Image Calibration IR arrays sky subtraction:

For ground-based IR, the airglow or thermal background is many orders of

magnitude brighter than the target sources. Errors in the flat field can leave
behind errors in sky subtraction that can dominate. So “chopping” is used.
Image Calibration: Summary

flat
Astronomical Errors

There are many ways to express errors which are equivalent:

“I measured the magnitude as mr = 12.0 ± 0.1”

“I measured the magnitude as mr = 12.0 with a S/N = 10”

“The one-sigma errors were 0.1 for a star with mr = 12.0”

Error Propagation

What is the magnitude error for a star with V = 20 with S/N = 20

The S/N is the fractional error on the measure flux.

Need to convert flux error into magnitude error.

Recall that magnitude is defined as m = -2.5 log10(flux):

N
m 2.5log10 (1 + )
S
2.5 N 1 N 2
= ( ) + ...
2.3 S 2 S <-- Taylor expand

1.086
=
S/N

For S/N = 10, this is a 10% error on the measured flux.

The error in magnitude for this star is V = 20 ± 0.11 mag.
Error Propagation

What is the magnitude error for a star with V = 20 with S/N = 20

The S/N is the fractional error on the measure flux.

Need to convert flux error into magnitude error.

Recall that magnitude is defined as m = -2.5 log10(flux):

N
m 2.5log10 (1 + )
S
2.5 N 1 N 2
= ( ) + ...
2.3 S 2 S <-- Taylor expand

1.086
=
S/N

For S/N = 10, this is a 10% error on the measured flux.

The error in magnitude for this star is V = 20 ± 0.11 mag.
Error Propagation

What is the magnitude error for a star with V = 20 with S/N = 20

The S/N is the fractional error on the measure flux.

Need to convert flux error into magnitude error.

Recall that magnitude is defined as m = -2.5 log10(flux):

N
m 2.5log10 (1 + )
S
2.5 N 1 N 2
= ( ) + ...
2.3 S 2 S <-- Taylor expand

1.086
=
S/N

For S/N = 10, this is a 10% error on the measured flux.

The error in magnitude for this star is V = 20 ± 0.11 mag.
Error Propagation

For a given star, I measure its magnitude in two filters:

mV = 12.5 ± 0.2
mR = 12.0 ± 0.3

The color of a star is defined as the difference in magnitude

V - R = mV - m R
= 12.5 - 12.0
= 0.5

How do we combine the error on this measurement?

Error Propagation

Given two independent set of measurements, how are these combined?

In general, given some function x = f(u,v,...), we want to know σx

2 2
x x
2
x = 2
u + 2
v
u v
For example, to determine the error in color, mV - m R:

color = 2
V + 2
R

For additive noise, errors add in quadrature

Error Propagation

For a given star, I measure its magnitude in two filters:

mV = 12.5 ± 0.2
mR = 12.0 ± 0.3

The color of a star is defined as the difference in magnitude

V - R = mV - m R
= 12.5 - 12.0
= 0.5

(V R) = 0.32 + 0.22
= 0.4
Error Propagation

For a given star, I measure its magnitude in two filters:

mV = 12.5 ± 0.2
mR = 12.0 ± 0.3

The color of a star is defined as the difference in magnitude

V - R = mV - m R
= 12.5 - 12.0
= 0.5

(V R) = 0.32 + 0.22
= 0.4
Going back to S/N calculations

Readout Noise

In addition to the errors from Poisson statistics (statistical noise), also need to consider
instrumental errors. An error source for CCD detectors is readout noise.

Readout noise is an additive noise (with zero mean!) that comes from the detector and is
independent of signal level.

The Readout noise in a CCD is typically σRN= 3-4 electrons per pixel.

Including this noise term in the S/N equation:

S S
=
N S + Npixel B + Npixel 2
RN
Signal-to-Noise: ADU vs. Electrons

The S/N equation has three major terms:

S S
=
N S + Npixel B + Npixel 2
RN

Although we measure S and B in ADU, the noise is calculated in electrons since

we have picked an arbitrary gain factor: # of electrons = gain * ADU

Thus, we really should write the S/N equation as:

S gSADU
=
N ( gSADU + Npix (gBADU + 2 ))1/2
RN
Signal-to-Noise: Example #1

The CCD used to take this image has:

Gain: g = 1.2 e-/ADU
Read Noise: σRN = 8 e-

The exposure time is t = 200 seconds.

The average background count is 65 ADU

For the bright star, I measure a flux of 40,000 ADU over 50 pixels
S gSADU
=
gSADU + Npix (gBADU + 2 )
N
√ RN

= 200
Signal-to-Noise: Example #1

The CCD used to take this image has:

Gain: g = 1.2 e-/ADU
Read Noise: σRN = 8 e-

The exposure time is t = 200 seconds.

The average background count is 65 ADU

For the bright star, I measure a flux of 40,000 ADU over 50 pixels
S gSADU
=
gSADU + Npix (gBADU + 2 )
N
√ RN

= 200
Signal-to-Noise: Example #2

The CCD used to take this image has:

Gain: g = 1.2 e-/ADU
Read Noise: σRN = 8 e-

The exposure time is t = 200 seconds.

The average background count is 65 ADU

For the fainter star, I measure a flux of 2000 ADU over 50 pixels
S gSADU
=
N √gSADU + Npix (gBADU + 2 )
RN

= 24
Signal-to-Noise: Example #2

The CCD used to take this image has:

Gain: g = 1.2 e-/ADU
Read Noise: σRN = 8 e-

The exposure time is t = 200 seconds.

