Home

PACS Map-making Tools: Analysis and - Herschel

1. amp 0 7 0 500 L 1 1 l L 1 l L 1 L 1 1 L 0 1 L 1 L L 1 1 L 500 L L 1 1 L L 1 1 L L 1 1 L L 1 1 L 0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000 MIPS 70 micron MJy sr MIPS 70 micron MJy sr MIPS 70 micron MJy sr IC348 Tamasis IC348 Unimap 2000 T T T T T T T T T T T T T T T T T T T T T 2000 T T T T T T T T T T T T T T T L PACS 35 6455 1 54040 MIP PACS 111 785 1 47912 MIPS 1500 F A L T 7 m Li si f 5 amp 1000 amp f c E c H o o El L 3 1000 2l E f g 10001 R s500F R g go a a soo E ap A 500 L L L 1 1 L L 1 1 L L L 1 L L 1 1 0 1 L L 1 L L 1 L L 1 L 500 0 500 1000 1500 2000 0 500 1000 1500 2000 MIPS 70 micron MJy sr MIPS 70 micron MJy sr Figure 21 continued Pixel to pixel PACS vs MIPS scatter plots and corresponding linear fit for 10348 at 70 um PACS 70 micron MJy sr Figure 21 continued Pixel to pixel PACS vs MIPS PACS Herschel LDN1780 JScanam Document Date Version LDN1780 Scanamorphos T T T T T T T PACS mapmaking November 1st 2013 1 0 LDN 1780 MADMop TET ED Sia fa n1 PACS 41 5106 A 3 18304 MIPS ga gg O 200 T T T T T T T T T 200 T T T T T T T T 3o00 T L PACS 1 29189 2 36917 MIPS L PACS 30 4894 2 17786 MIPS L L 250 150 150 t 5 or 1 2001 f 3 Ter a EL t 1
2. therefore in this case an 18 pixel radius was used Using these PSFs curves of growth were built normalized to the larger available radius Fig ure 15 illustrates the PSFs obtained from the maps generated with the different map makers at 100 um and 160 pm respectively The PEP HPF PSFs are not shwon as they are affected by HPF flux losses 23 and they need to be cut at a much smaller radius e g 5 6 pixels 5 3 4 Injection of artificial sources Artificial sources were injected on the maps generated by the different map making codes using the IRAF ARTDATA package The observed PSFs previously extracted were used as source profile in this process Ten thousand sources were added to each map according to a flux distribution given by deep FIR number counts 1 2 10 11 Since the area covered by these maps is very small the 10000 sources were split in 500 groups of 20 sources each In other words for each map 500 copies were created and each time only 20 artificial sources were added so that the latter are well isolated and the properties of the maps can be studied without incurring into crowding issues Sources were added down to fluxes 10 20 times deeper than the actual flux limits of the maps In this way the simulation is more realistic than just stopping at or close to the threshold In fact the extraction of sources from these images includes the possibility that faint objects effectively below the flux limit are extrac
3. 4n y O a s a 01 02 03 04 05 06 Truth Bright red Reprojected Jy Figure 12 continued Pixel to pixel scatter plot of S Struc vs Strue for the bright 160 um case No offset correction applied P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Faint red blue reprojected Bright red blue reprojected 0 008 0 008 0 006 0 006 E 5 i Y Y a 0 004 a 0 004 O O gt gt V Q d d gt 0 002 4 0 002 0 000 0 000 MAD Tama Scan Uni JScan MAD Tama Scan Uni JScan Figure 13 Slope corrected values of the standard deviation of S Swuc for the faint red blue case left panel and bright red blue case right panel Faint Bright red blue red blue MADmap 0 58 0 66 0 02 0 15 Tamasis 0 06 0 08 0 07 0 07 Scanamorphos 0 05 0 02 0 06 0 03 JScanam 0 00 0 00 0 00 0 00 Unimap 0 03 0 10 0 01 0 08 Table 3 Summary of the slopes obtained for all the mappers and cases faint bright red blue from the S Strue vs Strue scatter plots No errors are quoted 5 3 1 Bright sources 0 3 50 Jy In this case the test was done using the Rosette field i e an active star forming region known to contain at least 100 sources with fluxes ranging from 0 3 to 50 Jy 8 Aperture photometry at YO and 160 um was performed for each source following these steps 1 doing a 2D gaussian fitting at the reference coordinates of the source provided in 8
4. 1 97 1 04 1 13 2 10 1 07 Unimap 1 58 1 25 1 48 2 11 0 99 1 08 2 08 1 04 P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Field JScanam MADmap SANEPIC Scanamorphos amasis Unimap Antennae 1 11 1 77 1 10 1 10 1 10 Crab 0 98 0 94 1 44 0 92 0 94 0 94 IC 348 0 40 1 06 1 16 0 65 1 07 LDN 1780 0 73 0 61 0 52 0 46 0 50 M31 1 36 1 31 1 33 1 33 1 31 M8 amp 1 1 38 1 35 1 35 1 36 1 38 NGC 6946 1 48 1 39 1 38 1 39 1 42 Rosette 0 36 0 88 0 97 0 99 0 91 Table 9 Gain factors from 70 um comparison with MIPS Quoted values are from the fits obtained from considering only the surface brightness linear regime i e up to 100 MJy sr Values denoted with indicate low quality fits Field JScanam MADmap SANEPIC Scanamorphos Tamasis Unimap Antennae 1 32 3 62 1 41 1 32 1 40 LDN 1780 0 95 0 61 1 02 0 87 0 89 M31 0 89 1 01 0 87 1 08 0 97 0 93 M381 1 00 1 06 1 07 0 98 1 02 NGC 6946 1 22 1 35 1 35 1 26 1 35 Table 10 Gain factors from 160 um comparison with MIPS Field JScanam MADmap SANEPIC Scanamorphos Tamasis Unimap Antennae 0 74 2 01 0 79 0 72 0 81 LDN 1780 0 78 0 07 0 87 0 74 0 76 M31 0 96 0 94 0 81 1 03 0 80 0 88 M381 1 07 1 12 1 13 1 02 1 11 NGC 6946 0 24 0 48 0 63 0 78 0 64 Table 11 Gain factors from 160 um comparison with MIPS Quoted values are from the fits obtained from considering only the surfac
5. Faint blue Reprojected Jy Truth Faint blue Reprojected Jy E P 0 01 Pe a E ga 0 01 GE E 0 02 g Bn 0 03 x S 0 04 0 00 0 01 0 02 0 03 0 04 0 05 0 06 Truth Faint blue Reprojected Jy Figure 12 Pixel to pixel scatter plot of S S44 vs Strue for the faint 70 um case No offset correction applied P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Jy Difference from Truth MADmap Jy Difference from Truth Scanamorphos Jy Difference from Truth Unimap MEN CNN 0 02 po 0 02 0 00 fron anana a anana nannan ononososnonn go pan 0 02 Ea 002r ES A 0 04 dE 0 04 E 0 08 y 0 06 0 08 S 0 08f A J 0105 A Mwa 0 02 0 04 0 06 0 08 0 10 0 12 0 14 0 02 0 04 0 06 0 08 0 10 0 12 0 14 Truth Bright blue Reprojected Jy Truth Bright blue Reprojected Jy gt 0 02 fp A onon p 0 02 1 0 00 b 8 0007 MMMM Meee RN eo hb 0 02 T l ae 0 02 be 0 04 S O 04 0 06 o5 0 08 SN O9 NE nme il 0 08 EM 0 08 T 0 10 0 10 0 02 0 04 0 08 0 08 0 10 0 12 0 14 z 0 02 0 04 0 06 0 08 0 10 0 12 0 14 Truth Bright blue Reprojected Jy Truth Bright blue Reprojected Jy 0 02 0 00 Porro 0 02 0 04 0 06 0 08 0 10 O12 0 14 Truth Bright blue Reprojected Jy Figure 12 continued Pixel to pixel scatter plot of
6. i oe mitan tesco Sees ae alte aria ete 5 A i l im k A a gt E c 26r Ser MEN jpscan madmap sanepic scanamorphos iamasis unimap 0 20 40 60 80 100 Spatial Frequency Cycles in 204 samples 16 D ud MIGNE MG a a jpscan madmap sanepic scanamorphos tamasis E unimap Spectrum dB 0 2 10 15 20 20 Spatial Frequency Cycles in 201 samples Figure 24 continued Top panel 1D spectrum of the 160 um Atlas field processed with the different codes Bottom panel zoom of the 0 to 20 spatial frequency range The DC has been zeroed x indicates the half power frequency P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 presents at various frequencies characteristic peaks which are not found in the 1D spectra for the other mappers In the red band all the spectra peak at low frequencies and display a frequency dependent behaviour with a half power frequency shifted towards lower frequencies JScanam SANEPIC MADmap and Unimap have a similar spectrum albeit at different levels while the spectrum of Scanamorphos and Tamasis drops at high frequencies Figure 25 shows the 2D spectra In the blue band one can see that the map makers are not isotropic meaning that the noise is more pronounced along some directions In particular the scan directions clearly have an impact on the spatia
7. median Diff All the maps are on the same scale P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Scanamor pr JScanam 0 0148 0 0140 0 0128 0 0112 0 0090 0 0064 0 0032 0 0004 0 0045 Jy pixel Figure 11 continued Faint red 160 um simulations Right top corner truth reference map All other panels difference maps The quantity shown is Diff median Diff All the maps are on the same scale P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Scanamorpk Tamasis JScanam 0 0118 0 011 0 00984 0 00816 0 00599 0 00335 0 000243 0 00338 0 00745 Jy pixel Figure 11 continued Bright red 160 um simulations Right top corner truth reference map All other panels difference maps The quantity shown is Diff median Diff All the maps are on the same scale P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Eae EN 2 E 0 01 P gt 0 00 n enenennsessassssesesanseseasasassssnr E n E n E 9 0 01 ge E 0 02 ES y l ye 0 03 G api ERREUR XE RETO m 0 00 0 01 0 02 0 03 0 04 0 05 0 06 0 00 0 01 0 02 0 03 0 04 0 05 0 06 Truth Faint blue Reprojected Jy Truth Faint blue Reprojected Jy 5 E E 23 38 g ES ES ES RE S g 7 Y 95 va e X gu v T E A 0 00 0 01 0 02 0 03 0 04 0 05 0 06 0 00 0 01 0 02 0 03 0 04 0 05 0 06 Truth
8. striping effects For Herschel data processing the original C language version of the algorithm has been translated to Java to comply with HIPE requirements Additional interfaces are in place to allow PACS and SPIRE data to be processed by MADmap This implementation requires that the noise properties of the detectors are determined aprior These are passed to P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 MADmap via PACS calibration files and are referred to as the InvNtt files or noise filters T he InvNtt is short for Inverse Time Time Noise correlation matrix 2 3 SANEPIC Signal And Noise Estimation Procedure Including Correlation SANEPIC is a maximum likeli hood mapper capable of handling correlated noise between receivers It was first developped to handle the BLAST experiment data 16 and then fully rewritten parallelized and generalized to handle any kind of dataset The SANEPIC package now consists of several programs e SANEFRAMEORDER finds the best distribution of the input data files over several computer if SANEPIC is used on a cluster of computers e SANEPRE distributes the data to temporary directories The data are stored in a dirfile format each computer receiving the data segment it will process e SANEPOS computes the map size pixel indices and a naive map One can define a projection center or use a mask as a reference for projection Users can use all p
9. 0 08 0 06 0 04 0 02 0 00 0 02 Median Value 0 04 0 06 0 08 2000 0 2000 4000 6000 8000 10000 12000 14000 16000 18000 Readout Index Figure 5 Correlated signal drift of the PACS bolometer signal The change in signal intensity over time scales of gt 100s of readout is due to simultaneous drift in the raw signal intensity for all pixels of the array Strong astrophysical sources produce changes on much shorter time scales as they move in and out of the field of view of the bolometer array during a scan The spikes seen atop the drift signal are due to sources in the sky 4 2 1 Processing notes The pre processing steps described above were applied to each simulated and real data set In some instances it was necessary to flag and remove additional sections of readouts that showed artifacts resulting from a cosmic ray hitting directly the electronics These manifest themselves as the so called module or pixel drops and are recognized as a sudden and drastic change in the signal level of the module or the pixel which is sometimes sustained over thousands of readouts Once free of artifacts the MADmap algorithm is applied to the data to remove the 1 f noise and produce a final map 4 9 SANEPIC processing All the PACS simulated and real data were processed according to the same procedure 1 e e export the data from HIPE using export SpireToSanepic py or export PacsToSanepic py scripts e define
10. 2013 Herschel Version 1 0 e for blank deep fields compute the noise noise power spectra using sanePS with the raw data or bootstrap previously computed noise noise power spectra the number of common mode components varies 1 e 6 for the PACS green band e inverse the noise noise power spectra with sanePS and run sanePic The last two steps can must be iterated using the previous iteration map of sanePic as an input to be remove from the time stream by sanePS The process converge quickly in 3 to 4 iterations This allow to derive noise noise power spectra in the case where strong or weak emissions are present in the data This also allow to adapt the noise component to each data segment in particular in case of cooler burps In case of strong emission in the data noise noise power spectra from a previous empty fleld can be bootstrapped as the first iteration in the process 4 3 1 Processing notes SANEPIC make several assumptions on the data and nosie model which can leads to known caveats artifacts on the maps No gaps in the time stream processing data in the Fourier domain requires that the timestreams are contiguous 1 e without gaps in order to maintain consistency in the noise frequencies features In particular even if they are not used in the final map turnaround of PACS data must be present in the timestream Signal 1s circulant this is an intrinsic hypothesis when doing Fourier Transforms and im plies th
11. 21 Issue 8 118 Piazzo L Subspace Least Square Approach for Drift Removal with Applications to Herschel Data in preparation 19 PICC ME TN 037 http herschel esac esa int twiki pub Public PacsCalibrationWeb pacs bolo fluxcal report vl pdf 20 PICC ME TN 038 http herschel esac esa int twiki pub Public PacsCalibrationWeb cc report vl pdf 21 PICC NHSC TN 029 https nhscdmz2 ipac caltech edu pacs docs Photometer PICC NHSC TN 029 pdf 22 PICC NHSC TR 034 https nhscsci ipac caltech edu pacs docs Photometer PICC NHSC TR 034 pdf 23 Popesso P Magnelli B Buttiglione S et al astro ph 12114257 24 Roussel H 2012 astro ph 12052576 25 Roussel H 2012 proceedings of the annual meeting of the SF2A ed S Boissier P de Laverny N Nardetto R Samadi D Valls Gabaud and H Wozniak p 559 563 26 Tegmark M 1997 ApJ 480 87 27 UniHIPE Web Page http herschel asdc asi it index php page unimap html 28 Unimap Web Page http w3 uniromal it unimap
12. 5 4 2 and Table 7 The PGLS algorithm has only one parameter i e the half length of the median filter The default value is 20 which is normally adequate However in some cases longer filters are needed to fully remove the GLS distortions especially for red band PACS data The WGLS algorithm has a parameter controlling mask construction In version 5 3 or earlier it was difficult to set a satisfactory default value for this parameter Asa consequence for the maps produced for this report the parameter was set manually The problem was solved in version 5 4 4 6 1 Processing notes The input data for Unimap can be produced using a dedicated script named UniHIPE devel oped by the Science Data Center ASDC of the Italian Space Agency ASI 27 UniHIPE can run on any machine where HIPE is installed and will transform the HIPE Level 1 data into a format suitable for Unimap For PACS the Level 1 HIPE product is the perfect starting point for Unimap 5 Codes benchmarking For testing the performance of the map making packages described in Section 2 the following metrics were applied 1 Power spectrum analysis 2 Difference matrix 3 Point source photometry bright and faint sources 4 Extended emission photometry comparison with ancillary data sets IRAS Spitzer MIPS 5 Noise analysis Table 2 summarizes which map making code was tested with a given metric P ACS Document PACS mapmaking Date November 1st 201
