Identifying Optical GW Counterparts with Image Processing Noemi - - PowerPoint PPT Presentation

▶
identifying optical gw counterparts with image processing
SMART_READER_LITE
LIVE PREVIEW

Identifying Optical GW Counterparts with Image Processing Noemi - - PowerPoint PPT Presentation

Nov 04, 2022 366 likes •1.02k views

Identifying Optical GW Counterparts with Image Processing Noemi Glaeser, University of South Carolina SIST Final Talk 6 August 2018 Overview Background GW DES The DES-GW pipeline SE DiffImg Next steps 2 06/08/18

slide-1
SLIDE 1

Noemi Glaeser, University of South Carolina SIST Final Talk 6 August 2018

Identifying Optical GW Counterparts with Image Processing

slide-2
SLIDE 2
  • Background

– GW – DES

  • The DES-GW pipeline

– SE – DiffImg

  • Next steps

Overview

06/08/18

  • N. Glaeser | SIST Final Talk

2

slide-3
SLIDE 3

LIGO/T. Pyle 2016

“Perpetuating changes in the gravitational field caused by accelerating masses”

3

Gravitational Waves

06/08/18

  • N. Glaeser | SIST Final Talk
slide-4
SLIDE 4

LIGO/T. Pyle 2016

Binary Black Hole (BBH) Black Hole-Neutron Star (BH-NS) Binary Neutron Star (BNS)

4

Gravitational Waves

06/08/18

  • N. Glaeser | SIST Final Talk
slide-5
SLIDE 5

LIGO/Caltech/MIT/LSC 2017

Wave shape → event type Amplitude → distance

5

Gravitational Waves

06/08/18

  • N. Glaeser | SIST Final Talk
slide-6
SLIDE 6

6

Gravitational Waves

LIGO/Virgo/NASA/Leo Singer 2017 (Milky Way image: Axel Mellinger) Dark Energy Survey Collaboration 06/08/18

  • N. Glaeser | SIST Final Talk
slide-7
SLIDE 7

7

Dark Energy Survey (DES)

Reidar Hahn (Fermilab) DES Collaboration DES Collaboration

DECam, assembled at Fermilab, now mounted on the Blanco telescope in Chile

06/08/18

  • N. Glaeser | SIST Final Talk
slide-8
SLIDE 8

The DES footprint spans ~5000 square degrees (⅛ of the sky)

8

Dark Energy Survey (DES)

DES Collaboration 06/08/18

  • N. Glaeser | SIST Final Talk
slide-9
SLIDE 9

Independent measure of the Hubble parameter and other cosmological parameters

9

Dark Energy Survey (DES): Why?

DES Collaboration 06/08/18

  • N. Glaeser | SIST Final Talk
slide-10
SLIDE 10

10

Dark Energy Survey (DES): GW170817

GW170817 Optical Counterpart composite detection images (Soares-Santos et al. 2017)

  • The first optical counterpart of a GW event was observed by DES in August 2017!

– One point on the distance-redshift plot

  • In O3 we expect ~10x as many events

06/08/18

  • N. Glaeser | SIST Final Talk
slide-11
SLIDE 11

Independent measure of the Hubble parameter and other cosmological parameters

11

Dark Energy Survey (DES): Why?

DES Collaboration 06/08/18

  • N. Glaeser | SIST Final Talk
slide-12
SLIDE 12

DES-GW Image Processing Pipeline

12 06/08/18

  • N. Glaeser | SIST Final Talk
slide-13
SLIDE 13

Full Pipeline

13

DES-GW Pipeline

06/08/18

  • N. Glaeser | SIST Final Talk
slide-14
SLIDE 14

Full Pipeline

  • Single Epoch (SE)

processing

  • Intermediate

steps

  • DiffImg

14

DES-GW Pipeline

06/08/18

  • N. Glaeser | SIST Final Talk
slide-15
SLIDE 15

Full Pipeline

  • Single Epoch (SE)

processing

  • Intermediate

steps

  • DiffImg

15

DES-GW Pipeline

06/08/18

  • N. Glaeser | SIST Final Talk
slide-16
SLIDE 16

Full Pipeline

  • Single Epoch (SE)

