Download presentation
Presentation is loading. Please wait.
Published byBarnaby Norris Modified over 8 years ago
1
GLAST LAT Project DC2 Closeout Workshop, GSFC, 31 May-2 June 2006 1 Comparisons of Searches for Sources in the DC2 Data S. W. Digel Stanford Linear Accelerator Center S. Ciprini
2
GLAST LAT Project DC2 Closeout Workshop, GSFC, 31 May-2 June 2006 2 Introduction Thanks to the ‘content providers’ for being willing to go along with this DC2 exercise The DC2 data set is a large (for us) and (semi) realistic representation of the celestial sky, so of course trying out algorithms for source detection is hard to resist – but is not to be mistaken for a systematic study Systematic studies are also how the algorithms will be optimized, and this has not been done uniformly – as the presentations yesterday made clear Brief introduction to the source lists and some comparisons are in the following slides –Other investigations will be made within the Catalog group (attend the VRVS meetings) –If permission is granted, we can post the lists in Confluence – again these are all works in progress
3
GLAST LAT Project DC2 Closeout Workshop, GSFC, 31 May-2 June 2006 3 Recap of Source Detection Methods MRF2 – Multi Resolution Filter (Ballet) based on application of MR_FILTER (Starck) –Important details of its application include running MR_FILTER on separate bands and merging the results –Here will use MRF2-equ-all.txt – has merging of 4 bands and one iteration to detect fainter sources in the vicinity of brighter ones –Fluxes for many of the sources Optimal – Optimal filter (Ballet), also described yesterday –Merging of results from different bands and iteration were also applied; the result file that will be used here is Optimal-equ-all.txt –All-sky search, significances in the merged file are not well defined –Fluxes for many of the sources UW – Wavelet filtering (Burnett), described today –Works on 8 successively finer grids of HEALPix, with finer gridding used for higher energies. Detections are merged across bands –A significance is provided per band, but an overall significance is not provided; fluxes are not provided either
4
GLAST LAT Project DC2 Closeout Workshop, GSFC, 31 May-2 June 2006 4 Recap (cont) BIN (Casandjian) – basically a reimplementation of the binned likelihood analysis used for EGRET data (LIKE) –As described yesterday by Isabelle, many spurious detections at low latitude owing to an offset (0.25°) in latitude and probably also longitude between where the diffuse model thinks it is on the sky and where LIKE thinks it is –The BIN analysis is iterative and provides fluxes, counts, and significances. No confidence regions yet, and spectral indicies are fixed at -2 VR (Romeo & Cillis) – source detection by clustering of cells in Voronoi tessellation, along with a method to evaluate whether candidate sources are point-like –A work in progress; provides only candidate source position and position uncertainty
5
GLAST LAT Project DC2 Closeout Workshop, GSFC, 31 May-2 June 2006 5 Recap (cont) SB (Stephens) - Aperture Photometry – described yesterday – [SB?] –Provides sigificance and counts estimate (in 0.25 deg aperture) PGW (Tosti) – Perugia wavelet filtering – also described yesterday –Current application is for | | < 80°, although this is not an intrinsic limit and misses only 1.5% of the sky –Provides counts (2x2 deg region, >100 MeV) and significance DC2Cat (Ballet & Landriu) – DC2 source list (MRF…) – currently v2.1 –Provides flux, estimated counts, position uncertainty and significance (TS)
6
GLAST LAT Project DC2 Closeout Workshop, GSFC, 31 May-2 June 2006 6 Review of recap For SB, and PGW I’ll crudely estimate fluxes from counts based on weighted mean exposure for the position of the source –All-sky average ~>100 MeV for DC2 is 3.2 × 10 9 cm 2 s For UW, VR, Optimal, and MRF2, either fluxes or counts are not available for some or all sources Method# Sources MRF2644 Optimal560 UW1651* BIN540 VR2548 SB1463 PGW934 DC2Cat380 DC2Sky1720 * New version, not analyzed here, has ~3k sources
7
GLAST LAT Project DC2 Closeout Workshop, GSFC, 31 May-2 June 2006 7 Distribution in flux Whole sky Black – truth Yellow – DC2cat Blue – BIN Red – PGW Green – SB (fluxes likely to be seriously underestimated from counts in small aperture) Implied flux limits
8
GLAST LAT Project DC2 Closeout Workshop, GSFC, 31 May-2 June 2006 8 Distributions in latitude Density of sources vs true density of sources Black – DC2 sky model (sources with flux >1 × 10 -8 cm -2 s -1, >100 MeV) – this is the reference (683 sources)* *May exclude some very hard sources YellowDC2Cat BlueBIN RedPGW GreenSB Red dashedMRF2 Blue dashedOptimal Yellow dashedVR
9
GLAST LAT Project DC2 Closeout Workshop, GSFC, 31 May-2 June 2006 9 Comparison with the sky model Again selected only the ‘true’ sources above >1 × 10 -8 cm -2 s -1, >100 MeV)* NB: logarithmic scale For comparisons of completeness, make generous assumption that R = 1° is close enough to count as detecting a true source, and R > 1° is a false positive *Mean separation of sources at this level ~4.4° YellowDC2Cat BlueBIN RedPGW GreenSB Red dashedMRF2 Blue dashedOptimal Yellow dashedVR
10
GLAST LAT Project DC2 Closeout Workshop, GSFC, 31 May-2 June 2006 10 Comparison with true sources (cont) Preliminary, and using the crude source correspondence definition R = 1° For most methods, more ‘True’ sources are found than were in the DC2Cat, and the range of number of true detections is remarkably small (406- 443) The DC2Cat has the lowest number of spurious detections – an expected tradeoff with sensitivity Method# SourcesTrueSpurious MRF2644430214 Optimal560443117 UW16514221229 BIN540406134 VR25481292419 SB14634351028 PGW934443491 DC2Cat38033545 DC2Sky1720 Best values in these columns are in red
11
GLAST LAT Project DC2 Closeout Workshop, GSFC, 31 May-2 June 2006 11 Toward selecting an algorithm For what? –An algorithm to run as a Science Tool would be useful and would provide freedom from needing to have a pregenerated source list –And for catalog analysis – the source detection step of the pipeline –And for Automated Science Processing ( Quick Look ) –No, the optimization between spurious source rate and completeness is not the same for each of these For when? –We’ll have an informal discussion on this and other subjects in a rump meeting of the Catalog group tomorrow after the conclusion of the workshop
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.