Atmospheric calibration with SPT

From CMB-S4 wiki
Jump to navigationJump to search

April 15, 2020 - Julian Borrill, Tom Crawford, Reijo Keskitalo, Sasha Rahlin and Nathan Whitehorn


In this post we use the South Pole Telescope data to validate the TOAST atmospheric simulation module. The SPT dataset features trichroic bolometers, allowing us to test our assumptions about

  • correlations across frequencies
  • relative weights of detector noise and atmosphere as a function of frequency
  • validity of the atmospheric model at the South Pole

All validation to date has been performed on data from Chilean sites so this is a critical test to ensure the simulation tools are also suitable for Polar experiments.


We use two separate datasets:

  • "79891185" was collected during optimal observing conditions and minimal atmospheric interference
  • "68894255" was collected during suboptimal observing conditions and considerable atmospheric interference

Here is a plot of the timelines in the suboptimal weather case:

Tod 68894255.png

The very long time scale (subharmonic) fluctuations in the data make it difficult to use Fourier methods to study them. We project out a second order polynomial from each of the four subscans. This removes the elevation-driven offset and the associated step caused by changing the observing elevation midway through the data. We also flag and discard the turnarounds:

Tod 68894255.2.png

and here are the timelines during optimal weather, again with the polynomial filtering:

Tod 79891185.2.png


We use TOAST to simulate correlated atmospheric fluctuations and uncorrelated detector noise. The detector noise is particularly apparent in the 90GHz timelines. Here is a plot of the simulated timelines using the "79891185" pointing and randomized weather parameters compatible with the season and time of day of the real data:

Tod 68894255.2.sim.png

The relative levels of atmosphere and noise, both within and across frequency are all compatible with real data but we will use the statistical properties of the data to draw further conclusions.

Cross correlation functions and spectra

We proceed by estimating the correlation function between pairs of detectors. We limit the analysis to a) pairs of detectors that observe at the same elevation (are only offset in azimuth) and b) pairs of detectors that observe at the same azimuth but are offset in elevation

Here are the cross spectra for pairs of detectors at 90GHz. The plot circulates between four panels: 1. bad weather, real data 2. bad weather pointing, simulated data 3. good weather, real data 4. good weather pointing, simulated data.

Cov 090GHz x 090GHz.gif

The top row shows the dimensionless correlation coefficient as a function of lag. The bottom row shows the actual cross spectrum between the detectors.

It should be pointed out that we have not attempted to match the observing conditions between the real data and the simulation, only the pointing is shared. The two simulated cases seem to fall between the real data cases in terms of noise severity.

At 150GHz, the correlations are stronger:

Cov 150GHz x 150GHz.gif

and at 220GHz:

Cov 220GHz x 220GHz.gif

The correlated, low frequency noise is evident across the frequencies. Here we use pairs of detectors where one observes at 150GHz and the other at 220GHz:

Cov 150GHz x 220GHz.gif


We measured the statistical properties of the atmospheric signal in two short segments of SPT data and found low frequency, intra and inter frequency correlations most probably sourced by the atmosphere. We then ran a TOAST TOD simulation comprising instrumental and atmospheric noise. The TOAST correlation properties of the simulated TOD are in good agreement with the real data. The differences between simulated and real data are no larger the differences between the two real datasets, suggesting that the TOAST simulations are able to capture the defining features of atmospheric noise in South Pole data.