13. 5 the maximum Note that SANEPIC did not take part in one of the metrics e g difference matrix the positive exception of Scanamorphos and Unimap in the red band for which the noise is almost isotropic The difference maps also show that these two mappers e g Scanamorphos and Unimap have the lowest high frequency i e pixel to pixel noise Finally we found that some of the mappers appear to introduce a gain in flux calibration especially MADmap for faint background data both in the blue red bands Table A is an attempt to provide a preliminary assessment of the performance of the map making codes in light of the metrics described above For each metric we evaluated the average performance of a given mapper based on the plots and discussions included in Section 5 of the report and accordingly assigned a grade from 0 to 5 where O corresponds to the minimum grade and 5 to the maximum one We caution the reader against any straightforward interpretation of the grades in the Table Despite an effort of folding in the grades the nuances of the metric results it is understandable that a high level of sub jectiveness is involved in the process Therefore we highly recommend reading Section 5 Codes Benchmarking and Section 6 Conclusions before coming to any strong conclusions about the performance of the various mappers P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 PACS Map making Tools Ana
14. T T mT T T T T T T T T Ta 300 T T T T T T T T T T T T T T T T x a T T T T ES T T LJ PACS 5 14164 1 31806 MIES xy x PACS 16 5902 1 39877 MIRS Xx P x E 4 250 x 250 la L x L 8 2001 h E as i 2001 5 mu E E E E 5 150r 5 150r 9 t 9 F E E E L 8 100 8 10067 S sol 2 sol amp sor 7 amp sor ob ot 50L L L L L J L L L L L L 1 L L 1 L 1 L L L L 1 L L 50 L L L L 1 L L L L L L L L 1 L L L L 1 L L L 0 50 100 150 200 250 300 0 50 100 150 200 250 300 MIPS 160 micron MJy sr MIPS 160 micron MJy sr Figure 21 continued Pixel to pixel PACS vs MIPS scatter plots and corresponding linear fit for the Antennae data set at 160 um P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Crab JScanam Crab Scanamorphos Crab MADMop TTTTTTTTTTTUTTTTTTTT TTTTTTTTTTTTTTTTTTT TTTTTTTYYTTUTTYVTTTIJTTTTTTTTTTT TTTTTTTTTT TTTTTTTTT TTTTTTTTTTT TTTTTTTTTTTTTTTTTTTTTTTTTT UUTTTETTY TTTTTTTTT TTUTTYTTTTT TTTTTTTTTTTETTTTTTTTTTTUTTTTTTTTTVTTTTTY TT PACS 50 6182 1 28339 MIPS 7 PACS 42 6169 1 24189 MIPS 7 FP PACS 62 7080 1 24292 MIPS 600 Ly y 600 600 x E 5 5 5 an xD 400 400 400 c c p 2 e 9 9 8 E E E o o o m m m an an an 200 200 200 0 0 100 200 300 400 500 600 700 0 100 200 300 40
15. TIT ELT ES Tal TOT TOYCT tt ay TOT ae tai Document Date Version PACS 70 micron MJy sr SOQ TTTTTTTTT TT PACS mapmaking November 1st 2013 1 0 PACS 70 micron MJy sr NGC6946 Tamasis VoU Y T T IT 500F TTTTTTTTTTTTTT MIPS Ey 1 x X y 400 3 4 5 p i gt ren 3 300 M j 5 d S J E 200 x he y o a J rm J 8 1 a 100 j E ki RET as f esed oe oed 0 100 200 300 400 500 MIPS 70 micron MJy sr N a a o Q ae o o o PACS 70 micron MJy sr o o 400 Jy sr NGC6946 Unimop 500 a XA TITO L T Y T Jen DA T TT TTT wT E UT T Car tT ect Ge TT J J Li A i tt I LLL LL LL 4 Ati tt ti it tt es ee ee ee 1 Aa ee L1 4 1L 3 0 100 200 300 400 500 200 300 MIPS 70 micron M 200 300 400 500 MIPS 70 micran MJy sr MIPS 70 micran MJy sr Figure 21 continued Pixel to pixel PACS vs MIPS scatter plots and corresponding linear fit for NGC 6946 at 70 um NGC6946 BS JScanam 400 ta ae e N e e 100 PACS 160 micron MJy sr NGC6946 Scanamorphos 54 7 400 C ae o N e e 100 PACS 160 micron MJy sr LL ai NGC6946 BS MADMop 400 N tA Q 3 PACS 160 micron MJy sr a NGC6946 BS Tamosis T y T 400 CA ae e N e e 100 PACS 160 micron MJy sr rr er D PACS 29 7375 1 26229 MIPS
16. The user can select the polynomial order and decides if the drift is to be estimated for every single bolometer or for a whole array subarray e Noise this module estimates the noise spectrum and constructs the corresponding GLS noise filters e GLS this module estimates and removes the noise affecting the timelines by implement ing the GLS map maker It produces two output images in the form of fits files the naive map and the GLS map e PGLS this module estimates the distortion introduced by the GLS map maker It is based on the Post Processing for GLS PGLS algorithm described in 17 The estimated distortion is subtracted from the GLS map to produce a PGLS map which is saved in the form of a fits file e WGLS this module implements the Weigthed PGLS WGLS described in 17 where the distortion estimated by the PGLS is analysed and subtracted from the GLS image only when it is significant In this way the noise increase caused by PGLS is minimized A more detailed description of the map maker can be found in the user s manual which can be downloaded from the Unimap Home Page 28 3 Benchmarking data set To test the performance of the individual map making codes we made us of both hybrid simulated and real PACS observations 3 1 Hybrid simulated data The map makers performance can be accurately tested on hybrid simulated data in which simulated sky signal and pure instrument noise components are co added in the flux cal
17. a blank mask with the requested WCS add some margin pixels to accommodate for flag data on the edge as sanePic need to be able to project all data even flagged data needs to be present in the map although not in the final map e write the ini file for the processing defining all directories parameters file and choosing a very low frequency cut half length of the time stream allows to e distribute the data segments with sanePre and compute pixel indices and a naive map with sanePos P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 0 01 PE YO s a A JE m s e th S 0 00 5 go o0 T Ey 001 F o 3 SD 0 0 B 0 02 8 2 c 0 04 2 e S 0 03 a c 0 06 S 0 04 on Z Z e 0 08 S 005 gt 0 06 0 10 0 07 0 12 2000 0 2000 4000 6000 8000 10000 12000 14000 16000 18000 2000 0 2000 4000 6000 8000 10000 12000 14000 16000 18000 Readout Index Readout Index Figure 6 Correlated signal drifts in the PACS bolometer timelines top panels are fit by interactive curve fitting processing This Figure shows the resulting maps when the fit is inappropriately left side and when drift is mitigated properly right side While most users will likely show due diligence in removing the correlated drift the resulting systematics may be noticeable in automatic pipeline processing for which a one size fits all default values are applied P ACS Document PACS mapmaking Date November 1st
18. onto the pixellized sky defined by the astrometry parameters This module also takes care of performing an initial filtering of the data by rejecting timelines with a percentage of flagged redaouts higher than a user specified threshold and of setting the unit measure as specified by the user http pchanial github com pyoperators and http pchanial github com tamasis map P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 e Pre processing this module detects signal jumps due to cosmic rays Where jumps are detected the data are flagged and the timeline is broken into two independent timelines The module may also remove an initial signal tilt due to the memory of the calibration block which can be found at the beginning of the timelines As a last step the module linearly interpolates the flagged data e Glitch this module performs a high pass filtering of the timelines and carries out a glitch search on the high pass filtered data A sigma clipping approach is used according to which for each pixel the outliers i e readouts with a difference from the median value larger than a user selectable multiple of the standard deviation are marked as glitches After detection the marked values are reconstructed using linear interpolation e Drift this module estimates and removes the polynomial drift affecting the timelines It exploits an Iterative implementation of a Subspace Least Square approach
19. small medium very large medium small medium large large large medium large PACS mapmaking November 1st 2013 Coverage deep medium shallow medium deep medium deep medium deep deep medium medium medium 1 0 Background flat flat flat flat bright structured faint structured flat fairly flat fairly flat bright structured Table 1 Summary of the real PACS data sets involved in the map making codes benchmarking P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 4 1 1 Processing notes The main current limitation of JScanam is its large memory requirements Around 90 GB of memory were necessary to reduce the bigger fields Atlas map but most maps can be reduced with less than 10 GB The latest version of JScanam to be included in Hipe 12 has improved in this respect and requires half of the memory than the Hipe 11 version 4 2 MADmap processing NOTE Some of the images produced via the MADmap pipeline for this report did not have optimal signal drift correction in the pre processing stage see the discussion below on pre processing The resulting maps show significant sloping backgrounds which affect pixel to pixel and differenced image metrics most strongly The photometry from these maps 1s consistent with expectations and is considered reliable The artifacts are a result of improper pre processing of timelines resulting from the time constrai
20. that is 1 extracting a common region from PACS and IRIS data 2 putting the PACS and IRIS data in the same units 3 applying color corrections 4 convolving PACS data to IRIS resolution 5 rebinning the convolved PACS data on the IRIS grid 6 generating scatter plots of the PACS vs IRIS data sets 7 fitting the obtained distribution and deriving offset and gain P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 8 evaluating the results in the light of possible beam effects Below we discuss more in detail the crucial steps of the anlysis Unit conversion since the PACS units are Jy pixel and the IRIS units are MJy sr we con verted the PACS maps into surface brightness units Taking into account the pixel size of the M31 and NGC 6946 PACS data sets the conversion factor amounts to 4154 80 Note that the Tamasis units are already MJy sr Color corrections and wavelength scaling to compare data produced by two different instruments one has to apply a color correction to both instruments and then a wavelength scaling to one of the two The first correction which we denote cc allows the convertion of the the monochromatic flux density measured by each instrument at the reference wavelength to the true source Spectral Energy Distribution SED flux density This correction is always necessary since every in strument quotes the measured flux under a fixed assumption for the input
21. the discrete difference operator along the rows and columns of the map 4 5 1 Processing notes The Tamasis reconstructed maps show cross like artifacts around bright point sources This artifact is produced by a mismatch between the estimated trajectory of each bolometer on the sky and the real one This mismatch has several origins most notably the errors in the pointing reconstruction but also uncorrected optical distortions By slightly changing by around 196 the physical size of the PACS detectors cross like artifacts were significantly mitigated p AC S Document PACS mapmaking Date November 1st 2013 Version 1 0 Herschel Field Crab Field Crab band blue obsid 1342204441 group 0 band blue obsid 1342204441 group 0 median TOD TOD baseline fit Myjy sr Myjy sr 0 2000 4000 6000 8000 10000 12000 14000 16000 0 2000 4000 6000 8000 10000 12000 14000 16000 Frame Frame Figure 7 Left panel The blue line shows the median of TO Dobs TOD map over the detectors group 0 at the first baseline removal step see text in red the fitting linear relation that is subtracted from the signal Right panel Removal of the baseline with a spline Field Crab band blue obsid 1342204441 group 0 Field Crab band blue median TOD TOD Myy sr 22 0000 Dec 2000 0 0 2000 4000 6000 8000 10000 12000 14000 16000 5 35 00 Frame R A 2000 0 Figure 8 Left panel The blue line shows
22. the first processing a bug was found that prevented the average drift of the first scan leg of each obsid to be faithfully propagated in other words the average drift was correctly computed but not accurately subtracted The red band of two PACS real data sets i e NGC 6946 and M81 which were particularly affected by this bug were reprocessed with a revised version of the code e g 20 1 For these fields and band the improvement in the outer parts of the map is visible by eye 4 5 Tamasis processing Before executing the optimal reconstructions of the maps the data have to be pre processed The pre processing is divided in three steps The first step removes the baseline i e the drift caused by variations of the detector plate temperature The second step consists in the identification and masking of jumps in the timelines These jumps are caused by glitches in the electronics also called long glitches The final step is the second level deglitching to remove short glitches e Baseline removal As in the case of MADmap Tamasis optimal reconstruction gives the best results if correlated trends or baseline are removed from the signal timelines before the reconstruction The estimation of the baseline is not a simple task in particular when one wants to preserve extended emission In Tamasis this operation is made possible by assuming that if a field is observed more than once and in different directions then possible trends affe
23. the mappers from comparison of PACS re processed data with ancillary IRAS or Spitzer MIPS data At 100 um all the mappers are typically within 10 from the IRAS measurements with the only exception of MADmap which is an outlier with a 30 96 discrepancy At 70 and 160 um PACS re processed data are on average within 10 to 4096 with respect to the MIPS data with the largest discrepancies due to residual artifacts e g stripes in the MIPS data Noise for the faint background case and in the blue band Tamasis and Scanamorphos are the mappers which introduce less noise although of correlated type while the other mappers tend to be characterized by higher uncorrelated noise In the red band all the mappers introduce correlated noise Furthermore in both the blue and the red band for all the mappers the noise is typically more pronounced along the scan direction e g Tamasis in the blue band with P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Metric JScanam MADmap SANEPIC Scanamorphos Tamasis Unimap Power spectrum 5 4 5 5 5 5 5 Difference matrix 5 3 x Ook 3 3 4 Bright point source photometry 4 5 4 5 3 4 5 4 5 4 5 Faint point source photometry 3 9 3 9 2 9 O 3 6 4 2 Comparison with IRAS 5 3 0 5 5 5 5 Comparison with MIPS 4 5 3 9 3 0 4 8 4 5 4 7 Noise analysis J 4 3 4 7 4 4 5 Global assessment 30 9 29 9 Zor 31 6 29 9 31 9 Table A The grades are from 0 to 5 with 0 denoting the minimum grade and
24. using the centroid coordinates resulting from the fit for centering the source aperture using 6 and 12 aperture radii for respectively the blue and red band m Co N using 25 and 35 sky apertures around the source for respectively the blue and red band 5 estimating the photometric errors by placing apertures on empty regions around the source The extracted fluxes were then compared to the reference flux values obtained from the HPF maps and the ratio between the two was computed Additional aperture photometric measure ments were done by adopting larger apertures with respect to the standard ones i e 10 and P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 M 2 S 1 05 x x 2 po pn 1 o o C C o o 9 o 0 95 ed e 0 9 0 85 0 8 0 1 2 3 4 5 6 Mapmaker Mapmaker 2 2 x x 2 o o S 8 9 o e e Mapmaker Mapmaker Figure 14 Top panels photometric measurements in the Rosette field at YO um Left panel 6 aperture radius Right panel results for 10 aperture radius For each mapper the average ratio between the source flux exctracted from the reprocessed map and the reference flux in the HPF map is plotted Along the x axis from left to right 1 Scanamorphos 2 JScanam 3 Unimap 4 Tamasis 5 MADmap Bottom panels photometric measurements in the Rosette field at 160 um Left panel 12 aperture radius Right pane