processing

  • Intermediate

steps

  • DiffImg

16

DES-GW Pipeline

06/08/18

  • N. Glaeser | SIST Final Talk
slide-17
SLIDE 17

Full Pipeline

  • Single Epoch (SE)

processing

  • Intermediate

steps

  • DiffImg

Runtime: ~5-8 hr

17

DES-GW Pipeline

06/08/18

  • N. Glaeser | SIST Final Talk
slide-18
SLIDE 18

Full Pipeline

  • Single Epoch (SE)

processing

  • Intermediate

steps

  • DiffImg

Runtime: ~5-8 hr Goal: <1 hr

18

DES-GW Pipeline

06/08/18

  • N. Glaeser | SIST Final Talk
slide-19
SLIDE 19

Full Pipeline

  • Single Epoch (SE)

processing

  • Intermediate

steps

  • DiffImg

Runtime: ~5-8 hr Goal: <1 hr

19

DES-GW Pipeline

06/08/18

  • N. Glaeser | SIST Final Talk
slide-20
SLIDE 20

Single Epoch Processing

  • Image Correction

– Raw Images → “science-ready”

  • Astrometric calibration:

2MASS

  • Object Cataloguing
  • Photometric Calibration:

2MASS

20

Image Processing Pipeline: SE processing

06/08/18

  • N. Glaeser | SIST Final Talk
slide-21
SLIDE 21

Single Epoch Processing

  • Image Correction

– Raw Images → “science-ready”

  • Astrometric calibration:

2MASS

  • Object Cataloguing
  • Photometric Calibration:

2MASS

21

Image Processing Pipeline: SE processing

06/08/18

  • N. Glaeser | SIST Final Talk
slide-22
SLIDE 22

Single Epoch Processing

  • Image Correction

– Raw Images → “science-ready”

  • Astrometric calibration:

2MASS

  • Object Cataloguing
  • Photometric Calibration:

2MASS

22

Image Processing Pipeline: SE processing

06/08/18

  • N. Glaeser | SIST Final Talk
slide-23
SLIDE 23

Single Epoch Processing

  • Image Correction

– Raw Images → “science-ready”

  • Astrometric calibration:

2MASS

  • Object Cataloguing
  • Photometric Calibration:

2MASS

23

Image Processing Pipeline: SE processing

06/08/18

  • N. Glaeser | SIST Final Talk
slide-24
SLIDE 24

Single Epoch Processing

  • Image Correction

– Raw Images → “science-ready”

  • Astrometric calibration:

2MASS

  • Object Cataloguing
  • Photometric Calibration:

2MASS

24

Image Processing Pipeline: SE processing

06/08/18

  • N. Glaeser | SIST Final Talk
slide-25
SLIDE 25

Single Epoch Processing

  • Image Correction

– Raw Images → “science-ready”

  • Astrometric calibration:

2MASS

  • Object Cataloguing
  • Photometric Calibration:

2MASS

25

Image Processing Pipeline: SE processing

06/08/18

  • N. Glaeser | SIST Final Talk
slide-26
SLIDE 26

Single Epoch Processing

  • Image Correction

– Raw Images → “science-ready”

  • Astrometric calibration:

2MASS

– CCD-by-CCD

  • Object Cataloguing
  • Photometric Calibration:

2MASS

– CCD-by-CCD

26

Image Processing Pipeline: SE processing

06/08/18

  • N. Glaeser | SIST Final Talk
slide-27
SLIDE 27

Single Epoch Processing

  • Image Correction

– Raw Images → “science-ready”

  • Astrometric calibration:

2MASS

– CCD-by-CCD

  • Object Cataloguing
  • Photometric Calibration:

2MASS

– CCD-by-CCD

27

Image Processing Pipeline: SE processing

06/08/18

  • N. Glaeser | SIST Final Talk
slide-28
SLIDE 28

Single Epoch Processing

  • Image Correction

– Raw Images → “science-ready”

  • Astrometric calibration:

GAIA-DR2

– CCD-by-CCD

  • Object Cataloguing
  • Photometric Calibration:

