Diapositiva 1

1
MHW2 a Blind and Non-blind approach to detect
point sources in WMAP 5-year
M. López-Caniego1 and M. Massardi2 1 Cavendish
Laboratory, University of Cambridge, UK 2 SISSA,
Trieste, Italy In collaboration with J.
González-Nuevo, SISSA, Italy D. Herranz, IFCA,
Santander, Spain G. De Zotti, INAF-OAP, Italy
J. L. Sanz, IFCA, Santander, Spain
2
INTRODUCTION

• The detection of extragalactic point sources in
  maps of the Cosmic Microwave
• Background is very important
• Our knowledge of the populations of galaxies at
  frequencies gt5 GHz is
• very poor. In particular, in the range of
  frequencies observed by Planck.
• The resolved and unresolved point sources
  contaminate the data and we
• need to do something with them (mask,
  substract,calibrate the effect etc).
• In 2008 Planck will be launched, this will
  yield a wealth of information about EPS,
• but untill then we have to work with the best
  all-sky available data, WMAP.
• Several methods to detect point sources have
  been proposed
• .- Bayesian methods (PowellSnakes)
• .- Linear filters (Matched Filters, Mexican Hat
  Wavelets 1 2, Scale Adaptive)
• .- Combination of Linear Filters Bayesian
  Detectors (MF Neyman Pearson)
• .- Internal Linear Combination

M. López-Caniego Cavendish Laboratory
University of Cambridge
3
Detection of Point Sources in CMB Maps

• The detection of PS in CMB maps is, in general, a
  two step process
• Pre-processing of the data (filtering, ILC, etc).
• Detection of the objects (Thresholding, Bayesian
  detector, etc.)
• In the framework of PS detection in CMB all-sky
  maps, the detection process,
• in general, is done in a blind way, looking for
  maxima above a given threshold.
• Working in this way, important information
  already known is not used at all
• the positions of hundreds of thousends of
  objects observed at low frequency.
• We have been exploring a non-blind approach, as
  well as a combination of
• both, blind and non-blind.

M. López-Caniego Cavendish Laboratory
University of Cambridge
4
Two complementary approaches BLIND or NON-BLIND
What are the advantages/disadvantanges of a BLIND
and NON-BLIND search

• NON-BLIND
• - The position of the object is known, and only
  the flux must be estimated.
• - Better estimation of the noise in the vicinity
  of the source
• - Catalogues produced in this way may be easily
  be incomplete, depending
• on the selection criterion used to construct
  them
• surveys at other frequencies (mostly low
  frequency).
• different angular resolution.
• variability, observation at different epochs
• Low freq surveys may miss strong inverted
  sources.
• High resolution experiments may miss extended
  sources, etc.
• BLIND
• 2 additional parameters for the source position
  must be found, and below a certain flux level the
  number false objects that appear as point sources
  increases rapidly, affecting the reliability.

M. López-Caniego Cavendish Laboratory
University of Cambridge
5
Some recent BLIND works detecting Point Sources
in WMAP

• WMAP Team has produced 3 catalogues of point
  sources
• (WMAP1) 1 year data detected 208 sources
  (Bennett C. L., et al., 2003, ApJS, 148, 97).
• (WMAP3) 3 years data detected 323 sources
  (Hinshaw et al. ApJS 170, 228-334, 2007).
• (WMAP5)5 years data detected 390 sources (Wright
  et al. 2008 ApJS submited).
• Nie Zhang 2007 clean the 1st year maps and
  look for sources in the residual maps. 101
  sources, 26 (25) of them new to WMAP1 (WMAP3).
• Chen Wright 2008 combine the 61 and 94 maps to
  cancel the CMB, detected 31 (64 in 3 yrs data)
  sources, 21 of them new to WMAP3.
• Wright 2008 using the same method as before,
  detected 99 sources outside bgt10, 64 in WMAP5,
  17 new, 17 Galactic and 1 unidentified object.