25. x rr ETT a 2 MIPS 160 micron MJy sr E TET aT Lit n PACS 49 0257 1 35207 MIPS 400 0 NGC6946 Unimap BL FI T Td TY 400 CA ae e N e e 100 PACS 160 micron MJy sr PACS 44 6857 1 34779 200 MIPS 160 micron MJy sr Figure 21 continued Pixel to pixel PACS vs MIPS scatter plots and corresponding linear fit for NGC 6946 at 160 um PACS mapmaking November 1st 2013 Version 1 0 PAC S si Herschel Rosette JScanam 2000 1500 MJy sr o Q Q PACS 70 micron n ae o Rosette Scanamorphos 1000 PACS 56 8434 1 03520 MIPS 2000 1500 MJy sr 1000 500 PACS 70 micron Rosette MADMop 2000 1000 PACS 70 micron o 1000 0 1000 MIPS 70 micron MJy sr MIPS 70 micron MJy sr Figure 21 continued Pixel to pixel PACS vs MIPS scatter plots and corresponding linear fit for Rosette at 70 um Field JScanam Antennae 1 61 Crab 1 28 IC 348 0 95 LDN 1780 A SA M31 1 08 M8 1 13 NGC 6946 2 91 Rosette 0 66 Table 8 Gain factors from 70 um comparison with MIPS Values denoted with indicate low quality fits MADmap 2 20 1 24 0 35 o 15 1 20 1 14 2 09 0 98 2 00 SANEPIC Scanamorphos 1 56 1 24 1 54 29 1 02 1 12 2 06 1 03 Tamasis 1 59 1 26 1 54
26. 0 100 150 200 0 50 100 150 200 0 50 100 150 200 MIPS 160 micron MJy sr MIPS 160 micron MJy sr MIPS 160 micron MJy sr M31 SANEPIC M31 Tamasis M31 Unimop 200 T T T T T T T T T T T T T T T T T T 200 T T T T T T T T T T T T T T T 200 T T T T T T T T T T T T T T T T T T T L PACS 17 2898 0 870160 MIPS 4 L PACS 15 5946 0 966872 MIS m A 4 L PACS 16 8907 0 932200 MIPS 4 w xk li F L A ins aes F iso PEL m J _ 1504 1 _ 1501 1 150 e h H 4 h F 4 7 H 4 3 100 100 100 o Ir 7 o r 7 o r 7 E E F J 5 L E F E F E F 8 sot 4 S sor 4 8 so A m n n Ba uuu do s x So ou o x ES 50 AA ERE TO ap eee E IR E E 0 50 100 150 200 0 50 100 150 200 0 50 100 150 200 MIPS 160 micron MJy sr MIPS 160 micron MJy sr MIPS 160 micron MJy sr Figure 21 continued Pixel to pixel PACS vs MIPS scatter plots and corresponding linear fit for M31 at 160 um PACS Herschel MB1 JScanam Document Date Version MB1 Scanamorphos PACS 6 78930 1 11998 MIPS O O A al ua sac Eu ua eoa I aeo cae a E se ug a oa ua PACS mapmaking November 1st 2013 1 0 M81 MADMop 300 250 200 150 100 xit SE e rota Tob doa PAT PACS 70 micron MJy sr 50 koy ei y a ral co Nig ls prey el SI 300 E T T T T T T T T T T T T T T T T T T T T T T T T T T T
27. 0 500 600 700 0 100 200 300 400 500 600 700 MIPS 70 micron MJy sr MIPS 70 micron MJy sr MIPS 70 micron MJy sr Crab SANEPIC Crob Tamosis Crab Unimap EE AA O RISE EE oo ARS E Foe A a LL RE AAA E y E a Go Len E en anan nn LAA i TILDA TTT TTT Jou a oue A Leo LAA Lan TTTT 1400 PACS 104 121 2 56111 MIPS Ed PACS 14 5825 1 26211 MIPS y PACS 42 4585 1 24667 MIPS F 600 X 600 4 1200 L x a L ae M L A L E LI PII M t og T aus nm F Ca Ca gt 1000 k gt gt 3 ES E 400 400 6 8oop 6 6 E L 9 9 E L E E o 600 o o s s s s m r m a L 200 200 E 400 z E 200 L 0 0 0 0 100 200 300 400 500 600 700 0 100 200 300 400 500 600 700 0 100 200 300 400 500 600 700 MIPS 70 micron MJy sr MIPS 70 micron MJy sr MIPS 70 micron MJy sr Figure 21 continued Pixel to pixel PACS vs MIPS scatter plots and corresponding linear fit for the Crab data set at 70 um IC348 JSconom IC348 Sconomorphos IC348 MADMap 2000 T T T T T T T T T T T T T T T T T T T T T T T T T T T T T T T T T T T 2000 T T T T T T T T 2000 T T T T T T T T T PACS 56 5319 0 948279 MIPS PACS 44 2048 1 53561 L PACS 109 513 0 852306 MIPS 1500 F Il 1500 F Y queer 1 g 1000 1000F c F c s c H o o o L 3 1000 7 E L E E f E L R soo R R soo m m m 500 j
28. 1000 T T T T T T T T EN T T T T T 1000 T T T T T T T T T w T T T T T T T PACS 0 295327 1 58R63 PACS 17 9362 1 58099 MIP x 1 mus L i J L J 800 A 800 7 T ow x 2 600 i 600 s T gt o L J o li J ke E E E o 400L 4 o 400 4 m ds m de m L m L 200 200 sd 0 id L 1 1 L L 1 L 1 1 L i L L 1 1 L L 1 L L L 1 L i L 0 200 400 600 800 1000 0 200 400 600 800 1000 MIPS 70 micron MJy sr Antennae data set at 70 um PACS 160 micron MJy sr Antennoe JScanam MIPS 70 micron MJy sr the Antennoe Scanamorphos T T T T T T 300 T T T T T T T T le w T T T T T T T 3060 T T T T T T T T T T T T T T T x T x T T 306 T T T T T T T T T T T MIPS 18 9223 1 3264 PAS 5 J E PACS 18 7706 1 40790 MIPS un 7 250 J 250 x J 250 H ol x 200 gt zu 5 200 e E E 150 7 5 1507 7 5 150 7 E t 7 9 7 E E 100 4 g top 4 g 100 4 o la m 50 sor 4 50 0 4 0 4 0 4 50L L L L L J L L L L L L L 1 L L L 1 J L L L L 1 L L L a 50 L L L L J L L L L L 1 L 1 L L 1 1 J L L L L 1 L L L Hl 50 L L L L 1 L L L L 1 L 1 L L 1 1 L L L L 1 L L L 3 0 50 100 150 200 250 300 0 50 100 150 200 250 300 0 50 100 150 200 250 300 MIPS 160 micron MJy sr MIPS 160 micron MJy sr MIPS 160 micron MJy sr Antennae Tamasis Antennae Unimap 300 T T T T T T T T T T T T T
29. 3 Herschel Version 1 0 5 1 Power spectrum analysis One of the most powerful methods to evaluate the performance of a code for map making is through a power spectrum analysis which allows us to estimate how well a given map making algorithm is able to preserve the flux present in map at different angular scales For this metric we used version 3 of the reprojected hybrid simulations both for the faint and bright case see Section 3 1 For each simulated data set processed by the map making codes we generated a 2D angle averaged power spectrum using an IDL package written by Jim Ingalls The package performs the following operations e it computes the Fourier Transform FFT and applies a normalization using the number of elements in the image e it renormalizes the average 2D power spectra by the summed square surface brightnesses of the original image e it computes k bins where k is in units of X and X in arcmin is the maximum between the x and y dimensions of the image e it computes the average value in each k bin e if beam corrected the FF T image is divided by the FFT of the instrument beam before computing the power spectrum After computing the power spectra for the reprocessed simulated data both beam corrected and uncorrected we compared these with the power spectra of the original input simulated data sets which we denote true sky A power law is fitted to each power spectrum Note that a white nois
30. 501 o 5 L 3 E E H Ed 5 100F so R sop R t we Non x 8 m E af F ES E 50r 0 t Bonn a one L L 0 50 L L L L 1 L L 1 L 50 L L 1 4 1 L L L 1 L L L 1 L L L L 50 0 50 200 0 50 100 150 200 0 200 100 MIPS 70 micron MJy sr MIPS 70 micron MJy sr PACS 56 6995 1 97143 MIPS LDN1780 Tamasis T T T T T T T 150 e o rrr fap AAA AAA PACS 70 micron MJy sr a Q N Q o e e o o PACS 70 micron MJy sr a Q 100 MIPS 70 micron MJy sr fit for LDN 1780 at 70 um PACS 160 micron MJy sr Figure 21 continued Pixel to pixel PACS vs MIPS LDN1780 JScanam 200 150 100 50 PACS 24 1657 0 952315 MIPS 200 LDN1780 Unimap T T T T T T T 250 300 PACS 29 6520 2 11275 MIPS 100 MIPS 70 micron MJy sr scatter plots 200 150 100 50 PACS 160 micron MJy sr 100 MIPS 160 micron MJy sr PACS 23 5303 1 02365 MIPS LDN1780 Scanamorphos T T T T T T T T T T 50 100 150 200 MIPS 70 micron MJy sr T 1 L L 1 L 150 200 and corresponding linear LDN1780 MADMop T T T T T T T T 200 150 100 50 PACS 160 micron MJy sr 200 LDN1780 Tomosis 100 MIPS 160 micron MJy sr PACS 6 75542 0 615593 MIPS 200 LDN1780 Unimop T T T T T T T 100 200 MIPS 160
31. ADmap at 160 um for both the Antennae and NGC 6946 For IC 348 and LDN 1780 at 70 um the x of the fits are unusually high for all the mappers In these cases an analysis of the MIPS data evidences the presence of severe stripes which significantly affect the quality of the fits PACS 70 micron MJy sr Figure 21 Pixel to pixel PACS vs MIPS scatter plots and corresponding linear fit for PACS Herschel Document Date Version PACS mapmaking November 1st 2013 1 0 Antennae JScana Antennoe Scanomorphos Antennae MADMop 1000 T T T T T T T T T TH T T T T T T 1000 T T T T T T T T wu T T T T T T T L T T T T T T T T T T T T T T T T T T T T T T T T T T T PACS 20 7996 1 60002 MIP PACS 17 8830 1 55981 MIPS 1400 PACS 30 7239 2 19929 MIPS zj L i ux L a E 800 E 4 800 1200 4 L yo uw d L gt L 1000 3 x 2 fe 600 4 2 e00 3 L 8 5 800 E 9 L E E E 400 x o A400 o 600 4 k m E m F 4 L m L m L z aob 4 206 206 L L 200 L t e 1 L L 1 L L 1 1 L L L L L 1 1 L 1 1 L L 1 L L 1 1 L L 0 L 1 L 1 L L L 1 1 L L L L 1 1 1 L 1 1 1 L l L 0 200 400 600 800 1000 0 200 400 600 800 1000 0 200 400 600 800 1000 1200 1400 MIPS 70 micron MJy sr MIPS 70 micron MJy sr MIPS 70 micron MJy sr Antennae Tamosis Antennoe Unima
32. Extended emission photometry IRAS comparison the comparison between PACS and IRAS IRIS data at 100 um for only one data set M31 shows that there is generally a good agreement with IRIS fluxes for all the mappers except for MADmap for which the comparison retrieves a gain factor of the order of 30 e Extended emission photometry Spitzer MIS comparison the agreement be tween the PACS reprocessed data and the MIPS data in the surface brightness linear regime is typically within 10 40 Noticeable departures are however found such as SANEPIC at 70 um for the Crab and MADmap at 160 um for both the Antennae and NGC 6946 For IC 348 and LDN 1780 at 70 um the x of the fits are unusually high for all the mappers In these cases an analysis of the MIPS data evidences the presence of severe stripes which significantly affect the quality of the fits e Noise analysis in the blue band Tamasis and Scanamorphos introduce less noise al though this is of the correlated type orange peel while the other four mappers are characterized by flat uncorrelated white noise salt and pepper Moreover the mappers do not have an isotropic behaviour and in particular lamasis introduces more noise along the scan directions In the red band all the mappers introduce correlated noise but as in the blue band the spectrum of Scanamorphos and Tamasis drops at high frequen cies The spatial spectrum of SANEPIC and Unimap is almost isotropic while the ot
33. L a OF eJ 50L L L L L L L 1 L L L 1 L L L 1 L L 1 J L L L L 1 L 1 L 0 20 40 80 100 120 140 60 MIPS 160 micron MJy sr PACS 160 micron MJy sr 150 e o T T T T T T T T T T T T T T T T T T T tn o 0 20 MB1 Scanamorphos PACS 13 5576 1 06854 MIPS L L L L L L L 1 L L 1 L L L L 1 L 1 L 40 60 80 100 120 140 MIPS 160 micron MJy sr 150 PACS 160 micron MJy sr an te o o M81 MADMop Pixel to pixel PACS vs MIPS scatter plots and corresponding linear PACS 11 1448 1 06527 MIPS iaa ca a O JE aia e E a a Ea a gy 0 20 40 60 80 100 120 140 MIPS 160 micron MJy sr M81 Tamasis 150 e o PACS 160 micron MJy sr ta ke PACS 5 69236 0 976537 MIPS pg gg gg ga gig ap egy oo 20 40 60 80 100 120 140 MIPS 160 micron MJy sr PACS 160 micron MJy sr 156 T T T T T T T T T T T T T T T T T T T T T T T T T T T T PACS 12 0353 1 02389 MIPS 100 oo 4 50 lt or zl 50L L L L L L L L L L J L L L 1 L L L 1 L L L L 1 L J L o 20 40 60 80 100 120 140 M81 Unimap MIPS 160 micron MJy sr Figure 21 continued Pixel to pixel PACS vs MIPS scatter plots fit for M81 at 160 um and corresponding linear PACS Herschel 500F T7TTTTTTTTTTT 400 5 gt 3 300 c 5 2 E 200 o 2 amp 100 NGC6946 Scanamorphos Th EI TET
34. P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 PACS Map making Tools Analysis and Benchmarking Coordinator and report compiler Roberta Paladini Authors Alphabetical order Babar Ali Bruno Altieri Zoltan Balog Alexandre Beelen Stefano Berta Pierre Chanial Javier Gracia Carpio Vera K nyves Gabor Marton Roberta Paladini Pasquale Panuzzo Lorenzo Piazzo Helene Roussel Roland Vavrek Michael Wetzstein Summary The PACS photometer blue 70 wm 100 wm and red 160 wm detectors are bolometers This type of detectors has noise characterized by a power spectrum that rises at low temporal frequencies often referred to as 1 f noise The removal of 1 f noise typically takes place during the map making process i e the process of turning time ordered data hereafter TOD into an image in the sky If the correction for 1 f is not accurate stripes even severe are left in the final maps compromising their quality In addition to correct for 1 f noise the map making process also removes from the signal the bolometers common mode drift This step of data reduction is typically denoted pre processing as it occurs prior to correction for uncorrelated 1 f In this report we summarize the results obtained from the comparison of the performance of six map making packages for PACS photometer data The codes which participated in this exercise are in alphabetical order JScanam MADmap SANEPIC Sca
35. PDF p AC S Document PACS mapmaking Date November 1st 2013 Version 1 0 Herschel E A E IES A ki C hor ee 2H d pe m ALT Spectrum dB jpscan madmap sanepic scanamorphos tamasis unimap 0 50 100 150 200 250 Spatial Frequency Cycles in 501 samples jpscan madmap sanepic 19 AN Se r ao A tal r TUER M e r ken scanamorphos n i tamasis B V em MEM unen m 21 Ne Ne Dm ane pp pon ere sasa esse nn asa se Bt a Aa ro ene e e ASS M te man E ti ka 2 i 5 20 EE YAA een en a ode na c ec tee eee eee mn ES UAE Me E EE LEE k a on ee E TN amp 23 A E a O ER C A UM ES i a mei A LL dr A E cai T en m h EN INE MM SINN RAE ho Me RAE CERE CO DRE eT a pera IR SE A ct D toD EE ERE MEO A ERI RN MS eget 25 M m t kana otan ao ae pe ne Tt n iii nn ka nan O 10 20 30 40 o0 60 Spatial Frequency Cycles in 501 samples Figure 24 Top panel 1D spectrum of the YO um Atlas field processed with the different codes Bottom panel zoom of the 0 to 50 spatial frequency range The DC has been zeroed zx indicates the half power frequency P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 16r i 1 1 i a i 18 p A JN f nee uU v n 2 ee lie LN ye on AS ojere Tm RR d Dod e ISA e See eo tn EOSS PE k p Man JA L
36. S Strue VS Strue for the bright 70 um case No offset correction applied P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 a 2 O00 fennen ensnnassnsesessonnnesennesoconaonn 3 E EE ga a BE O ga oF _9 10 L E E a b 9 0 15 Lo 0 15 5 0 05 010 0 15 0 20 0 05 010 0 16 0 20 Truth Faint red Reprojected Jy Truth Faint red Reprojected Jy 5 A 38 30 ES ER L Eo EO SE pE g 0 ph o5 F Eo gu yo E E sa 2 A 0 05 0 10 0 15 0 20 0 05 0 10 0 15 0 20 Truth Faint red Reprojected Jy Truth Faint red Reprojected Jy 2 Z 0 00 se En ga 0 05 9E 5 9 Q 10 S 0 15 z 0 05 010 O15 0 20 Truth Faint red Reprojected Jy Figure 12 continued Pixel to pixel scatter plot of S Suus vs Strue for the faint 160 um case No offset correction applied P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 2 T 1 Vil T T 2 A E 3 E E Fa 0 1 a Eg ES LA 02 L a g yA oe D 0 03 mwe z L b AAA Inn TE A GARRAS 01 02 03 04 05 08 amp 01 02 03 04 05 06 Truth Bright red Reprojected Jy Truth Bright red Reprojected Jy Se 33 ES ES ES E RE J ta m Y o amp og 28 F ye mi L E L a 01 02 03 04 05 06 5 01 02 03 04 05 08 Truth Bright red Reprojected Jy Truth Bright red Reprojected Jy D A p ag SE
37. T 300 E PACS 6 32636 1 12789 MIPS E 250 L 4 250 L F zx F L x 4 at E T 200C nx u a T 200L N DOL AS 7 f di f 1 i f e E 1 e E E ko E E E o 1007 zi o 1007 E 7 m L o L l m z SOT 7 E 50 0 4 0 50 L 1 4 4 L L L L 1L L 4 4 L L L L 1 L L ki 50 L 0 50 100 150 200 250 300 0 50 MIPS 70 micron MJy sr 300 250 200 150 100 PACS 70 micron MJy sr 50 0 50 M81 Tamasis 100 PACS 33 3575 1 13560 MIPS idea aoc po og a 7E n ji a Ea a a Ea a a O O pra O O A MEME 100 150 200 250 MIPS 70 micron MJy sr 300 150 MIPS 70 micran MJy sr PACS 70 micron MJy sr PACS 19 0766 1 FINT NUT ip A O LUE api rp a ST Tm TET NT T a a O api 14164 MIPS aga eo yg ax eg pep ye Eae eg A a a arnica on 50L 200 250 300 0 50 100 M81 Unimop 300 T T T T T T T T T T T T T T T T T T T T T T T T T T T T T T T T Fi PACS 6 50246 1 07957 MIPS 250 200 150 100 y 50 x or e OO a ii a aa a A a a ua T ag ci a a a s Eg as 7 50 0 50 100 150 200 250 MIPS 70 micron MJy sr 150 200 250 300 MIPS 70 micron MJy sr 300 Figure 21 continued fit for M81 at 70 um MB1 JScanam 150 T T T T T T T T T T T T T T T T T T T T T T T T T T T T T PACS 13 1647 1 00549 MIPS 1 5100 2 L e H 4 g E si E 501 7 e F 4 g
38. T T T T T ye T 200 T T T T T T T T P 5 me T T MT T 200 T T T T T T T T x y T T T L PACS 12 2263 1 04219 MI a 4 L PACS 8 09916 0 991181 MIPS 4 L LI a x hw 4 E Mou x Li El le x x n J L x e x 5 x 4 _ 1507 z 1507 I Le x 7 E FO f ee DW ZO l 100 E amp 100 c F 4 Ss H 4 P t J 2 t J e l 4 e L 4 E E a R sor R 50 m l 1 o L Hi x i Hi a vu 7 n 7 50L L 1 1 L L 1 1 L l L 1 1 L L 1 1 L 50L L 1 1 L L 1 1 L l L 1 1 L L 1 1 L 0 50 100 150 200 0 50 100 150 200 MIPS 70 micron MJy sr MIPS 70 micron MJy sr Figure 21 continued Pixel to pixel PACS vs MIPS scatter plots and corresponding linear fit for M31 at 70 um M31 JScanam M31 Scanamorphos M31 MADMap M31 red fits 200 T T T T T T T T T T T T T T T T T T T 200 T T T T T T T T T T T T T T T T T T T 7 200 T T T T T T T T T T T T T T T T T T L PACS 20 5706 0 895565 MIPS 4 PACS 15 7965 1 07651 MIPS 4 L PACS 14 1551 1 00811 MIPS J L J L Zon x 1 E I 150 7 150 OS 150 a t L 4 t L 4 E A J Boul i 3 esr x B avel i 100 100 100 e e e e 7 e F 4 El H 4 E F E r E F sot E sop F 8 sor E m m m L oH 4 oH 4 or 4 AAN AE 4 uoa us dE su a A A E A soi A M ays ap ete ES n rare ae A 0 5
