 3.5 deg LSE-29, LSR-REQ-0004
Focal plane coverage (fill factor in active area of FOV) >90% LSE-30, OSS-REQ-0259
Focal plane coverage (fill factor in active area of FOV) 91% Calculated from focal plane models
[2]The area-weighted clear aperture is 6.423 m across the entire field of view, although this varies with location. Near the center, the clear aperture is 6.7 m, while near the edge of the field of view it rolls off by about 10%. 6.423 m is the area-weighted average across the full field of view.

Throughput curves: syseng_throughputs github repo:

The QE curve for the CCD is measured from prototype devices delivered by the two vendors under consideration. The filter transmission curves match those provided as specifications to vendors, and are derived from LSE-30, OSS-REQ-0240. Mirror reflectivities are based on lab measurements of pristine witness samples; the losses and lens transmission curves are based on expected performance curves. The atmospheric transmission is based on MODTRAN models of the atmosphere at Cerro Pachon, with the addition of a conservative amount of aerosols. The throughput curves are consistent with the relevant requirements documents, LSE-29 and LSE-30. More information on the throughput curves for each component, along with the time-averaged losses applied to each component due to surface contamination and condensation, is available in the README.

The throughput curves in the syseng_throughputs repository track the expected performance of the components of the LSST systems. There are versions of these throughput curves packaged for distribution in the throughputs github repository, along with jupyter notebook examples of calculating SNR using these curves and the sims_photUtils package, such as this notebook.

The dark sky sky brightness values come from a dark sky, zenith spectrum which produces broadband dark sky background measurements consistent with observed values at SDSS and other sites. We have a new skybrightness package in development which is also in general agreement with these dark sky values. The new sky brightness simulator includes twilight sky brightness, as well as explicit components contributed by the moon, zodiacal light, airglow and sky emission lines - it is based on the ESO sky calculator with the addition of a twilight sky model based on observational data from the LSST site.

The conversion from atmospheric FWHM to delivered image quality is based on ray-trace simulations by Bo Xin (LSST Systems Engineering). The atmospheric FWHM measurements come from an on-site DIMM, described in more depth in the Site Selection documents. The DIMM measurements were cross-checked with measurements coming from nearby atmospheric monitoring systems from other observatories.