DR5 2008-2018 Coadd Maps

Product Name
ACT DR5 Explanatory Supplement


Sky maps

The main data products in this data release are the ACT+Planck and ACT-only sky maps:


These are 32-bit float FITS images with shape 43200,10320,3. The first two axes are RA and dec in the Plate Carreé projection, covering the area 180° > RA > -180° and -63° < dec < 23° at 0.5 arcmin resolution. The last axis represents the three Stokes parameters I, Q and U in the IAU polarization convention. The maps are in units of μK CMB temperature increment. Note that the axes appear in the opposite order when loaded as a pixell enmap, since enmap (like numpy) uses row-major ordering instead of column-major ordering like FITS does.

See below under Known Issues for recommended use of these maps.

Inverse variance maps

Associated with each of these maps is a noise floor inverse variance map, which has the same shape and contains an estimate of the non-atmospheric inverse variance in 1/μK2 per pixel. This does not include the contribution from Planck due to Planck's limited multipole range. These files are labeled ivar, e.g. act_dr5.01_s08s18_AA_f090_night_ivar.fits.

Detailed noise model

A more detailed noise model is provided in the files


These are 32-bit float FITS images with shape 720,172,50,15,3, and provides the noise inverse variance in units of 1/μK2 per square arcmin as a function of position, ℓ and detector array, albeit at reduced resolution to make the file size managable.

The first two axes are RA, dec in the same projection as the main maps, but at only 0.5° resolution. The third axis represents 50 exponentially spaced multipoles from 100 to 20,000: ℓi = 100 ⋅ 200i/50. The noise power is a smooth function of ℓ, so these 50 sample points should suffice for most purposes.

The fourth axis represents the 15 different detector arrays that contribute to these coadds: pa1_f150, pa2_f150, pa3_f090, pa3_f150, pa4_f150, pa4_f220, pa5_f090, pa5_f150, pa6_f090, pa6_f150, ar1_f150, ar2_f220, planck_f090, planck_f150 and planck_f220. This axis can be summed over if one is not interested in how much each of these contributes to the total inverse variance. Because the sky maps are inverse variance weighted combinations of the individual input maps, one can recover the relative weight of each array in the combination by its inverse variance by the total, resulting in a set of per-array weights.

As with the sky maps, the axes in these files appear in the opposite order when loaded as an enmap, i.e. 3,15,50,172,720. The list of multipoles and arrays is also included in the file act_planck_dr5.01_s08s18_fullivar_info.txt.


act_planck_dr5.01_s08s18_bandpasses.txt provides the bandpasses of each of the 15 detector arrays in units of μK/(MJy/sr)/GHz. The first column is the frequency in GHz, followed by a column for each array, in the same order as for the detailed noise model. This can be combined with the per-array weights to estimate the effective bandpass at any point and multipole in the map. However, this would only be approximate because this file does not capture the scale-dependence of the individual detector array bandpasses.

The file act_planck_dr5.01_s08s18_bandpasses_scaledep.hdf provides the full scale-dependence of the array bandpasses. It is a HDF file containing 4 data sets: arrays, ls, freqs and bandpass. arrays and ls just list the array names and multipoles, which are the sames as those given above, while freqs is the list of the 434 frequencies the bandpasses are resampled to (66 GHz to 283 GHz with 0.5 GHz steps). bandpasses is a data set with shape 2,15,50,434. The first axis corresponds to the bandpass value and uncertainty, while the remaining axes are the arrays, multipoles and frequencies respectively.

The code fragment below illustrates how to compute the effective ℓ-dependent bandpass for total intensity component of the ACT+Planck f150 day+night map at RA = 0°, dec = 0°, which corresponds to the noisebox pixel [126,359]:

noisebox    = enmap.read_map("act_planck_dr5.01_s08s18_AA_f150_night_fullivar.fits")
weights     = noisebox / np.sum(noisebox,1)[:,None]
bands_array = hget("act_planck_dr5.01_s08s18_bandpasses_scaledep.hdf", "bandpass")
band_at_pos = np.sum(weights[0,:,:,None,126,359]*bands_array[0],0)

The result in this case is a 2D array with shape [Nℓ,Nfreq]. One can similarly get the average bandpass over the whole map. To exclude Planck, simply set the Planck entries in noisebox to zero (noisebox[:,-3:]=0) before computing the weights.