39. amely 1 Power spectrum analysis 2 Difference matrix 3 Point source photometry bright and faint sources 4 Extended emission photometry comparison with ancillary data sets IRAS Spitzer MIPS 5 Noise analysis The result of the metrics are summarized below and in Table A Removal of emission on some angular scales no significant flux removal on scales larger than a few instrument beams is found for any of the mappers T he only exception is represented by MADmap for the faint background case both in the blue and red band for which we report a slight flux removal at relatively large angular scales Point source photometry in the bright flux regime 0 3 50 Jy all mappers provide flux measurements consistent within 5 10 with High Pass Filtering HPF For fainter fluxes 0 001 0 1 Jy in the blue band all mappers are in agreement with HPF down to 0 03 In the red band the agreement with HPF is found only down to 0 3 Jy i e an order of magnitude higher than in the red band The only exception for both the bright and faint flux regime and both the blue and red band is SANEPIC for which larger discrepancies with respect to HPF are evidenced by this photometric analysis Finally we find no indication at this stage that any of the mappers allows us to reach a higher flux depth compared to HPF Extended emission photometry no systematic discrepancy neither in the red nor in the blue band is found for any of
40. at any signal gradient between the beginning and the end of the timestream will be removed if the observation does not end where it started on the sky then any large scale gradient between those two points will be filtered out This leads to very large scale filtering of bright gradient in PACS maps as the observations often start and end on the two extreme points on the map Note that apart for very large scales that are not measurable by a Fourier analysis of the map all the other scales are preserved This implies that any difference from a truth map will show a large gradient while any Fourier analysis of the map will show a transfer function close to unity Noise 18 stationary the noise properties are described by a single power spectrum for a given data segment implying that across the data segment the noise must be stationary having the same properties from start to end This is very well the case for PACS receivers If there are strong noise properties changes one could partition the data segment in several parts in which the noise is stationary Sky is constant over a pixel As for all the mapmakers SANEPIC assumes that the sky is constant flat over a pixel in the final map This assumption could be broken in the case of 1 a strong gradient in a single pixel 2 astrometric mismatch 3 gain or calibration mismatch between data segments l hese problems in case of strong sources could lead to a wrong determination of the sky
41. bolometers the offset is elec tronically set to approximately 0 The PACS bolometers are however multiplexed and only the mean signal level for individual modules or array can be set to 0 leading to the observed variations in the pixel to pixel signal level This is purely an electronic and design effect Mitigation of this effect entails subtracting an estimate of what the zero level should be per pixel from each of the readouts of the pixel The median of the entire pixel history is used to set the zero point The idea is to compute the median of the entire history of signal values per pixel and subtract this median from each of the pixel readouts The task photOffsetCorr applies this correction in HIPE e Global drift correction Figure 5 illustrates the concept of correlated signal drift It shows the median of an individual readout image in the observation cube plotted against P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 time readout index The rise and decline of the median signal in Figure 3 is not due to change in the intrinsic intensity of the observed source but due to a change drift in the raw signal level that is correlated in time for all pixels 3 discuss the origin of this drift as likely due to drifts in the focal plane temperature of the bolometer array The observed behavior is usually a monotonic decline but may also be more complex such as the one shown in Figure 5 Strong a
42. bright background red blue the stdev of Tamasis in the red band is the highest Scanamor phos Unimap and JScanam show equally low slope corrected stdev 5 3 Point source photometry One of the most important metrics of the codes benchmarking is point source photometry Goal of this test is to check the quality of the flux ectracted from the reprocessed maps taking as a reference the flux estimated from the HPF maps For this purpose two flux ranges were considered one for bright sources 0 3 50 Jy and one for faint sources 0 001 0 1 Jy P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Scanamorphos ii Y C gi JScanamofphos 0 0015 Jy pixel 0 0029 0 0028 0 0025 0 0021 0 0016 0 001 0 00031 0 00053 Figure 11 Faint blue 70 um simulations Right top corner truth reference map All other panels difference maps The quantity shown is Diff median Diff All the maps are on the same scale P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Scanamorpfios TamasisSe Unimap JScanamorphos EE 5 dy pa e X Ww NC EA TO IR EAS TS pus imus 3 a 0 0078 0 0074 0 0066 0 0054 0 004 0 0022 0 00016 0 0023 0 005 Jy pixel Figure 11 continued Bright blue 70 um simulations Right top corner truth reference map All other panels difference maps The quantity shown is Diff
43. ce source images convolve the sky model as well as the point source images with Gaussian PSFs and create composite images used for timeline projection e Using the map index project the sky simulation composite images onto an LI flux cube e Using the standard projection WCS project with photProject the L1 pure sky simulation flux cube with pixfrac 0 1 This will serve as the reference map containing the pure geometrical footprint of the naive projection mechanism e Correct for digitization effect to simulate proper quantization noise e Co add noise and projected sky model flux cubes and copy back to Level 1 simulated cubes e Create Observation Context with simulated Level 1 cubes P AC S Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Co 15 1E Do 18 12d u 15 Figure 3 Intermediate steps of the simulation process from top left to bottom right mapl pure noise donor cube projected onto input sky WCS 1 pixel size map2 the same as mapl but after applying HPF with filterSize 100 map3 jaive projection of pure sky model cube onto standard projection WCS 3 pixel size in the red band map4 pure noise donor cube projected onto small input WCS after applying HPF map5 similar to mapl but projected onto larger pixelseize standard projection WCS map6 naive projection of sky noise cube applying HPF on the timeline P ACS Document PACS mapmaking Date November 1st 2013 Herschel Versio
44. cedure is repeated three times with a threshold decreasing at each repetition In the case of jumps in single bolometers a similar procedure to the one now described is followed for each bolometer TOD s TODmap is computed where the map is generated using all data except for those from the bolometer being analyzed Jumps are identified with the same detection algorithm as above and the bolometers affected by jumps are masked for the entire observation e Second level deglitching Short glitches are corrected using a second level algorithm This method identifies glitches as timeline outliers with respect to the reconstructed map e Reconstruction The map reconstruction is treated as an inversion problem y Hx n 2 where y is the observed timeline or TOD x the map to be estimated and n the additive noise of time time correlation matrix N H is a model of the PACS instrument and can be written as H M C R P where P is the projection of the sky light into the spatially extended PACS detectors R is the thermal response of the bolometers C is the on board lossy compression and M is the masking operator so that masked samples such as the glitches are not taken into consideration in the reconstruction Assuming a smoothness prior on the map the regularized least square solution is obtained by minimizing the criterion J x y Hx N y Hx A Dx 3 where A is the hyperparameter controlling the map smoothness and D is
45. cting the data average out when all the data are combined into a final P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 map Asa first step and for each observation the median of the signal computed over the entire observation is subtracted from the signal timeline of each bolometer e g the TOD A map is then generated by using the data of all observations which results in a map where part of the baseline is averaged out This map x is simply generated by back projecting the TOD y on the sky and by dividing it by the coverage map also called naive projection i e P y 1 X A TOD from this map is obtained from the projection matrix P i e we project the map on the detectors and the difference between this TOD TOD map i e y Px and the TOD i e y from the observation is computed For each group of detector matrix in the red band a group corresponds to a 16x16 detector matrix while in the blue and green bands it corresponds to 2 adjacent matrices the median of TOD TOD over the detectors in the group is estimated and this is fitted with a linear relation as a function of time see Figure 7 left panel This linear relation can be considered as a crude first order estimation of the average baseline for each group and it is subtracted from the TOD of each detector in the group A new map is then generated with the corrected TOD This procedure is repeated twice At this
46. d were used while for the rest of the programs the pixel size was set to 2 for the blue band and to 4 for the red band The only exception is represented by the Abell 370 data set for which a pixel size of 1 2 and 2 4 was used for the blue and red band respectively 4 Data Processing overview 4 1 JScanam processing The processing of the simulated and real PACS data was done with the latest version of the JScanam tasks included in Hipe 11 The JScanam package comes with an accompanying pipeline script that can be easily accessed through the Hipe menu Photometer Scan map and minimap Extended source JScanam scanmap Extended emission JScanam The script processes the data from a Level 1 HSA product to a Level 2 projected map It comes with a set parameters that can be modified by the user but the proposed defaults produce satisfactory results in most cases For this comparison exercise the default parameters were used for almost all datasets In particular the galactic option was always set to true even in those cases were the emission was concentrated on a small region The drift removal is implemented in two separate tasks scanamorphosDestriping and scanamorphosIndividualDrifts The first one removes drifts from the signal with time scales longer than a scanleg while the second one works on shorter time scales Before these two tasks are executed some data pre processing is required Strong signal jumps produced by cosmic ray
47. e brightness linear regime i e up to 50 MJy sr Values denoted with indicate low quality fits P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 scanamorphag tamasis Figure 22 The Atlas 70 um data set processed by the different map making codes P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 5 5 Noise analysis For this metric we selected the Atlas field which is a large sky area essentially free of emission except for a few scattered sources As a consequence the map is dominated by the noise and can be considered a sample of the spatial noise introduced by each map maker The Atlas maps produced by the different mappers are shown in Figure 22 The colorbar is adjusted to leave out 196 of the samples from both tails of the distribution The field was observed twice ie using cross scans but the pixels at the coverage boundary have only one observation These pixels are affected by a higher noise and should be excluded from the measurements together with the pixels containing sources lo this end the pixels not covered or covered by only one observation are flagged Moreover the sources are detected using a simple algorithm and flagged too No flagged data will be used in the measurements We first computed the image variance and the Probability Density Function PDF Prior to the generation of the PDFs we subtracted the mean value from each map
48. e comparable dis tributions The MADmap and JScanam distributions are not symmetric In the red band Scanamorphos has the lowest variances JScanam and MADmap have the highest and the other mappers are intermediate cases In addition JSscanam and MADmap have again a wider distribution All distributions are approximately symmetric Figure 24 illustrates the 1D spectra for the two bands x denotes the half power frequency ie the frequency with respect to which the power is symmetrically distributed i e half of it lies on the left and half on the right In the blue band we note that all the spectra peak at low frequencies At higher frequencies JScanam SANEPIC MADmap and Unimap yield essentially a flat spectrum although with different amplitudes and the half power frequencies lies in the middle of the spectrum On the contrary 5canamorphos and Tamasis performance is frequency dependent The noise drops at high frequencies and is shifted towards low frequencies as indicated by the location of the half power frequency Interestingly the JScanam spectrum P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 tamasis Figure 22 continued The Atlas 160 um data set processed by the different map making codes P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Gunimap Noise variance otamasis oscanamorphos Mapper FO M dE A MN DS a M dis iudei ual Vad
49. e component is added to the true sky before performing the spectral analysis in Fourier space This is obtained by creating a 2D image of normally distributed values with 1 0 width equal to the HSPOT predicted sensitivities at YO um and 160 um The level of white noise added to the true sky amounts to 1 24 mJy and 4 21 mJy in the blue and red band respectively The results are shown in Figure 10 The shaded area in each figure panel highlights the range of angular scales for which the simulations do not accurately reproduce real PACS data These scales typically corresponds to the size of a few instrument beams and below We strongly advise against drawing conclusions on the performance of the map making codes based on the power spectrum behavior at these angular scales At scales larger than the beam for all considered cases bright and faint background blue and red band all the mappers appear to reproduce equally well the power spectrum of the ruth map The only exception appears to be MADmap for the faint background case both in the blue and red band where a slight loss of power at relatively large angular scales is found The HPF case is included as a reference and is a clear outliers in all the plots This is due to the fact that HPF maps are background subtracted thus contain little power on scales larger than the beam P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Truth JPScanam Scanamorph