GAIA-DR2

– CCD-by-CCD

28

Image Processing Pipeline: SE processing

06/08/18

  • N. Glaeser | SIST Final Talk
slide-29
SLIDE 29

Single Epoch Processing

  • Image Correction

– Raw Images → “science-ready”

  • Astrometric calibration:

GAIA-DR2

– CCD-by-CCD

  • Object Cataloguing
  • Photometric Calibration:

GAIA-DR2

– CCD-by-CCD

29

Image Processing Pipeline: SE processing

06/08/18

  • N. Glaeser | SIST Final Talk
slide-30
SLIDE 30
  • “Fast” CCDs don’t have to

wait for slower ones

30

Image Processing Pipeline: SE Modifications

Photometric calibration Raw image CCD62 CCD1 CCD2 ...

06/08/18

  • N. Glaeser | SIST Final Talk
slide-31
SLIDE 31
  • “Fast” CCDs don’t have to

wait for slower ones

31

Image Processing Pipeline: SE Modifications

Raw image CCD1 CCD2 ... CCD62 Photometric calibration

06/08/18

  • N. Glaeser | SIST Final Talk
slide-32
SLIDE 32
  • “Fast” CCDs don’t have to

wait for slower ones

32

Image Processing Pipeline: SE Modifications

Raw image CCD2 ... CCD62 Photometric calibration CCD1

06/08/18

  • N. Glaeser | SIST Final Talk
slide-33
SLIDE 33
  • “Fast” CCDs don’t have to

wait for slower ones

33

Image Processing Pipeline: SE Modifications

Photometric calibration CCD62 CCD1 CCD2 ...

06/08/18

  • N. Glaeser | SIST Final Talk
slide-34
SLIDE 34
  • “Fast” CCDs don’t have to

wait for slower ones

34

Image Processing Pipeline: SE Modifications

Photometric calibration Science-ready image

06/08/18

  • N. Glaeser | SIST Final Talk
slide-35
SLIDE 35
  • “Fast” CCDs don’t have to

wait for slower ones

35

Image Processing Pipeline: SE Modifications

Photometric calibration Science-ready image to DiffImg

06/08/18

  • N. Glaeser | SIST Final Talk
slide-36
SLIDE 36
  • “Fast” CCDs don’t have to

wait for slower ones

36

Image Processing Pipeline: SE Modifications

Photometric calibration Raw image CCD62 CCD1 CCD2 ...

06/08/18

  • N. Glaeser | SIST Final Talk
slide-37
SLIDE 37
  • “Fast” CCDs don’t have to

wait for slower ones

37

Image Processing Pipeline: SE Modifications

Photometric calibration Raw image CCD62 CCD1 CCD2 ...

06/08/18

  • N. Glaeser | SIST Final Talk
slide-38
SLIDE 38
  • “Fast” CCDs don’t have to

wait for slower ones

38

Image Processing Pipeline: SE Modifications

Photometric calibration Raw image CCD62 CCD1 CCD2 ... to DiffImg

06/08/18

  • N. Glaeser | SIST Final Talk
slide-39
SLIDE 39

Single Epoch Parallelization Speedup (series: 0.9625 h; parallel: µ=0.15, median=0.13, σ=0.05)

  • “Fast” CCDs don’t have to

wait for slower ones

39

Image Processing Pipeline: SE Modifications

06/08/18

  • N. Glaeser | SIST Final Talk
slide-40
SLIDE 40

Single Epoch Parallelization Speedup (series: 0.9625 h; parallel: µ=0.15, median=0.13, σ=0.05)

  • “Fast” CCDs don’t have to

wait for slower ones

  • But…

40

Image Processing Pipeline: SE Modifications

06/08/18

  • N. Glaeser | SIST Final Talk
slide-41
SLIDE 41
  • “Fast” CCDs don’t have to

wait for slower ones

  • But…

– do the results still make sense? – did we break something down the line? – Test on GW170817

41

Image Processing Pipeline: SE Modifications

06/08/18

  • N. Glaeser | SIST Final Talk

Single Epoch Parallelization Speedup (series: 0.9625 h; parallel: µ=0.15, median=0.13, σ=0.05)

slide-42
SLIDE 42
  • Run newly SE-processed

images through DiffImg

– Make sure we still identify the counterpart

42

Image Processing Pipeline: SE Modifications

06/08/18

  • N. Glaeser | SIST Final Talk
slide-43
SLIDE 43

Non-matching candidates with old SE (all 2MASS) vs. new SE (all GAIA)

  • Run newly SE-processed

images through DiffImg

– Make sure we still identify the counterpart

  • 135 unmatched objects

(45%)!