Responses and bandcenters

The main use of the bandpasses is to find the map response to individual signal components like the CMB, tSZ, dust and synchrotron. For convenience we provide map-averaged versions of these, labeled response_cmb, response_tsz, response_dust and response_sync. These are text tables with columns ℓ, I, dI, Q, dQ, U, dU, where I, Q, U are the response in each Stokes component (The distinction between Q and U is mostly meaningless here, since they have the same noise properties in the maps, which translates into the same effective bandpasses), and dI, dQ, dU are their spatial standard deviation. The response is in units of μK/μK for the CMB (This is always 1, making the CMB response files redundant), μK/y for tSZ and μK/arbitrary for dust and synchrotron.

We also provide band-centers in the same format, labeled band_center_cmb, band_center_tsz, band_center_dust and band_center_sync, in units of GHz, but for most purposes the response files will be more relevant.


The map beams transfer functions are available in the following files.


These have two columns, ℓ and B(ℓ), which applies to all Stokes parameters, and to both the ACT+Planck and ACT-only maps.

Point source subtracted maps

There is also a source-subtracted version of each map, labeled map_srcfree instead of map, e.g. act_planck_dr5.01_s08s18_AA_f090_night_map_srcfree.fits. These have the same format and noise properties as the normal maps.

Known issues

Due to the preliminary nature of the data going into these maps, they should not be used for precision cosmology such as CMB power spectrum analysis and cosmological parameter recovery (further examples below). The ACT-only maps lack power on large scales due to a combination of difficulties recovering these scales from the ground and the ground pickup filter we applied. This filter is much smaller in the ACT+Planck maps, <1% in TT but but reaching up to 10% in EE for ℓ < 350 due to Planck's low statistical weight there. See Appendix E in the paper.

Very bright regions in the galaxy were mistakenly identified as glitches in our data cuts, resulting in holes in the map surrounded by artifacts at low galactic latitudes. This will be fixed in a future data release.

The noise model is built on the assumption that difference maps only contain noise. This breaks down for the Planck maps in regions with strong, variable point sources. This manifests as tiles of too low inverse variance in the Planck parts of the detailed noise model. This has minimal impact on the actual act+planck maps because the steepness of the ACT noise curve means that even a large change in the Planck noise model only slightly changes the multipole where the map transitions from Planck-dominated to ACT-dominated. But it means that care should be taken when using this as a noise model for analysing the Planck maps in their own right.

Examples of applications these maps are useful for include point source studies, Galactic science, per-frequency cluster studies, and kSZ cross-correlations. These analyses are appropriate where amplitude calibration of the maps at a level better than a few percent is not required. We do not recommend their use for precision cosmology, where this includes TT/TE/EE power spectrum estimates, the reconstructed lensing power spectrum, birefringence, lensing cross-correlations, primordial bispectra, and component separation of these maps for CMB and Compton y-maps. Null tests have either not been run or not passed for these statistics, and simulations of these coadded DR5 maps are not available. The beam uncertainty and leakage has not been well characterized for these maps, and bandpass-beam coupling effects have not been accounted for. For precision cosmology analyses using these statistics, we recommend that the community use the public DR4 maps that include data through 2016, and then update to the next DR6 map release currrently in preparation with data through 2019.

Back to Product Page

A service of the HEASARC and of the Astrophysics Science Division at NASA/GSFC
Goddard Space Flight Center, National Aeronautics and Space Administration
HEASARC Director: Dr. Alan P. Smale
LAMBDA Director: Dr. Eric R. Switzer
NASA Official: Dr. Eric R. Switzer
Web Curator: Mr. Michael R. Greason