50. e fact that Unimap can store the intermediate results and restart the processing from any module In most cases the default parameters were adequate to obtain good quality maps However a few needed a finer tuning Generally speaking the default parameters are ok for the first five modules see Section 2 6 The only three parameters that really need to be tuned to the specific image are discussed in the following The GLS map maker needs an initial guess to start the iterations The initial guess can be either a zero map or the naive map We verified that when the observation is signal rich it is better to start from the naive map because starting from the zero map may require too many iterations to converge On the contrary when the observation is essentially a flat background with a few sources it is better to start from the zero map since convergence is faster In general images dominated by the background are more difficult to reduce and require longer processing times P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Metric JScanam MADmap SANEPIC Scanamorphos amasis Unimap Power spectrum M Y N Y Y Y Difference matrix Y Y N Y Y Y Point source phot bright M Y in part Y Y Y Point source phot faint Y Y Y Y dE d Comparison with IRAS Y Y in part Y Y Y Comparison with MIPS Y Y imparte Y Y Y Noise analysis Y Y Y X X X Table 2 Notes see Section 5 3 1 see Section 5 4 1 see Section
51. e remaining drifts are weaker than in the previous step which facilitates the drift removal e the PACS observations were done in pairs of scans and cross scans with close to per pendicular scan directions This is necessary because in some of the JScanam tasks the information from the cross scan is used to remove the drifts from the scan and vice versa and this is only possible when the drifts projections on the sky have perpendicular directions e short term drifts from different bolometers in the array are independent If the map pixel size is large enough to include a significant number of pixel crossings at different times and from different bolometers the drifts average contribution to the map should be close to zero If these three conditions are fulfilled the JScanam tasks are able to remove most of the 1 f noise from the PACS photometer signal preserving the flux from real sources at all spatial scales shorter than the map size We should note that even if they are based on the same principles the J5canam and Scanamorphos implementations are quite different in many points and assumptions This explains why they do not produce always the same results 2 2 MADmap The Microwave Anisotropy Dataset mapper MADmap is an optimal map making algorithm which is designed to remove the uncorrelated 1 f noise from bolometer time ordered data see 5 The removal of 1 f noise creates final mosaics without any so called banding or
52. easy to use and optimal map making tools for Herschel and future generation of sub mm instruments The project is a collaboration between 4 institutes ESO Garching IAS Orsay CEA Saclay Univ Leiden Tamasis code is massively parallel MPI OpenMP and is written in FORTRAN for the number crunching and Python for the abstraction easiness Tamasis is publicly available to the community Tamasis is used not only to reconstruct optimal maps but also to manipulate timeline data and to compute maps with naive projection 2 6 Unimap Unimap is a map maker based on the Generalised Least Square GLS approach which is also the Maximum Likelihood ML method when the noise has Gaussian distribution The method is known since the ninetiees e g 26 and several practical implementations were proposed in the last decade Unimap is specialised for handling Herschel data PACS and SPIRE Unimap is written in Matlab and can be compiled to run on every machine where Matlab can be installed including Windows Linux and Mac Note that it is not required that Matlab is installed to run Unimap Only some runtime libraries are needed Unimap version 5 5 is divided into several modules which are summarised in the following e Data loading the input data to Unimap is a set of fits files each one storing an observation The first module performs the loading of these files and creates the internal data structures including the projection of the data
53. enerate an optimal convolution beam This was done by adopting as a circular gaussian kernel as a first guess and then by determining the position offset and the FWHM of the Gaussian by means of a grid of offsets and FWHM values The results of this analysis are not given in this report as this is still work in progress 5 4 2 Comparison with Spitzer MIPS data The test consists in comparing a sub sample of reprocessed 70 and 160 ym PACS maps with the corresponding MIPS archival data At these wavelengths MIPS has an angular resolution of 19 and 40 respectively The list of data sets used for this metric is provided in Table 7 Note that data selection was carried out with the purpose of covering as wide a parameters space as possible Table 7 shows that the comparison includes both galactic and extragalactic fields and both the faint and bright end in surface brightness The table also indicates which code for every given data set took part in the MIPS comparison The test was done following a procedure similar to the one adopted for the IRAS case and along the lines of what is described in PICC NHSC TR 034 22 i e 1 generating MIPS mosaics with the MOPEX package 12 2 extracting a common region from PACS and MIPS data 3 putting the PACS and MIPS data in the same units P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Map maker MJy sr SANEFIC 0 99 Tamasis 1 02 MADmap 1 28 Scanamarp
54. energy distribution of the source For both PACS and IRIS the adopted convention is that the flux scales as follows an exponential that is flux density F scales with frequency v as F x v The specific corrections to apply depend as well as on the instrument convention for the input energy distribution on the details of the SED of the source and on the wavelength response of the system Here we assumed an ISM SED and adopted the PACS and IRIS color corrections provided in PICC NHSC TN 029 The color corrected PACS and IRIS data can be expressed as MaPpacs MAPpACS CCPACS 5 MaPrRIS MAPIRIS CCIRIS 6 lhe second correction brings the monochromatic color corrected flux at the reference wave length of one instrument to the reference wavelength of the other instrument If the reference wavelength for the two instruments is the same as in this case no correction is needed Convolution the high angular resolution PACS data were convolved to the lower IRIS angu lar resolution by approximating the IRIS beam with a circular gaussian kernel normalized to unity to 7 5 radius This step was carried out using the IDL routines PSF GAUSSIAN and CONVOL Alignining and rebinning all the PACS and IRIS maps were rebinned onto a common grid Non common pixels are set to NaN values The reprojection is performed with the IDL routine HASTROM Possible astrometric offsets were searched for using peaks of emission Sca
55. espectively a relative gain and absolute offset See Section 5 4 1 for more details MIPS flux non linearity The MIPS 70 and 160 um arrays are gallium doped germanium Ge Ga photoconductor detector These types of arrays are characterized by a relatively low absorption coefficient 2 cm 1 and therefore are usually very bulky 0 1 2 mm This large physical size has important consequences on the detectors performance which typically exhibit a variety of non linear behaviours The MIPS 70 um array is linear only below 1 Jy roughly corresponding to a surface brightness threshold of 100 MJy sr For the 160 um detector non linearity effects manifest 2 Jy which correspond to 50 MJy sr In the light of these considerations we adopted a conservative approach and applied the linear fit i e Equation 13 to both the original scatter plots distributions and the distributions obtained by truncating the dynamic range to respectively 100 MJy sr at 70 wm and 50 MJy sr at 160 um The results of the analysis are provided in Figure 21 and in Table 8 to 11 In the interest of space the scatter plots are shown for the original distributions only while the gain factors are provided for both the original and truncated distributions The agreement between the PACS reprocessed data and the MIPS data in the surface brightness linear regime is typically within 10 40 Noticeable departures are however found such as SANEPIC at 70 um for the Crab and M
56. fecting the MIPS detector In particular Unit conversion since the PACS units are Jy pixel and the MIPS units are MJy sr we converted the PACS maps into surface brightness units Taking into account the different pixel size of the PACS data sets the conversion factors amount to 10636 3 Antennae Crab NGC 6946 and 4154 80 LDN 1780 M31 M81 Rosette at 70 jum 2659 1 Antennae NGC 6946 and 1038 70 LDN 1780 M31 M81 at 160 um Note that the Tamassis units are already MJy sr Color corrections and wavelength scaling as mentioned in Section 5 4 1 the PACS convention is F v 1 while for MIPS the flux is assumed to scale with frequency according to a blackbody at 10 000 K In addition we also need to bring the PACS and MIPS measured fluxes at the same reference wavelength that is the PACS surface brightnesses from 160 um to 155 9 um and from YO um to 71 4 um i e the MIPS reference wavelength at 70 and 160 um respectively In order to compute color corrections and wavelength scaling we assumed a fixed dust tem perature for each target in Table 7 In particular the following temperatures were adopted 20 K Antennae M31 M81 NGC 6946 LDN 1780 30 K IC 348 Rosette 50 K Crab Fol lowing these assumptions we then applied the PACS color corrections and wavelength scaling provided in PICC ME TN 038 see Table 3 3rd and 4th columns 20 and the MIPS color correctons given in the MIPS Data Handbook 14 In
57. field at 70 um Shown in the table are the ratios between the reference HPF flux values and the fluxes extracted from the data processed by the mappers R Scanamorphos JScanam Unimap Tamasis MADmap SANEPIC 12 0 97 0 12 1 01 0 12 0 95 0 12 1 08 0 16 0 96 0 104 1 19 0 16 20 0 960 138 1 01 0 15 0 97 0 15 1 08 0 14 0 98 0 19 1 15 0 16 Table 5 Photometric measurements in the Rosette field at 160 um Shown in the table are the ratios between the reference HPF flux values and the fluxes extracted from the data processed by the mappers 5 3 2 Faint sources 0 001 0 1 Jy To evaluate the quality of the photometry in the faint end regime the PEP Abell 370 field was used and the following pieces of analysis were carried out 1 extraction of observed Point Spread Functions PSFs using the Starfinder code 6 following the same procedure as for the PACS Evolutionary Probe PEP survey 2 injection of artificial sources into the science maps using the observed PSF as object pro file and adopting number counts consistent with deep extragalactic observations Sources are injected down to flux limits 10 20 times deeper than the observed flux thresholds 3 extraction of sources from the modified science maps ie those including artificial sources using Starfinder tuned as for the PEP survey 4 derivation of c s i q r semi interquartile range MAD of the in out flux comparison after correcting for possible flux trends in t
58. finite time constant and on board compression effects In addition pointing jitter is not simulated i e an infinite accuracy knowledge of attitude is assumed The Level 1 timelines provided by the simulations are already de glitched using the standard PACS pipeline and both 2nd order de glitch mask and MMT glitch masks are provided in the simulation L1 frames products 3 2 Real Data PACS real data selection was performed to allow us to cover a parameter space as wide as possible in terms of 1 source surface brightness 2 background surface brightness P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 3 background morphology 4 size of sky area covered by the observations 5 observing mode 6 depth i e of repetitions Following the criteria above we selected 10 fields see Table 1 These cover galactic e g LDN 1780 and extragalactic targets e g Antennae scan e g NGC 6946 and parallel mode observations e g LDN 1780 small w 15 x 15 e g IC 348 and large a few degrees e g M31 areas relatively shallow e g Atlas and deep e g Abell 370 programs faint flat e g Crab and bright structured background e g Rosette In generating the maps with the various mapping codes different pixel sizes were adopted In particular for programs covering large areas of the sky Atlas LDN 1780 M31 M81 a pixel size of 3 2 for the blue band and of 6 4 for the red ban
59. hand when a cosmic ray hits the readout electronics the effect is a jump in the signal positive or negative of all bolometers that are read by the affected electronics that is either a single bolometer a line of 16 bolometers or an entire detector group These jumps are seen as a sudden temporal variation either positive or negative of the detector signal The signal goes back to the previous level within a timescale of several tens of seconds T he removal of jumps is crucial for a good reconstruction In Tamasis jump detection in PACS data is performed by adapting an algorithm initially designed to detect jumps for SPIRE bolometers see P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 13 The algorithm consists of two steps both identifies and masks jumps but the first one does that for lines of bolometers while the second one for individual detectors In the first step TO Do6s TOD map is computed for each line of bolometers where the map is generated using all data except for those from the line of bolometers being analyzed Then the median of TODo6s TOD map over the 16 bolometers of the line is computed The resulting value as a function of time see Figure 9 is analyzed with a step detection algorithm based on the Haart wavelet which allows the identification of jumps above a given threshold The line of bolometers for the scan in which the jump is located is subsequently masked l his pro
60. he in out ratio due to flux boosting at the faint level in the Starfinder source extraction process see PEP documentation 5 extraction of catalogs from original science maps i e extraction of potentially real sources and comparison with the PEP official HPF based blind catalogs Each step of analysis is described below 5 3 3 Extraction of observed PSFs The PSFs were extracted using Starfinder piling up the brightest and most isolated point like sources in the field Each PSF i e for each mapper band combination is obtained stacking P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 0 8 Z Abell 370 Green de Abell 370 Red j MADMap p MADMap a SANEPIC a S SANEPIC g 0 8 Unimap E l Unimap JScanam JScanam E Tamasis E ji Tamasis A A sw sw 0 2 0 5 10 15 20 25 0 10 20 30 40 50 radius arcsec radius arcsec Figure 15 Curves of growth comparing the PSFs obtained from maps generated by the different map makers 100 um left panel and 160 um right panel respectively typically 4 5 objects The PSFs are cut to a radius of 15 pixels as they are dominated by noise beyond this radius Using the maps pixel scale this translates to 18 and 36 radii at 100 and 160 um respectively Maps obtained with Scanamorphos represent an exception because their scale is 1 0 pixel and 2 0 pixel in the two bands instead of 1 2 and 2 4
61. hers are more affected by the scan directions References 1 Berta S Magnelli B Lutz D et al 2010 A amp A 518 30 2 Berta S Magnelli B Nordon R et al 2011 A amp A 532 49 3 Billot N et al 2010 SPIE 7741 1 4 Calabretta M R amp Greisen E W 2002 A amp A 395 1077 P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 5 Cantalupo C et al astro ph 09061775 6 Diolaiti E Bendinelli O Bonaccini D et al 2000 A amp AS 147 335 7 Greisen E W amp Calabretta M R 2002 A amp A 395 1061 8 Hennemann M Motte F Bontemps S et al 2010 A amp A 518 84 9 Herschel Map Making Workshop http herschel esac esa int 2013Mapmaking shtml 10 Lutz D Poglitsch A Altieri B et al 2011 A amp A 532 90 111 Magnelli B Popesso P Berta S et al 2013 A amp A 553 132 12 Makovoz D amp Khan I 2005 In ASP Conf Ser 132 113 Meixner M Panuzzo P Roman Duval J et al 2013 AJ 146 62 114 MIPS Instrument Handbook http irsa ipac caltech edu data SPITZER docs mips mipsinstrumenthandbook 51 15 Miville Deschenes M A amp Lagache G 2005 ApJS 157 302 16 Patanchon G Ade P A R Bock J J et al 2008 ApJ 681 27 17 Piazzo L Ikhenaode D Natoli P et al 2012 Artifact removal for GLS map makers by means of post processing IEEE Trans on Image Processing Vol