The average background count is 65 ADU

For the fainter star, I measure a flux of 2000 ADU over 50 pixels
S gSADU
=
N √gSADU + Npix (gBADU + 2 )
RN

= 24
Signal-to-Noise: Example #1: Signal-limited observations

If the first noise term dominates, then the observations are ‘signal-limited:
S S
=
N S + Npixel B + Npixel 2
RN

S
S
S

The S/N improves by increasing the Signal, e.g., by increasing exposure time.

Example, in signal limited case, if we want to increase our S/N from 10 to 100
(e.g., increase precision from 10% to 1%), need to increase exposure by 100x.

The fact that S/N usually varies as sqrt(time) drives development of bigger telescopes
Signal-to-Noise: Example #2: Background-limited

If the second noise term dominates, then the observations are ‘background-limited:
S S
=
N S + Npixel B + Npixel 2
RN

S
Npix B

For faint targets, observations are often background limited. S/N still improves
with exposure, but faster improvement if we reduce background level or region
over which measurement is made.

This drives observations during dark moon time, or better seeing conditions.
Signal-to-Noise: Example #3: Detector-limited

If the third noise term dominates, then the observations are ‘detector-limited:
S S
=
N S + Npixel B + Npixel 2
RN

S
Npix B

If observations are detector (read-noise) limited, can improve S/N by integrating

longer (and move into S or B-limit), reduce number of pixels... or get a better CCD.

=> this suggests a minimum exposure time for optimal observing.

 Read Noise Regime Photon Noise Regime

log (Noise)

Counts (ADU)
Magnitudes and Photometry

• Recall: m = m0 - 2.5log(Ncounts) + 2.5log(∆t)

Instrumental magnitude
Magnitudes and Photometry

How to determine m0, the magnitude zeropoint?

 Magnitudes and Photometry

How to determine m0, the magnitude zeropoint?

Always let someone else do the dirty work: standard stars

Historically, optical standards: Historically, IR standards:

Landolt (AJ 104 340 1992) Elias et al.
UBVRI Vega-based magnitudes (AJ 87 1029 1982) JHKL
for several 100 stars, near the magnitudes for 50 stars.
equator at variety of RA’s.

Nowadays: SDSS Nowadays: 2MASS, WISE

Magnitudes and Photometry
A bit of trivia...
 Magnitudes and Photometry
Don’t forget about extinction!

Standard star method assumes that all factors comprising m0 are

identical for the observations of targets and of standards.
Transmission of the atmosphere is likely to change.
 Magnitudes and Photometry
Don’t forget about extinction!

Standard star method assumes that all factors comprising m0 are

identical for the observations of targets and of standards.
Transmission of the atmosphere is likely to change. Measure:

Once m0 and cX are determined from the set of

standards, we can use this to get calibrated magnitudes
of any target at any airmass
Photometry

The signal from a given star/galaxy is spread over

several pixels.
Photometry

normal star

saturated star
Photometry

There are two main methods to measure the brightness of stars:

-- Aperture photometry: Count up all flux inside an aperture,
Count up background flux in annulus
Subtract background from star flux

-- PSF photometry: Determine point spread function (PSF) of instrument

Fit PSF + background to each stars
Aperture Photometry

• Choice of aperture? Strictly, standards need to be

measured through same aperture as targets.

• Also this presumes that the PSF is exactly the

same for the standards and targets.
 Aperture Photometry

Choice of aperture?
Aperture Photometry

• Choice of aperture?

• For a given aperture size, same fraction of a stars

counts falls within the aperture regardless of
brightness. Use bright stars to measure an aperture
correction.

• First example of a more general principle in

photometry: necessary to know the PSF from
bright stars to do a good job.
1. Find Centers of All Stars

Identifying unresolved stars

(not resolved galaxies), we
can use the shape of objects.

-- Object needs to cover more

than one pixel (remove CR).

-- Object needs to cover less than

4-5 pixels (remove galaxies)
1. Find Centers of All Stars

Identifying unresolved stars

(not resolved galaxies), we
can use the shape of objects.

-- Object needs to cover more

than one pixel (remove CR).

-- Object needs to cover less than

4-5 pixels (remove galaxies)
1. Find Centers of All Stars

One method to identify the center of a star (xc, yc) is via ‘marginal sums’.

xc = Ixy
y
y
yc = Ixy
x
x
With noise and multiple sources you have to decide what is a source and isolate it:

Marginal Distributions:
Decide on a ‘typical’ star size (L)
Extract small postage stamps size 2L+1
Find center by collapsing along x/y
Remove non-unique detections
1. Find Centers of All Stars

One method to identify the center of a star (xc, yc) is via ‘marginal sums’.
xcen = ∑x∑yx(R[x,y] - B)
∑xx∑c y=
(R[x,y]-B)
Ixy
y
y
yc = Ixy
x
x
With noise and multiple sources you have to decide what is a source and isolate it:

Marginal Distributions:
Decide on a ‘typical’ star size (L)
Extract small postage stamps size 2L+1
Find center by collapsing along x/y
Remove non-unique detections
 PSF Fitting Spitzer IRS red and blue PSF’s

• When the stars in an image

become crowded, we need a
different approach from
aperture photometry.

• PSF fitting: involves creating a

model of the image and
adjusting its parameters until it
best matches the observed
function. Can be quite a
formidable task.

DAOPhot, DoPhot,
etc.
Real PSF’s for the High Resolution Camera on HST

Fig. 1.— The x and y profiles of the PSFs found for the various filters.
50 kpc
25 kpc

Nidever+ 2010
Large Magellanic Cloud
3 Epcoh Analysis (ACS+WFC3) Random Errors:

ACS1
ACS2
WFC3
-64

-66

-68

-70

-72

-74

90 80 70
RA