43

Image Processing Pipeline: SE Modifications

06/08/18

  • N. Glaeser | SIST Final Talk
slide-44
SLIDE 44

Non-matching candidates with old SE (all 2MASS) vs. new SE (all GAIA)

  • Run newly SE-processed

images through DiffImg

– Make sure we still identify the counterpart

  • 135 unmatched objects

(45%)!

– 8 with ML score > 0.5

44

Image Processing Pipeline: SE Modifications

06/08/18

  • N. Glaeser | SIST Final Talk
slide-45
SLIDE 45

Non-matching candidates with old SE (all 2MASS) vs. new SE (all GAIA)

  • Run newly SE-processed

images through DiffImg

– Make sure we still identify the counterpart

  • 135 unmatched objects

(45%)!

– 8 with ML score > 0.5 – 1 with ML score > 0.7 (ML=0.96)

45

Image Processing Pipeline: SE Modifications

06/08/18

  • N. Glaeser | SIST Final Talk
slide-46
SLIDE 46

Non-matching candidates with old SE (all 2MASS) vs. new SE (all GAIA)

  • Run newly SE-processed

images through DiffImg

– Make sure we still identify the counterpart

  • Unmatched objects are

mostly junk

46

Image Processing Pipeline: SE Modifications

06/08/18

  • N. Glaeser | SIST Final Talk
slide-47
SLIDE 47
  • Run newly SE-processed

images through DiffImg

– Make sure we still identify the counterpart

  • Unmatched objects are

mostly junk

  • Almost identical difference

images

Difference images: new SE (left) vs. old SE (right)

47

Image Processing Pipeline: SE Modifications

06/08/18

  • N. Glaeser | SIST Final Talk
slide-48
SLIDE 48
  • Run newly SE-processed

images through DiffImg

– Make sure we still identify the counterpart

  • Unmatched objects are

mostly junk

  • Almost identical difference

images

– Still found the counterpart!

Difference images: new SE (left) vs. old SE (right) with counterpart

48

Image Processing Pipeline: SE Modifications

06/08/18

  • N. Glaeser | SIST Final Talk
slide-49
SLIDE 49
  • Run newly SE-processed

images through DiffImg

– Make sure we still identify the counterpart

  • Unmatched objects are

mostly junk

  • Almost identical difference

images

– ...Remember the lone unmatched object?

Difference images: new SE (left) vs. old SE (right) with unmatched object (ML=0.96)

49

Image Processing Pipeline: SE Modifications

06/08/18

  • N. Glaeser | SIST Final Talk
slide-50
SLIDE 50
  • Both search & template image calibration in parallel with

GAIA (instead of 2MASS)

– Understand the results

  • Was the 0.96 ML object an anomaly in camera pointing?
  • Why was it ultimately eliminated as a candidate?
  • Integrate SE, verifySE, and DiffImg into one script

– raw image → counterpart identification

  • Test the full DES-GW pipeline in the mock observing run

next week

Next Steps

50 06/08/18

  • N. Glaeser | SIST Final Talk
slide-51
SLIDE 51
  • Both search & template image calibration in parallel with

GAIA (instead of 2MASS)

– Understand the results

  • Was the 0.96 ML object an anomaly in camera pointing?
  • Why was it ultimately eliminated as a candidate?
  • Integrate SE, verifySE, and DiffImg into one script

