progress report on psfs and pixels jay anderson, elena sabbi, kailash sahu, and matthew bourque tips...

Progress Report on PSFs and Pixels

Jay Anderson, Elena Sabbi, Kailash Sahu, and Matthew Bourque

TIPS Feb 19, 2015

Lots of Inter-related Issues• Why do we need a good PSF? (Or do we?)• How do we get a good PSF? • How do we use a good PSF?

• Issues involve:– Distortion solutions– Image alignment and reference frames– High-precision transformations– PSF modeling (spatial and time variability)– Forward modeling into data domain– Deconvolution– Background matching– Pixel-area corrections

ChickenandEgg

Problems

Why Do We Need a Good PSF?• Not everyone does…• High-precision science– PSF-fitting of bright stars

• Photometry to 1%• Astrometry to 0.01 pixel

– PSF-fitting of faint stars• Simultaneous photometry to optimize S/N

– Fitting barely-resolved galaxies• Weak-lensing analysis, profiles, point-source removal,

deconvolution

How Do We Get a Good PSF?• Deal with undersampled nature of detectors

– WFC3/IR: up to 40% of light in central pixel – ACS/WFC: up to 23% of light in central pixel– WFC3/UVIS: up to 18% of light in central pixel

• Deal with spatial variations– Every ~500 pixels the PSF changes appreciably– Geometric optics, Charge diffusion variation– Tiny Tim models not very good

• Deal with time variations– Focus changes… model?– Often not enough stars in an image to extract– “Library” PSFs actually quite useful

• No standard format for PSF models, no obvious tools– “effective” PSF is one approach– Purely empirical -vs- model-based– Will suggest a “standard” format

How to Use a Good PSF?• Push through astrodrizzle– Must plant stars in multiple FLTs consistently– “Fuzzy” constraints

• uneven sampling, correlated noise

• Fit to the individual FLT/FLC pixels– The only “hard”, independent constraints

• Can properly treat the errors

– Easier for bright isolated point-sources• “one-pass”

– Harder for faint or resolved objects• Must simultaneously interrelate the FLT pixels with scene• Must properly convolve model of scene to match pixels

The Impasse• Many tasks at once

– Provide PSFs– Provide access to pixel mappings– Provide tools to use

• Institute is in a unique position– Users can’t be expected to solve all these probs

• Requires astronomer-decades…

– Legacy value for HST (higher-level data products)– Practice for JWST (on hook)

• This talk– Progress report– Stimulate ideas, pique interest

This Talk• Intro• Mapping– Frontier Fields: hst2galign routine– Enables inter-image mapping

• PSFs– PSF study of F606W WFC3/UVIS archive– Suggested “standard” PSF format

• Tools– “bundles”: convenient structure for modeling

Frontier Fieldshst2galign

Frontier Fieldshst2galign

“CENTROID” POSITIONS

“TEMPLATE-FIT” POSITIONS

Other Filters/Detectors

hst2galignsoftware takes *_flc.fits list and flagsit outputs “mat” files of point-association

lists allows mapping of FLT to REF frame

uses distortion reference images

This Talk• Intro• Mapping– Frontier Fields: hst2galign routine– Enables inter-image mapping

• PSFs– PSF study of F606W WFC3/UVIS archive– Suggested “standard” PSF format

• Tools– “bundles”: convenient structure for modeling

PSFs: Spatial variation F606W WFC3/UVIS

• Spatial variations

Star Images from Entire UVIS Archive5,013,607 star images extracted

raster of 21×21 pixels extracted for eachrecord (i,j) location and sky and fitted position/flux

Star Images from Entire UVIS Archive5,013,607 star images extracted

raster of 21×21 pixels extracted for eachrecorded (i,j) location and sky and fitted position/flux

S/N Distribution

PSF Moments

“A” excess

“B”

exce

ss

PSF Moments

“A” excess

“B”

exce

ss

“A” corner

How the upper-left PSF changes with

Focus

How the upper-left PSF changes with

Focus

Solving for the Focus in an exposure

To do: * extend analysis to entire chip* determine how many stars are needed* correlate with focus model

Encouraging!

Suggested “Standard”PSF format

• A 3-D fits image cube– NX×NY = NTOT PSFs

• Specify i,j fiducial locations

– Each PSF is 101×101 pixels• Super-sampled ×4• Covers 25×25 pixels (out to r=12.5)• Normalized to have flux of 1.0

within 10 pix

– Already available for many filters/detectors

• Tools:– get_psf[iCEN,jCEN,psfloc()]

• Returns local PSF at detector location

– rpsf_phot[Δx, Δy,psfloc()]• Returns fraction of light in particular pixel

NPSFs

%

101101

Example ofthe fitsheader fora STD PSF

This Talk• Intro• Mapping– Frontier Fields: hst2galign routine– Enables inter-image mapping

• PSFs– PSF study of F606W WFC3/UVIS archive– Suggested “standard” PSF format

• Tools– “bundles”: convenient structure for modeling

A High-level Product: Pixel Bundles• Putting it all together– Uses the alignment from hst2galign– Identify an object of interest and size (ie, 21×21 pix)

• Bundle-extraction routine: hst2bundle

bundle_21x21_00630_00979.fits

The Bundle Itself• A set of NRAST × NRAST × NIMs fits images– in 8 extensions

1-3 FROM FLT4-5 DIRECT MAPPING EACH PIXEL TO REF-FRAME6-7 REVERSE MAPPING FROM REF-FRAME TO LOCAL PIXEL 8 PSF (actually 101 × 101 × NIMs)

The Pixels in theBundle

A Bundle is Like an “object”

• Can imagine python “methods” to analyze• Bundle2process example routine– DUMP (output P,E,DQ,I,J,U,V,PSF…)– PIXCOMP (to rem artifacts, match sky)– XYZFIT_IND – XYZFIT_SIM– STACKING

• Point-source fitting• Deconv fwd-model

• Documentation…

This Talk• Intro• Mapping– Frontier Fields: hst2galign routine– Enables inter-image mapping

• PSFs– PSF study of F606W WFC3/UVIS archive– Suggested “standard” PSF format

• Tools– “bundles”: convenient structure for modeling

• Discussion…