M. López-Caniego Cavendish Laboratory
University of Cambridge
6
Our NON-BLIND approach in the WMAP 3 year data
Recent NON-BLIND works detecting Point Sources in
WMAP Lopez-Caniego et al. 2007 NEWPS catalogue
using WMAP 3 year data
M. López-Caniego Cavendish Laboratory
University of Cambridge
7
Our NON-BLIND approach in the WMAP 3 year data
Recent NON-BLIND works detecting Point Sources in
WMAP Lopez-Caniego et al. 2007 NEWPS catalogue
using WMAP 3 year data

• Non-blind positions where to look for point
  sources defined in a 5 GHz catalogue,
  1,980,491 sources, careful study of 4050 objects
  with flux gt 500 mJy at 5 GHz.

M. López-Caniego Cavendish Laboratory
University of Cambridge
8
Our NON-BLIND approach in the WMAP 3 year data
Recent NON-BLIND works detecting Point Sources in
WMAP Lopez-Caniego et al. 2007 NEWPS catalogue
using WMAP 3 year data

• Non-blind positions where to look for point
  sources defined in a 5 GHz catalogue,
  1,980,491 sources, careful study of 4050 objects
  with flux gt 500 mJy at 5 GHz.
• Local approach filters that are specific for
  the statistical properties of the
  neighbourhood of the sources

M. López-Caniego Cavendish Laboratory
University of Cambridge
9
Our NON-BLIND approach in the WMAP 3 year data

Filter with MHW2

M. López-Caniego Cavendish Laboratory
University of Cambridge
10
Our NON-BLIND approach in the WMAP 3 year data
Recent NON-BLIND works detecting Point Sources in
WMAP Lopez-Caniego et al. 2007 NEWPS catalogue
using WMAP 3 year data

• Non-blind positions where to look for point
  sources defined in a 5 GHz catalogue,
  1,980,491 sources, study 4050 objects with
  flux gt 500 mJy at 5 GHz.
• Local approach filters that are specific for
  the statistical properties of the
  neighbourhood of the sources
• Better beam modelling symmetrized radial beam
  profiles instead of
  idealized Gaussian beams

M. López-Caniego Cavendish Laboratory
University of Cambridge
11
Our NON-BLIND approach in the WMAP 3 year data
WMAP 3rd year symmetrized radial beam profiles
http//lambda.gsfc.nasa.gov/product/map/current/m_
images.cfm
M. López-Caniego Cavendish Laboratory
University of Cambridge
12
Our NON-BLIND approach in the WMAP 3 year data
Recent NON-BLIND works detecting Point Sources in
WMAP Lopez-Caniego et al. 2007 NEWPS catalogue
using WMAP 3 year data

• Non-blind positions where to look for point
  sources defined in a 5 GHz catalogue,
  1,980,491 sources, study 4050 objects with
  flux gt 500 mJy at 5 GHz.
• Local approach filters that are specific for
  the statistical properties of the
  neighbourhood of the sources
• Better beam modelling symmetrized radial beam
  profiles instead of idealized
  Gaussian beams
• Selection criterion in order to obtain
  complete sample of extragalactic
  point sources down to a fixed flux detection
  threshold

M. López-Caniego Cavendish Laboratory
University of Cambridge
13
NEWPS 3yr Non-Blind Local filtering and noise
estimation
For each position in the initial catalog A
flat patch centered at such position is
obtained Each patch is filtered with the MHW2
Determination of the optimal scale maximum
amplification minimum variance Local noise
estimation in a corona around the center of the
patch Look for peaks above a certain SNR
Estimate the flux density of the source
M. López-Caniego Cavendish Laboratory
University of Cambridge
14
The NEWPS 3yr source catalogue Summary of the
results
WMAP 3yr 323 obj Catalogue
(4050 Obj.) Initial Catalogue
25 Missed Objs.
297 Common Objs.
If, for example, we look at the 23 GHz
5s SubCatalogue (SC)
WMAP
80 new 5s detected objects from IC in WMAP
11 5s dets. out 25 14 non det. 5s Objs.
258 Common 5s dets. 39 non detected 5s
Total 349 (8025811) 5s at 23 GHz
M. López-Caniego Cavendish Laboratory
University of Cambridge
15
The NEWPS 3yr source catalogue Summary of the
results

• Galactic cut b lt 5 deg LMC (5 deg).
• Bright Galactic objects outside the previously
  introduced cut are removed.