– raw image → counterpart identification

  • Test the full DES-GW pipeline in the mock observing run

next week

Next Steps

51 06/08/18

  • N. Glaeser | SIST Final Talk
slide-52
SLIDE 52

Full Pipeline

52

DES-GW Pipeline

06/08/18

  • N. Glaeser | SIST Final Talk
slide-53
SLIDE 53

Full Pipeline

53

DES-GW Pipeline

06/08/18

  • N. Glaeser | SIST Final Talk
slide-54
SLIDE 54
  • Both search & template image calibration in parallel with

GAIA (instead of 2MASS)

– Understand the results

  • Was the 0.96 ML object an anomaly in camera pointing?
  • Why was it ultimately eliminated as a candidate?
  • Integrate SE, verifySE, and DiffImg into one script

– raw image → counterpart identification

  • Test the full DES-GW pipeline in a mock observing run

Next Steps

54 06/08/18

  • N. Glaeser | SIST Final Talk
slide-55
SLIDE 55
  • Parallelization: DES-GW image processing pipeline is faster

without sacrificing accuracy

– Allows us to discover counterpart ASAP, enabling detailed spectroscopic follow-up from the early stages of the kilonova

  • Integration: Full process, from image capture to counterpart

identification, is more streamlined

  • The new pipeline will be employed by DES-GW in LIGO O3

in early 2019

Conclusion

55 06/08/18

  • N. Glaeser | SIST Final Talk
slide-56
SLIDE 56
  • The Department of Energy
  • The SIST Committee & the Computing Division
  • Ken Herner, Jim Annis, Marcelle Soares-Santos, and the

remainder of the DES-GW group

Acknowledgements

56 06/08/18

  • N. Glaeser | SIST Final Talk
slide-57
SLIDE 57

57

Questions?

06/08/18

  • N. Glaeser | SIST Final Talk
slide-58
SLIDE 58

58

Backup

06/08/18

  • N. Glaeser | SIST Final Talk
slide-59
SLIDE 59
  • Run newly SE-processed

images through DiffImg

– Make sure we still identify the counterpart

  • Unmatched objects are

mostly junk

  • Similar machine learning

(ML) scores

Plot of ΔML as a function of averaged ML score: mostly clustered around 0, but many negative values for low ML scores

59

Image Processing Pipeline: SE Modifications

06/08/18

  • N. Glaeser | SIST Final Talk
slide-60
SLIDE 60
  • Run newly SE-processed

images through DiffImg

– Make sure we still identify the counterpart

  • Unmatched objects are

mostly junk

  • Similar machine learning

(ML) scores

– Low scores are typically even lower in the new processing

Plot of ΔML as a function of averaged ML score: mostly clustered around 0, but many negative values for low ML scores

60

Image Processing Pipeline: SE Modifications

06/08/18

  • N. Glaeser | SIST Final Talk
slide-61
SLIDE 61
  • Run newly SE-processed

images through DiffImg

– Make sure we still identify the counterpart

  • Unmatched objects are

mostly junk

  • Similar machine learning

(ML) scores

– Low scores are typically even lower in the new processing

Plot of ΔML as a function of averaged ML score: mostly clustered around 0, but many negative values for low ML scores

61

Image Processing Pipeline: SE Modifications

06/08/18

  • N. Glaeser | SIST Final Talk
slide-62
SLIDE 62
  • Run newly SE-processed

images through DiffImg

– Make sure we still identify the counterpart

  • Unmatched objects are

mostly junk

  • Similar machine learning

(ML) scores

– Left-skewed normal distribution

∆ML spread

62

Image Processing Pipeline: SE Modifications

06/08/18

  • N. Glaeser | SIST Final Talk
slide-63
SLIDE 63
  • Also found in the z-band

63

The Mysterious ML=0.96 Object

06/08/18

  • N. Glaeser | SIST Final Talk
slide-64
SLIDE 64

64

Kilonova Light Curve

06/08/18

  • N. Glaeser | SIST Final Talk

Detailed spectroscopic follow-up of GW170817 counterpart. This shows the importance of fast counterpart identification, which is enabled by the new

  • pipeline. (Soares-Santos et al.)
slide-65
SLIDE 65

65

Search-Temp = Diff

06/08/18

  • N. Glaeser | SIST Final Talk