Blending in LSST Data Products Jim Bosch, DM DRP Scientist / Princeton
Blending Families Two Footprints: 2 above-threshold regions with peaks. 4 One isolated object ( 1 ). 3 One Parent ( 2 ): 5 blends measured with no deblending. Three Children ( 3 , 4 , 5 ): blends measured after deblending. 1 2
Blending Families 1 2 2 this is a tree: 4 3 4 5 3 5 id parent deblend_nChild 1 0 0 represented as 2 0 3 a table: 3 2 0 4 2 0 1 5 2 0 3
Useful Subsets 1 2 2 The full table is not a 4 useful subset: 3 4 5 3 ( 3 , 4, 5 ) and 2 are 5 mutually exclusive. id parent deblend_nChild 1 0 0 2 0 3 3 2 0 4 2 0 1 5 2 0 4
Useful Subsets 1 2 2 Usually you want 4 both isolated and deblended objects: 3 4 5 3 5 deblend_nchild = 0 id parent deblend_nChild 1 0 0 2 0 3 3 2 0 4 2 0 1 5 2 0 5
Useful Subsets 1 2 2 If you're interested in 4 really bright objects (bright enough to 3 4 5 3 ignore their 5 neighbors), and you don't trust the id parent deblend_nChild deblender: 1 0 0 2 0 3 parent = 0 3 2 0 4 2 0 1 5 2 0 6
Useful Subsets 1 2 2 If you don't trust the 4 deblender, and don't mind an incomplete 3 4 5 3 sample: 5 deblend_nchild = 0 AND id parent deblend_nChild parent = 0 1 0 0 2 0 3 3 2 0 4 2 0 1 5 2 0 7
Footprints and HeavyFootprints 1 2 3 4 5 id parent deblend_nChild 1 0 0 2 0 3 3 2 0 4 2 0 5 2 0 8
Footprints and HeavyFootprints 1 2 3 4 5 id parent deblend_nChild 1 0 0 2 0 3 3 2 0 4 2 0 5 2 0 9
Footprints and HeavyFootprints 1 2 3 4 5 id parent deblend_nChild 1 0 0 2 0 3 3 2 0 4 2 0 5 2 0 10
Footprints and HeavyFootprints 1 2 3 4 5 id parent deblend_nChild 1 0 0 2 0 3 3 2 0 4 2 0 5 2 0 11
Footprints and HeavyFootprints 1 2 3 4 5 id parent deblend_nChild 1 0 0 2 0 3 3 2 0 4 2 0 5 2 0 12
Extending the Tree id parent deblend_nChild 1 2 1 0 0 2 0 3 3 2 2 3 4 5 4 2 0 5 2 0 6 3 0 6 7 7 3 0 13
Association Flags When merging detections from different bands, we set flags to indicate where the Object came from: merge_footprint_<band>: there is a Footprint in <band> that overlaps the parent Footprint. merge_peak_<band>: there is a peak in <band> near the peak that spawned a child object. 14
Deblender Flags These are algorithm-dependent, and they may look completely different in DR1 (or even a few months from now, if we've switched to Scarlet): deblend_deblendedAsPsf: Deblender thought this source looked like a PSF deblend_tooManyPeaks: source had too many peaks; only the brightest were included deblend_parentTooBig: Parent footprint covered too many pixels deblend_masked: Parent footprint was predominantly masked deblend_skipped: Deblender skipped this source deblend_hasStrayFlux: This source was assigned some stray flux 15
Blendedness We want a metric with the following properties: ● zero for isolated objects; ● approaches unity for when an object is much fainter than its neighbor(s); ● related to how much the photometry of the primary object could have been affected by its neighbors; ● can be derived from real data (doesn't require ground truth). 16
Blendedness If we knew the true child and parent profiles, we would use: 17
Blendedness For real data (and real deblends), we use: 18
Blendedness Using the deblended Child and original-data Parent is noisy. ● That's why we use the Gaussian model instead of the Child itself in some places: ● We also compute a variant that uses a de-biased absolute value of the Parent and Child: See HSC Pipeline Paper (Bosch et al 2018) for more details. 19
Summary ● The outputs of deblending are a tree, even though we flatten that tree into a table. ● The best way to learn what happened in deblending is to look at the actual results - the HeavyFootprints. ● There will be lots of deblender and association flags. But the flags we have now will probably change with the algorithms. ● Blendedness is one useful estimator for how how affected an object was by blending; there may be others. 20
Recommend
More recommend