62. hits on the electronics are identified and masked Frames affected by strong drifts produced after the observation of the calibration source are also masked An especially important pre processing step is the creation of a good source mask T his mask is used by several JScanam tasks but it is particularly important for the scanamorphosDestriping task where small pointing offsets between the scan and cross scan can produce significant effects close to bright sources if they are not masked A good source mask should include the brightest sources but does not need to include all the extended emission It should never cover more than one third of the map PACS Herschel Field Abell 370 Antennae Atlas Crab IC 348 LDN 1780 M31 M81 NGC 6946 Rosette OBSIDs 1342223332 1342223333 1342187836 1342187837 1342189661 1342139662 1342204441 1342204443 1342238774 1342238775 1342238830 1342238831 1342239043 1342239044 1342239433 1342239434 1342240719 1342240720 1342241389 1342241390 1342241460 1342241461 1342241515 1342241516 1342241647 1342241648 1342241697 1342241698 1342241927 1342241926 1342242058 1342242059 1342224989 1342224990 1342224991 1342211604 1342211605 1342186085 1342186086 1342204421 1342204422 1342186121 1342186122 Obs mode scan map scan map parallel mode scan map scan map parallel mode parallel mode scan map scan map parallel mode Document Date Version Field size
63. hos 1 07 Unimap 1 03 JPSeanam 1 02 10 20 30 40 50 IRIS MJy sr Figure 20 Binned scatter plots of the PACS vs IRIS data for M31 at 100 um The PACS data reprocessed by each of the mappers are shown in the figure Note that the offset 1 e parameter B has been removed in order to generate the plot Source MIPS pid A Scanamorphos JScanam MADmap SANEPIC Tamasis Unimap um Antennae 32 70 160 X X X X X Crab 130 70 X X X X X X IC348 40372 58 70 X X X X X LDN1780 40154 10 160 X X X X X M31 99 10 160 X X X only 160 X X Ms1 717 70 160 X X X X X NGC6949 159 10 160 X X X X X hosette 58 10 X X X X X Table 7 MIPS data sets used for the comparison with the PACS reprocessed maps The table content is source name 1 column MIPS program ID pid 2 column wavelength at which MIPS data are available 3 column map making code and availability of corresponding reprocessed data set 4 to 9 column P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Page 53 applying color corrections and scaling to a common reference wavelength convolving at each wavelength the PACS data to the MIPS angular resolution aligning and rebinning the convolved PACS data on the MIPS grid generating scatter plots of the PACS vs MIPS data sets fitting the obtained distribution and deriving offset and gain Oo 00 N C OO A evaluating the results in the light of non linearity effects af
64. ibrated P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Level 1 detector timeline As a result of the map making process an ideal projected map carries only sky signal while the noise component is being separated by the projection algorithm This hypothesis can be well tested by comparing the unbiased input sky model with a compromised quality projected map Deviations from the sky model qualifies the map making process and can be used to characterize the spectral response of the map maker 3 1 1 Hybrid simulations components Hybrid simulations are made up of three components 1 Noise donor calibration observation A long staring observation at a low FIR background position of the sky This dataset consists of pure instrument 1 f noise and telescope background measured for a nominal detector setup detector bias and readout mode A section of this noise sample is copied into the flux cube of the receptor science observation 2 Receptor science observation Placeholder of the simulated dataset Its flux cube is cancelled out the noise donor replaces the Level O timeline of the science data frames when reaching processing Level 1 the sky model is superimposed over the timeline The AOT setup is typical for a PACS scan map observation in terms of scan speed redundancy map repetitions and observing strategy multiple coverages at orthogonal scan angles In addition the spacecraft attitude has been
65. in order to remove the offset We then calculated a 1D spectrum which is obtained as follows Given a window length denoted by L we extracted from the image all the rows We partitioned each row into blocks of 2L 1 samples with an overlap of L samples We computed the Discrete Fourier Transform DFT of all the blocks except for the blocks containing flagged data The procedure is repeated for all the columns of the image Finally the spectrum is given by the average of the squared magnitudes of all the DFTs so obtained The DC is zeroed to get rid of the offset Last but not least we estimated a 2D spectrum according to the following method Given a grid step denoted by G we covered the image with a grid of rows and columns spaced by G samples Next for each row column crossing point we extracted from the image a square block of 2G 1 by 2G 1 samples centered on the crossing We computed the 2D DFT of all the blocks except for the blocks containing flagged data Finally the 2D spectrum is obtained as the average of the squared magnitudes of all the 2D DFTs so obtained The DC is zeroed to get rid of the offset Figure 23 shows the obtained variances and the PDFs in the blue and red band In the blue band Scanamorphos and Tamasis have the lowest variances JScanam and MADmap have the highest and SANEPIC and Unimap fall in between and have the same variances Moreover JScanam and MADmap have wider distributions while the other four hav
66. inally we summarize our conclusions in Section 6 P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Scanam SANEPIC Scanamorphos Tamasis Figure 1 PACS 160 um scan mode data for the Crab processed by the six map making codes P AC S Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Figure 2 PACS 160 um parallel mode data of M31 processed by the six map making codes P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 2 Overview of map making codes for PACS Figure 1 and 2 illustrate examples of the processing of PACS data sets by the six map making tools that we consider for this analysis Below we provide a brief overview of these packages 2 1 JScanam JScanam is the Hipe PACS implementation of the Scanamorphos algorithm developed in IDL by Helene Roussel http www2 iap fr users roussel herschel As Scanamorphos it is de signed to remove the low frequency noise also called 1 f noise typically found in bolometer arrays which produces strong signal drifts in the pixel timelines The algorithm is based on the following assumptions e the drifts power increases with the length of the considered time scale 1 f noise For that reason JScanam starts removing the drifts with the longest time scale beginning from the scan duration continuing with the scanleg duration and finally on time scales shorter than a scanleg In each step th
67. it Qu jpscan madmap 60 A METTE OS soit ken ee S ur I ae sanepic 3 scanamorphos tamasis 50 ane BD Ae re kk peni AD QT RAE ruga ETC AW E a unimap m et UE MO GA e ooo t n o a L o 30 MENU NN ua A Ya ee MEET SEE ea A A RNA AR E 3 O OW bcc A bee ee eel be ee Me ee ee eee ee Nas e n y TE ae 1 H 0 08 006 004 002 0 002 0 04 Value Figure 23 Top panel variance in the Atlas 70 um field after processing with different map making codes Bottom panel corresponding PDF p AC S Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Page 67 Noise variance Mapper jpscan NS d madmap 50 MN NE ER L ai E Aa a w a sanepic ski HN scanamorphos q tamasis ENS unimap AD L ft bi I RIMIS DRE A SEDEM Ag m li IN l A Li Ji 939p amp rann hi i STER NE E n i 1 5 A li il Bi i f i E ap kare kako eee Man kann Mer B ee TO QUUM A han banan ann 01 f AU a f LS n o A ee a Ht ee RAS Wn WU PO S ADMIN a i A Wn i A oA NAN er li NL 2 3 FI 1 yy 1 i RE 008 006 004 002 0 0 02 Value Figure 23 continued Top panel variance in the Atlas 160 um field after processing with different map making codes Bottom panel corresponding
68. l results for 20 aperture radius Along the x axis from left to right 1 Scanamorphos 2 JScanam 3 Unimap 4 Tamasis 5 MADmap 6 SANEPIC The blue dots refer to the case when the outliers are removed from the sample while the red dots represent the results for the ensemble of the sources 20 for the blue and red band respectively This operation allowed us to investigate possible PSF distortion effects in the maps processed by the different codes The results are shown in Figure 14 and in Table 4 and 5 All mappers provide accurate photometric measurements in the considered flux range In particular when the outliers are excluded from a statistical evaluation of the results they all retrieve fluxes within 596 1096 from the reference values ob tained from HPF maps Moreover when the recommended aperture radii 6 and 10 for the blue red band respectively are used and the outliers are rejected JScanam and MADmap in the blue band and JScanam Scanamorphos Unimap and MADmap in the red band retrieve the best photometric results In the red band SANEPIC is an outlier with recovered fluxes typically 20 higher than the HPF values P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 R Scanamorphos JScanam Unimap Tamasis MADmap 6 1 04 0 09 1 00 0 06 1 04 0 06 1 01 0 08 1 03 0 05 10 1 00 0 09 0 98 0 08 1 00 0 09 1 00 0 09 1 00 0 07 Table 4 Photometric measurements in the Rosette
69. l map are fine sampled pixel of 1 convolved with Gaussian beams sampled by 1 pixels and scaled to the desired brightness The simulation procedure is as follows see Figure 3 e Take Level 0 data of a real PACS observation of a dark field i e the noise donor and copy into the Level 0 flux cube of the receptor observation e Add improved pointing product to receptor observation e Run pipeline HIPE v11 876 on receptor cube up to Level 1 frames generation This will produce flux calibrated 1 f noise timeline e Create WCS for input sky model as well as standard WCS for projections based on the receptor cube footprint e Calculate map index obtain the RA and DEC for every sampling point in the Level 1 timelines of the receptor cube then assign pixels from the sky model map at the RA and Dec of a given sampling point in the receptor timeline Map indexing takes into account the fractional overlap of detector pixels and sky model pixels Map indexing from L2 sky model to L1 receptor cube has been done with pixfrac 1 0 e Apply broad width 100 High Pass Filter HPF on the receptor cube with pure instru ment noise Using photProject and the input sky WCS project this cube noise with pixtrac 0 1 e From the above created noise map determine suitable scaling factors required for bright and faint simulation cases e Create a point source catalogue of uniformly distributed fluxes in the 10 0 500 0 mJy range and produ
70. l spectrum especially for Tamasis In the red band while SANEPIC and Unimap are essentially isotropic the other mappers are affected by the scan directions although to different extents Note that the aforementioned noise drop behavior of Scanamorphos and Tamasis at high frequencies both in the blue and red band is clearly visible in Figure 25 In conclusions in the blue band Tamasis and Scanamorphos introduce less noise although this is of the correlated type orange peel while the other four mappers are characterized by flat uncorrelated white noise salt and pepper Moreover the mappers are not isotropic and in particular lamasis introduces more noise along the scan directions In the red band all the mappers introduce correlated noise but again the spectrum of Scanamorphos and Tamasis drops at high frequencies The spatial spectrum of SANEPIC and Unimap is almost isotropic while the others are more affected by the scan directions 6 Conclusions In the following we summarize the results obtained from application of the metrics described in Section 5 e Power spectrum analysis at scales larger than a few times the beam for all considered cases bright and faint background blue and red band all the mappers reproduce equally well the power spectrum of the truth map The only exception appears to be MADmap for the faint background case both in the blue and red band where a slight loss of power at relatively large angular scale
71. level over the filtering length of the data segment thus leaving strong artifacts crosses on the maps In order to avoid those problem SANEPIC P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 can remove the crossing constraints between two data segments using a mask for strong sources whose flux level is determined using a simple mean between data segments Bad Data Glitches Steps Moving Objects since SANEPIC is only a mapmaker data must be properly flagged before being projected Strong glitches steps or any strong non physical gradients induced from simulations with mismatched background level for e g not well described in the Fourier domain will need to be detected and flagged prior to theprocessing This could lead strong features in the maps even crosses for strong glitches or in the case of several faint unflagged glitches to an overestimate of the white noise level 4 4 Scanamorphos processing The processing options used for each dataset among parallel galactic jumps pacs can be found in the fits headers of the processed maps The log files available upon request contain a summary of the processing steps drifts amplitudes observation duration and processing time The processing was first done with version 19 of the code The glitch mask determined within HIPE was disabled and the deglitching was done entirely in Scanamorphos 4 4 1 Processing notes After
72. lysis and Benchmarking Abstract In this report we summarize the results obtained from comparing the performance of six map making packages for PACS photometer data The codes which participated in this exercise are in alphabetical order JScanam MADmap SANEPIC Scanamorphos Tamasis and Unimap All these codes are publicly available and are designed to correct for noise in the data while preserving emission on all angular scales To test the performance of these codes we considered a combination of simulated and real data and a set of five different metrics namely 1 power spectrum estimation 2 difference matrix 3 point source photometry for both bright and faint sources 4 comparison with ancillary data IRAS and Spitzer MIPS 5 analysis of the noise The analysis evidences that all the codes preserve emission on large angular scales and that they are able to handle the PACS data with a high level of photometric accuracy Finally we point out that differences in the generated maps are typically small yet noticeable 1 Context The PACS photometer blue 70 wm 100 wm and red 160 um detectors are bolometers This type of detectors has noise characterized by a power spectrum that rises at low temporal frequencies often referred to as 1 f noise The removal of 1 f noise typically occurs during the map making process i e the process of turning time ordered data hereafter TOD into an image in the sky If the correction fo