Why? Sources in the WMAP catalogues are not 5s
at all frequencies 5s in at least one and at
least gt2s in the other NEWPS5s all sources are
5s
98 NEW 5s SOURCES
NEWPS-3yr López-Caniego et al. ApJS 170,108-125,
2007
M. López-Caniego Cavendish Laboratory
University of Cambridge
16
One Step further NEWPS 5yr Blind Non-Blind
Approach
Combine the power of both techniques, the Blind
and Non-Blind approach. Simple Blind Detection
(SB) Non-Blind Detection (NB)
Divide the sky into flat patches Filter with the
MHW2 Detect objetcs SNR gt 5 and estimate S Make
histogram of the fluxes of sources Flag area
around 5 brightest sources Flag a border around
the patch Estimate noise from unflagged pixels
Using a list of known positions For every
object, a patch centered in this posiiton is
obtained Filter with the MHW2 Detect objetcs SNR
gt 5 and estimate S Make histogram of the fluxes
of sources Flag area around 5 brightest
sources Flag a border around the patch Estimate
noise from unflagged pixels ONLY in a CORONA
around the object
Combined Blind (CB) First a SB and Second a NB
in the Positions of the SB
M. López-Caniego Cavendish Laboratory
University of Cambridge
17
The Bright Source Sample Benchmark for testing
the Blind Non-Blind Approach
We have used the BSS (Massardi et al. 2008) to
study the flux estimation, completeness and
reliability of our blindnon-blind approach, in
WMAP 5 year data in the same area covered by
AT20G. The BSS Bright Source Sample is a
subsample of 320 objects observed by
Australian Telescope 20 GHz Survey (AT20G),
294 of them with bgt5 and dlt-15, (194 Sgt1
Jy). It is a complete sample down to S200.5
Jy, and the observations are contemporary to
WMAP. Results from the comparison at 23 GHz
channel with SNR gt 5
M. López-Caniego Cavendish Laboratory
University of Cambridge
18
The Bright Source Sample Flux density estimation
Error in the estimation of the flux good
agreement above aprox. 840mJy
Systematic overestimation of the fluxes in WMAP
data below 840 mJy.
840 mJy
M. López-Caniego Cavendish Laboratory
University of Cambridge
19
The Bright Source Sample Reliability
Simple Blind 41 new ps not in BSS 15
extragalactic 9 Galactic 17
no counterpart (14 blt20 deg) We have produced
a noise map at 23 GHz corresponding to HEALPix
nside 32 using the noise estimations of the
individual sources in the SB.
M. López-Caniego Cavendish Laboratory
University of Cambridge
20
The Bright Source Sample Reliability
Simple Blind 41 new ps not in BSS 15
extragalactic 9 Galactic 17
no counterpart (14 blt20 deg) We have produced
a noise map at 23 GHz corresponding to HEALPix
nside 32 using the noise estimations of the
individual sources in the SB. A cut of b gt 10
deg will remove most of the contaminated pixels,
but some valuable regions as well. The usual
bgt5 deg cut is a better compromise between
removing contaminated regions and preserving
clean ones.
We can make a nosie mask, removing the blt5
deg regions. The median of the remaining pixels
outside this cut is ltsgt 169 mJy.
M. López-Caniego Cavendish Laboratory
University of Cambridge
21
The Bright Source Sample Reliability
Dropping areas with sgt 1.5 smedian 253 mJy at
23 GHz removes 17 Objects, 9 of which where
doubtful. If the remaining 8 objects where
actually spurious, the sample reliability will
be 95.5.
M. López-Caniego Cavendish Laboratory
University of Cambridge
22
The Bright Source Sample Reliability
Threshold at 1.5ltsgt 253 mJy
M. López-Caniego Cavendish Laboratory
University of Cambridge
23
The Bright Source Sample Completeness
The final sample is almost 100 complete over
the unmasked area covered by the BSS gt 2 Jy
(because one of the BSS objects has SNR lt 5
). Considering only SB detections, the
completeness is 89 above 1 Jy. Considering the
combination of the SBNB, increases to
91. Taking these results into account, we
propose a recipe to do the PS Detection in the
full-sky maps of WMAP 5yr.
M. López-Caniego Cavendish Laboratory
University of Cambridge
24
Blind and Non-Blind detection in WMAP 5-year
data Recipe