73. map makers Top panels 100 um Bottom panels 160 um P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Unimap S HPF S test S HPF S HPF S test S HPF 1 0 001 0 01 0 1 0 001 0 01 0 1 S 100 PEP HPF Jy S 100 PEP HPF Jy Figure 18 Comparison of SANEPIC and Unimap based 100 mum fluxes to HPF based PEP fluxes 5 3 7 Catalog extraction Starfinder was also ran on the original map in order to extract the fluxes of real sources The Starfinder configuration adopted in this case is identical to the one used when working on simulations Likewise the PSF cut used in the extraction is again at 6 pixel which is then properly corrected for aperture losses using a combination of PSF curves of growth and the Mars EEF All map makers recover the PEP HPF based official fluxes within 5 10 at most see Figure 18 and 19 Differences might be due to the adopted EEF SANEPIC represents a catastrophic exception having a factor 6 of flux excess to be understood 5 4 Extended emission photometry and comparison with ancillary data sets lo check the overall reliability of the extended emission photometry in the processed maps we performed comparison with ancillary data in particular IRAS 100 wm and Spitzer MIPS 70 and 160 um We emphasize that these comparisons are affected by several caveats that the user should keep in mind while reading Section 5 4 1 and 5 4 2 A non comphrensive li
74. micron MJy sr 200 7 7 if T E a 200 L PACS 11 9309 0 873894 MIPS 150 150 a L 3 f gt 2 100 100 o r o S F S E F E 8 sot 8 so 2 2 a a 0 0 50l i ue je YN a A l 50 L 0 50 100 150 200 0 MIPS 160 micron MJy sr fit for LDN 1780 at 160 um PACS 20 6835 0 895052 MIPS Ll 100 MIPS 160 micron MJy sr scatter plots i 150 200 and corresponding linear P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 M31 JSconam M31 Scanomorphos M31 MADMop 200 T T T T T T T T m m m E T T T 200 T T T T T T T T T T T T T T Tw T T T T A 200 T T T T T T T T T T m 7 ha T T T T PACS 14 7235 1 0778 MIPS PACS 8 09760 1 02421 MIPS x 4 L PACS 7 85219 1 20181 MING 4 L J 4 Me al b x Li EN L al L 4 J L u n Li n A F a 150 em 150 q T il F 5 6 L J E E F 7 3 3 f os H 4 es 4 t 100 I si E E 4 c 4 c t a o o al o L al 100 4 S E E L 4 E E L el Rot 3 R R 50 4 an F 7 u ue Lod 4 la ab 7 le y L 1 1 L L 1 1 L L 1 L L 50 L 1 1 L L 1 L L L 1 1 L 4 1 L L 50L L 1 1 L L 1 L L 1 1 1 4 1 1 L L 0 50 100 150 200 0 50 100 150 200 0 50 100 150 200 MIPS 70 micran MJy sr MIPS 70 micron MJy sr MIPS 70 micron MJy sr M31 Tamasis M31 Unimop
75. morphos_IDL Tamasis Unimap MADMap HEF Total power Jy pixel 109 E Im LA pa S o Wave number arcmin Truth JPScanam Scanamorphos_IDL Tamasis Unimap 110 Normalised corrected power Jy Beam 0 1 1 0 10 0 Wave number arcmin Figure 10 continued Faint background case Blue band Uncorrected top panel and beam corrected bottom panel power spectra The shaded area encompassing angular scales comparable or smaller than the beam denotes the range of scales where the simulations become unreliable P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Truth JPScanam Scanamorphos IDL Tamasis a Unimap MADMap X HPF a yo gt 2 D O CE 9 O He k k k k 107 KE k k 10 Wave number arcmin Truth JPScanam Scanamorphos_IDL Tamasis Unimap MADMap FE Normalised corrected power Jy Beam Wave number arcmin Figure 10 continued Faint background case Red band Uncorrected top panel and beam corrected bottom panel power spectra The shaded area encompassing angular scales comparable or smaller than the beam denotes the range of scales where the simulations become unreliable P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 5 2 Difference matrix This metric consists in subtracting from each map reprocessed by the different map making codes a t
76. n 1 0 Receptor cube ld OBSID Band Orientation Size Repetitions Scan speed Start time PA R 1 1342216420 blue2 green 45 35x35 1 20 3 20 11 7250 5 Rit2 1342216421 blue2 green 135 35x35 1 20 3 20 11 7250 5 Sky donor image Noise donor cube Projections 00 lid File in Dropbox Pixel size Scaling factors ld OBSID Band on narmalized image 541 Pink noise simulation E bright 1 30 N81 1342182424 blue red 2 simDonor sky cirrus fits faint 1 90 NRZ 1342182427 greerrred Figure 4 Summary of PACS Type 2 simulation components 3 1 3 Final simulated data sets We decided to restrict the PACS simulations parameter space to a single simulation attribute the extended emission brightness The input sky model is a single image which is being scaled to the faint and bright map cases The point source fluxes do not change i e the source to background contrast is larger in the faint simulation case In total we generated 2 simulated Level 1 data sets for the nominal and orthogonal scanning directions The components of the simulation are summarized in Figure 4 3 1 4 Common WCS for model and projected map Irrespective of the photometric band the PSF convolved input sky simulations are provided on a common WCS with a pixel size of 1 3 1 5 Simulations Caveats The simulations include representative instrumental noise and outliers due to glitches in the noise donor dataset However they do not include detector dynamics i e
77. n the degree of accuracy achieved in applying polyno mial models to the correlated drift This is difficult to quantify since it depends partly on user s due diligence and partly on how well behaved is the background drift Note that the PACS bolometers show negligible correlated drift towards the end of the observation day and data taken during these stable periods requires no correction for the correlated drift Experience suggests that projected maps fluxes may be corrupted by as much as 0 30 compared to the best reduction For data that are dominated by the background the trend is relatively easy to model as source contamination is negligible However a more generic approach is needed that is also able to account for data that contain significant astrophysical emission To do this 1000 readout wide bins are created and it is assumed that the minimum values in these bins corresponds to the few actual blank background measurements The idea is that over these 100 readouts at some point the scan pattern observed a sourcefree part of the sky This will appear as the minimum value of the bin The minimum value of each bin is extracted and the resulting trend in the minimum values is fitted with a polynomial usually 2nd or 3rd degree The best fitting polynomial is then assumed to described the correlated global drift and is subtracted from the each readout P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0
78. namorphos Tamasis and Unimap All these codes are publicly available and are designed to correct for noise in the data while preserving emission on all angular scales To test the performance of the individual map making codes we used a combination of sim ulated data and real PACS observations The simulated data were obtained by co adding a synthetic sky signal to pure instrument noise The input sky model is a single image which is being scaled to the faint and bright background cases Point sources are also included in the simulations However their flux is not scaled along with the background hence the source to background contrast is higher in the faint background case In total we generated 2 simulated data sets for the nominal and orthogonal scanning directions Readers should be aware that at this time the synthetic data do not account for detector dynamic i e finite time constant on board compression and pointing jitter effects The simulations described above were complemented by 10 real PACS data sets selected from the archive in order to cover a parameter space as wide as possible In particular we considered P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 faint and bright background sources galactic and extra galactic flelds scan and parallel mode observations small and large maps shallow and deep programs The test campaign was carried out using a set of 5 different metrics n
79. ns of the noise properties as they exclude outliers As a result few map makers produce noise levels very similar to the official PEP HPF based maps see Figure 16 and 17 P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 MADMap Median 0 5 S out S in S out e an 0 001 0 01 0 1 S 100 out Jy Figure 16 Median c s i q r and M A D median absolute deviation of the source distribution computed for each bin The example shown in the figure is for MADmap at 100 um All other mappers are similar except for different noise levels PEP HPF MADMap Unimap PEP HPF MADMap Unimap PEP HPF MADMap Unimap 9 en Tamesis Tamasis Tamasis e S out S in S out S out S in S out a S out S in S out 0 5 siqr MAD 0 001 0 01 0 1 0 001 0 01 0 1 0 001 0 01 0 1 S 100 out Jy S 100 out Jy S 100 out Jy PEP HPF MADMap Unimap PEP HPF MADMap 9 en Tamasis e an e e S out S in S out S out S in S out v S out S in S out 0 5 sigr MAD 0 5 0 001 0 01 Q 1 z 0 001 0 01 0 1 0 001 0 01 0 1 S 180 out Jy S 180 out Jy S 180 out Jy Figure 17 Left to right comparison of c s i q r M A D curves obtained for maps generated by different
80. nt PACS mapmaking Date November 1st 2013 Version 1 0 Page 72 madmap a 100 150 200 50 100 150 200 scanamorphos 30 100 150 200 30 100 150 200 unimap 50 100 150 200 30 100 150 206 Figure 25 continued 2D spectrum of the 160 um Atlas field processed with the different codes The DC has been zeroed P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Page 73 are rejected JScanam and MADmap in the blue band and JScanam Scanamorphos Unimap and MADmap in the red band retrieve the best photometric results In the red band SANEPIC is an outlier with recovered fluxes typically 20 higher than the HPF values e Point source photometry faint sources 0 001 0 1 Jy in this faint flux regime for the blue band the photometric measurements of the different mappers are comparable to the performance of HPF down to fluxes of the order of 0 003 Jy Below this threshold JScanam appears to provide the largest departures from the HPF fluxes while Unimap and Scanamorphos provide the closest In the red band the thereshold of comparable fluxes between the mappers and HPF is an order of magnitude higher with respect to the blue band i e 0 03 Jy Below this threshold MADmap JScanam and Scanamorphos give the noisiest measurements while Unimap retrieves the closest to HPF values None of the codes appear to introduce significant distortions of the PSF e
81. nts imposed by the report schedule and the interactive nature of processing requirement by the pre processing code These artifacts are not a result of the GLS algorithm Future versions of PACS pre processing code will mitigate these artifacts for all fields The point of using MADmap is to account for signal drifts due to 1 f noise while preserving emission at all spatial scales in the final mosaic Thus the MADmap algorithm and indeed most optimal map makers assume and expect that the noise in the time streams is entirely due to 1 f variations The MADmap pipeline processing was started at the Level 1 stage The PACS bolometer timelines at this stage contain two additional sources of noise that must be mitigated before the MADmap algorithm is applied i the PACS bolometers have pixel to pixel electronic offsets in signal values These offsets must also be homogenized to a single base level for all pixels prior to combining signals across pixels ii the timelines contain spatially correlated pixel to pixel correlated drift in the signal level as a function of time The MADmap algorithm assumes that the noise is spatially i e pixel to pixel uncorrelated and will incorrectly interpret any systematic non 1 f like drifts as real signal The removal of these artifacts is done in the pre processing stage prior to map making itself Each of these pre processing steps is discussed below e pixel to pixel offset correction for most single channel
82. os IDL Tamasis 10 Unima MADMap HPF 10 Total power Jy Pixel 10 Wave number arcmin t0 Truth JPScanam Scanamorphos_IDL Tamasis Unimag PP MADMap AIRE Normalised corrected power Jy Beam tt 0 1 1 0 10 0 Wave number arcmin Figure 10 Bright background case Blue band Uncorrected top panel and beam corrected bottom panel power spectra The shaded area encompassing angular scales comparable or smaller than the beam denotes the range of scales where the simulations become unreliable P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 105 Truth JPScanam Scanamorphos_IDL Tamasis 10 nimap EM MADMap HPF 10 Total power Jy pixel 10 ar 1 0 Wave number arcmin TO Truth 10 JPScanam Scanamorphos_lQ Tamasis Unimap MADMap HPF Normalised corrected power Jy Beam 0 1 1 0 Wave number arcmin Figure 10 continued Bright background case Red band Uncorrected top panel and beam corrected bottom panel power spectra The shaded area encompassing angular scales comparable or smaller than the beam denotes the range of scales where the simulations become unreliable P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Truth JPScanam Scana
83. point any linear component of the baseline is efficiently removed but non linear components are still present in the signal To remove these the above procedure is executed 4 times by using a spline to fit the difference TODo s TOD map The number of spline knots depends on the length of the observation and it is increased at each repetition For the latest repetition it is assumed that the knots are at a distance in time corresponding to 4 scans while imposing that the number of knots is smaller than 13 After the processing now described the main components of the baseline are removed What is left is the differential drift between a single detector and its own group The quantity TODos TOD map is then computed and fitted for each bolometer with a linear relation which is then subtracted from TOD s The procedure described above is quite efficient in removing baselines This can be shown by computing TODo6s TlOD map at the end of the whole procedure see for example Figure 8 left panel for the Crab field For comparison a map of the same region generated via the simple projection of the TOD is shown in the right panel of Figure 8 e Jump detection Cosmic rays hitting the PACS instrument produce different features in the recorded signal depending on the component that is hit When a cosmic ray hits a PACS bolometer or an inter bolometer wall the signal shows a classic short timescale glitch in one or more bolometers On the other
84. r 1 f is not accurate stripes even severe are left in the final maps compromising their quality In March 2012 we started a comparison exercise of six different map making implementations for bolometer detectors These implementations were specifically conceived for Herschel PACS and or SPIRE timelines with the exception of the SANEPIC package which was initially designed to treat data from the BLAST balloon borne experiment see Section 2 3 The preliminary results of the analysis were shown at a dedicated workshop which took place at ESAC Spain on January 28 31 2013 9 The codes performance investigation continued after the workshop In this time period the workshop results were carefully validated and when necessary updated with new results This document intends both to summarize the methology used for the comparison exercise and to present the current status of the analysis results We emphasize that the developers of the codes whose performance is here under investigation have actively participated in the testing process and are all co authors of the document The report is organized as follows In Section 2 we briefly review the characteristics of each map making code In Section 3 we describe how the simulated data were generated and the selection of archival observations In Section 4 we provide details on data processing for each of the participating mapper In Section 5 we discusses the benchmarking analysis and its results F
85. reconstructed on the best effort basis using the ICC pointing reconstruction algorithm under development includes gyro drift and sub pixel correction The residual pointing reconstruction error is considered as low as 0 1 0 3 rms 3 Sky model donor image A realistic sky model image either a superior quality mea sured dataset or a simulated image The pixel size of the input sky map model sky granularity should not be larger than half the pixel size in the final projected map 3 1 2 Generation of simulated data The simulated Level 1 hybrid cubes consist of three components 1 Noise donor cube a staring calibration observation originally designed for detector low frequency noise characterization over the ISOPHOT dark field This allows the simulation of PACS instrumental noise 2 Receptor science observation a 35x35 arcmin observation taken at nominal and orthogonal scan directions with single map repetition and 20 sec scan speed This setup is considered being a representative PACS large field observation 3 Sky model 2D pink noise simulation representative for sky cirrus observations The surface brightness is adjusted to generate bright and faint cases Point sources of uni form flux distribution are superimposed over this simulated image The faint and bright background cases differ by a factor 3 in surface brightness P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 The sky mode
86. rojection supported by WCSLIB 4 7 The conversion to from ecliptic and galactic coordinates is also possible A mask for strong sources can also be define to remove crossing constraints between different datasets e SANEPS computes noise noise power spectra The data are pre processed and decom posed in the Fourier domain into uncorrelated and n correlated components using a mixing matrix of the correlated component all components common noise power spec tra and mixing coefficients are found using an expectation maximization algorithm e SANEINV inverts the noise noise power spectra by mode as needed for the full inversion made by sanePic e SANEPIC iteratively computes the optimal map using a conjugate gradient method All programs take inputs from a single ini file which describe all parameters in particular the frequency of the high pass filtering needed before being able to transform the data in the Fourier domain see Section 4 3 1 for limitations 2 4 Scanamorphos Scanamorphos is an IDL software making maps from flux and pointing calibrated time series exploiting the redundancy in the observations to compute and subtract the total low frequency noise both the thermal noise strongly correlated among detectors and the uncorrelated flicker noise The required level of redundancy is reached in PACS observations a fiducial value that is convenient to remember is 10 samples per scan pair and per FWHM 4 pi
87. ruth reference map For the reference maps only the simulated data were used As in Section 5 1 we made use of version 3 of the reprojected hybrid simulations both for the faint and bright case more details in Section 3 1 Prior to taking the difference each map both reprocessed and reference ones was trimmed at the edges to avoid possible artifacts To evaluate the result three quantities were considered 1 pixel to pixel scatter plots of S Struc VS Strue where S is the flux in a given pixel 2 slopes of the scatter plots 3 slope corrected standard deviations hereafter stdev of S Sirue The results are illustrated in Figure 11 12 and 13 and in Table 3 The slopes in the differ ence map scatter plots indicate that there is a gain in the flux calibration which is corrected differently by the map makers In particular e JScanam its scatter plot is the flattest in all cases faint bright red blue e MADmap presents a significant slope for the faint background case both in the blue and red band e Scanamorphos shows a slight slope which changes from a correlation red to an anti correlation blue bahaviour The slope gain corrected stdev of S Strue gives the high frequency pixel to pixel noise In particular e faint background red blue the highest stdev is for MADmap Scanamorphos and Unimap show the lowest pixel to pixel noise Trends are similar for the red and blue bands e
88. s is found e Difference matrix some of the mappers appear to introduce a gain in flux calibra tion especially MADmap for faint background data both in the blue red bands On the contrary J5canam reconstructed maps show no sign of artificial gains MADmap also appears to have the highest high frequency i e pixel to pixel noise again in the faint background case while Scanamorphos and Unimap have the lowest For bright background data Tamasis seems to be characterized by the highest high frequency noise while Scanamorphos Unimap and JScanam present equally low noise values e Point source photometry bright sources 0 3 50 Jy all mappers provide accurate photometric measurements in this flux range In particular when the outliers are excluded from a statistical evaluation of the results they all retrieve fluxes within 596 1096 from the reference values obtained from HPF maps Moreover when the recommended aperture radii 6 and 10 for the blue red band respectively are used and the outliers P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Page 71 jpscan madmap 100 200 300 100 200 300 100 200 300 sanepic scanamorphos 100 200 JOU 100 200 300 tamasis unimap Figure 25 2D spectrum of the 70 um Atlas field processed with the different codes The DC has been zeroed PACS Herschel jpscan 50 100 150 200 sanepic 50 100 150 200 tamasis Docume
89. st of these limitations includes e uncertainties in beam convolutions e uncertainties in color corrections e uncertanties associated with wavelength scaling e uncertainties associated with flux non linearity effects if applicable P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 S HPF S test S HPF 0 001 0 01 0 1 S 160 PEP HPF Jy Figure 19 Comparison of Unimap based 160 um fluxes to HPF based PEP fluxes 5 4 1 Comparison with IRAS data For the purpose of comparing PACS and IRAS data instead of the original ISSA plates the Improved Reprocessing of the IRAS Survey IRIS 15 data were used The advantage of the IRIS over the IRAS data is that their absolute zero level is estimated from DIRBE measurements yet at no cost of decrease in angular resolution In addition they provide an improved zodiacal light subtraction and destriping of the data The comparison was performed at 100 um where the IRIS angular resolution is 4 3 For this metric only one data set was suitable from the list in Table 1 namely M31 i e a nearby galaxies which is fully resolved by PACS and sufficiently extended for IRAS to provide enough pixels to make the comparison meaningful The remaining fields were rejected either because the emission is too faint e g Atlas or there are insufficient pixels at IRAS resolution This test was carried out using the same approach discussed in PICC NHSC TN 029 21
90. strophysical sources produce spikes rapidly changing signal atop the drift signal as the source moves in and out of the field of view of the array during a scan In PACS data processing the correlated signal drifts are modeled as low order polynomial This requires separating the drift component from any strong astrophysical sources that are also present in the signal and further requires user interaction to select appropriate polynomial model Timelines that are not properly fit by a single polynomial are sub divided segmented in smaller groups to ensure an accurate determination of the baseline Figure 6 shows two examples of the correlated drift removal with two significantly different outcomes The left two panels in Figure 6 show a 2nd order polynomial fit to baseline This case is presented to illustrate the difficulty in baseline fitting and the need for user interaction We do not expect that users are so obviously careless in their data processing In this case the baseline is the minimum value of bins with user defined bin widths The minimum value rejects the signal from the astrophysical source The 2nd order polynomial however provides a poor fit to the baseline and consequently the resulting map bottom left hand panel shows systematic variations in the final image The two right hand side panels in Figure 6 show the results when careful segmented mitigation is applied to the data The resulting error is dependent o
91. summary denoting with cc the color correction as in the previous section and with le the wavelength scaling and using the formalism introduced in PICC ME TN 038 we have roya OO A MAPPACS 155 9um map p Acs 160um CCPACS X le 160um 155 9um MAPPACS 71 4um MAPPACS 70um CCPACS X le 70uma71 44m 9 and MAP MIPS 155 9um MAP MIPS 155 9um CCmrps 10 MAP MIPS Ti Aum MAPMIPS 71 4um CCMIPS 11 P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Convolution the high angular resolution PACS data were convolved to the lower MIPS angu lar resolution using a gaussian kernel with an effective Full Width Half Maximum FWHM 4 given by FWHM yFWHMipgs FWHM pos 12 where we adopted FWHM y rpszo 19 FWHM vrpsigo 40 and FWHM p 105870 6 FWHM pacs7o 11 This operation was done using the IDL routines PSF GAUSSIAN and CONVOL Alignining and rebinning all the PACS and MIPS maps were rebinned onto a common grid Non common pixels are set to NaN values The reprojection is performed with the IDL routine HASTROM Scatter plots and fitting the scatter plots were generated by plotting the pixel distribution of one data set e g PACS with respect to the corresponding pixel distribution for the other instrument e g MIPS The pixel to pixe distributions so obtained were fitted with a linear function of the form Mappacs AX mapyrps B 13 where A and B represent r
92. t we are working in S out space rather than in S in as often is done This is because in the real world the quantity S in is not known and one has to stick to S out only Therefore these simulations provide noise curves to be applied directly to observed quantities without the need of further assumptions For faint fluxes Starfinder produces a boosted median flux with respect to the input fluxes This is a known effect and it is due to the fact that very faint objects can be extracted with fluxes significantly higher that their true injected fluxes and thus be kept with S out larger than the 3 c noise threshold even if in principle they were not supposed to be detected because too faint This effect can happen in the real world as well and can only be seen in simulations that include very faint artificial sources as in the present case Simulations limited to 3c fluxes do not show this effect When needed possible trends of S out S in S out as a function of S out were corrected on the basis of the simulations themselves See the PEP global data release documentation for more information about this issue The in out flux comparison is split into flux bins and then the median o s i q r semi interquartile range and M A D median absolute deviation of the source distribution in this plane are computed for each bin o is heavily affected by flux boosted sources while s i q r and M A D are better representantio
93. ted with brighter fluxes due to noise flux boosting P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 5 3 5 Extraction of sources Source extraction was performed with Starfinder using the same configuration scheme adopted in the official PEP blind source extraction A PSF cut at 6 pixels was used T his conservative cut is necessary to avoid the outer part of the PSF which can be rather noisy Fluxes were then corrected to the 15 pixels radius adopted as input using the c o g of each specific PSF Furthermore fluxes were aperture corrected using the Enclosed Energy Function EEF described in PICC ME TN 037 see Table 15 19 Although this might not be an optimal choice because the Mars and Vesta observations used in that case were reduced via HPF while here we are using different mappers it likely represents a good approximation Extracted sources fluxes were then matched to the catalogs of input artificial sources by means of a simple closest neighbor algorithm using a 2 0 maximum matching radius at 100 um and 3 0 at 160 um The comparison between input and output fluxes shows an agreement of the order of 5 1096 for all the mappers with the only exception of SANEPIC 5 3 6 Statistics of the output vs input fluxes The input and output fluxes are compared by plotting the quantity S out S in S out 4 vs S out for each mapper at each band It is worth to emphasize tha
94. the median of TOD s TODmap for the detectors group O after baseline removal Right panel The blue band map of data after baseline removal P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Field M81 blue band obsid 1342186086 scan 56 line 64 1500 1000 500 Mjyisr c 500 1000 1500 B800 600 400 200 0 200 400 600 Frame Figure 9 The blue points show the median of TO Dobs TO Dmap over a line of 16 bolometers during a jump while the red line shows a fit with an exponential law 4 6 Unimap processing All the processing was carried out either on a laptop with a 8 GB RAM or on a desktop with a 12 GB RAM The desktop is really required only for the largest PACS blue tile the Atlas field while all other tiles including Atlas red can be reduced on the laptop The reduction time varies from a few minutes to several hours depending on the observation The maps discussed in this report were generated using Unimap version 5 3 or earlier In particular the maps from the real PACS observations were delivered in August September 2012 while the processing of the simulated data was completed in January 2013 Unimap comes with a default set of parameters values The processing approach was to firstly use the default parameters and inspect the results If required additional iterations were carried out in order to improve the quality This process is simplified by th
95. tter plots and fitting the scatter plots were generated by plotting the pixel distribution of one data set e g PACS with respect to the corresponding pixel distribution for the other P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 Field Scanamorphos JScanam Unimap Tamasis MADmap SANEPIC M31 1 07 1 02 1 03 1 02 1 28 0 99 Table 6 Gain factors derived by comparing the PACS M31 data set reprocessed by the various map making codes with the IRIS data instrument e g IRIS The distributions were binned and the mean value computed for every bin The binned distributions were fitted with a linear function of the form Mappacs AX MAaPrris B 7 where A represents the slope or gain factor and B the offset Noteworthy for a given data set the offset will vary from code to code since PACS did not perform absolute measurements over the course of the mission so the zero level retrieved by the mappers is arbitrary The results are shown in Figure 20 and Table 6 There is generally a good agreement with IRIS fluxes for all the mappers except for MADmap We believe the large gain factor is the result of a sloping background from incorrect drift removal Beam effect the major caveat of this comparison is the significant difference in angular resolution between PACS and IRIS A priori both source confusion and beam convolution provide plenty of pitfalls This problem was tackled by attempting to g
96. xel Its capabilities In a naive map the data are simply re mapped onto the image plane with no additional data processing on the timelines P ACS Document PACS mapmaking Date November 1st 2013 Herschel Version 1 0 also include the detection and masking of glitches and of brightness discontinuities caused by either glitches or instabilities in the multiplexing circuit low level interference patterns sometimes affecting PACS data are not handled The algorithm is described accompanied by simulations and illustrations in 24 and 25 The repository of the software and up to date documentation is http www2 iap fr users roussel herschel The output consists of a FITS cube of which the third dimension is the plane index The first plane is the signal map then comes the error map statistical error on the weighted mean the subtracted drifts map and the weight map The weight of each sample is the inverse square high frequency noise one value for each bolometer and each scan Whenever present the fifth plane is the clean signal map where the mean signal from each scan has been weighted by its inverse variance it is provided to ease the detection of remaining artefacts in the map by comparison with the first plane not for scientific purposes 2 5 Tamasis Tamasis stands for Tools for Advanced Map making Analysis and SImulations of Submillimeter surveys It is an ASTRONET funded project aiming at providing reliable

PACS Map-making Tools: Analysis and - Herschel

Contents

Download Pdf Manuals

Related Search

Related Contents