• The analysis of the BSS provides a recipe to
  study the WMAP 5yr and to
• update NEWPS-3yr to NEWPS-5yr catalogue
• Carry out a Simple Blind detection over the full
  sky maps
• Generate noise maps of the mean noise values per
  pixel for nside32 (spixel) and compute the
  median excluding the blt5 deg (smedian).
• Generate a mask with all the pixels with noise
  level gt 1.5 smedian
• Consider all the sources outside the mask and SNR
  gt 5 as true detections.
• Carry out a Non-Blind search on the maps at the
  positions of the
• NEWPS-3yr-3s sample.
• 6. Combine results from both methods

M. López-Caniego Cavendish Laboratory
University of Cambridge
25
Summary of the results for the three methods
M. López-Caniego Cavendish Laboratory
University of Cambridge
26
Summary of the properties of the NEWPS-5yr 5s
catalogue
Areas with spixel gt 1.5 smedian have been
excluded.
516 different sources 480 EPS, 20 GS, 16
unidentified - at least 96.9 reliable In the
region covered by NEWPS, WMAP has 388 PS 356
SNR gt5 and 32 SNR lt 5 NEWPS-5yr has 164 new
sources not in WMAP-5yr from them, only 14 may
be false We recover the 64 objects of Chen
Wright (08) and the objects by Nie Zhang (07)
M. López-Caniego Cavendish Laboratory
University of Cambridge
27
NEWPS-5yr Number Counts
M. López-Caniego Cavendish Laboratory
University of Cambridge
28
Conclusions
We have studied the efficiency in source
detection and flux density estimation of a
BlindNon-Blind approach based on the MHW2
applied on WMAP 5yr. Using the ATC20G Bright
Source Sample (Massardi et al. 2008) we have
been able to estimate the completeness,
reliability and accuracy of the flux
density Determination for the samples obtained
with the blind and the non-blind approach. The
flux density are unbiased down to 840 mJy where
we overestimate the Fluxes (a manfestation of the
Eddington bias). This will affect the number
counts, but it can be corrected using high
resolution surveys that go below WMAP detection
limit, AT20G (Massardi 08, Ricci 04) and 9C
(Waldram, 03). At higher flux density, most
candidates to be spurious are low Galactic blt20
deg. Excluding objects in contaminated areas (s gt
1.5 smedian outside bgt5) the number of dubious
objects reduces by half. If all of them are
spurious, reliability 95.5
M. López-Caniego Cavendish Laboratory
University of Cambridge
29
Conclusions
In the full-sky study Blind finds 488
sources (bgt5, noise gt 1.5 smedian) Non-Blind
finds 28 additinal sources Total 516
(4802016 )sources with SNR gt 5 compared with
the 388 objects in the WMAP-5yr
catalogue. From the 516, since only 16 of them
are dubious, then the reliability at least
96.9.
M. López-Caniego Cavendish Laboratory
University of Cambridge