Optical Photometry

Sergio Ilovaisky (OHP)

5th NEON School

23 July-5 August 2006
 Foreword

• Starlight coming to us has :

– traversed the Earth's atmosphere

– reflected off the telescope mirrors
– passed through the filters and dewar window
– absorbed by the CCD chip
– before the data was read out and digitized

Our task is to remove the signature of the

atmosphere and instrument and try to recover as
much information as possible from what entered the
Earth's atmosphere
 summary

• Getting to know your instrument

• CCD and filter characteristics
• Atmospheric extinction
• Flat-fielding: is it an art or a science ?
• PSF-fitting vs. Aperture photometry
• Differential vs. All-sky photometry
• Transforming into a standard system
Getting to know your instrument
• The camera is attached to the telescope
using an adapter or bonette

• The CCD is housed in a dewar and is

held at low temperature (-90°C)

• Software controls all instrument

functions

• The adapter normally holds:

– A filter wheel
– A shutter
– A viewing / autoguider system
 Something about shutters
• There are two types of mecanical shutters used in
CCD cameras:
Iris-type: Since the opening and closing are done in a radial
mouvement, the exposure in the center of the image will be
greater than at the edges. You must calibrate this effect and
correct for it in the case of short exposures

Curtain-type: Here all pixels get the same amount of

exposure if the opening and closing sequence are done in
the same sense. This type is to be preferred, but you cannot
change what an observatory will offer you…
 CCD parameters

• A given CCD camera system is characterized, among other things, by:

• Dynamic range (full-well capacity)
• Number of bits of A/D converter
• Read-out noise (RON)

• In order to use the full dynamic range of the system, the gain setting
should be : gain ~ (dynamic range)/(2Nbits), with the added constraint
that : gain < read-out noise

• For the TK 1024 camera of the 1.2-m OHP telescope, the well depth is
200000 e- and Nbits = 16 (65536 discrete levels). The best gain value
should thus be ~ 3e–/ADU

FACTS :
For the 1.2-m system the gain is 3.5 e –/ADU and the RON is 8.6 e –
Example of CCD wavelength response
UBVRI filter transmission curves

This is the so-called Johnson-Kron-Cousins system

Other filter systems exist…

• Johnson : UBVRI + JHKL

• Geneva : UB1BB2V1VG
• Walraven : VBLUW
• Stromgren : uvby Hβw Hβn
• Thuan-Gunn : uvgr + iz
• Sloan Survey : griz
• Hipparcos+TYCHO : HP, BT VT
Not all R and I filters are alike !
 Air mass
Flat Earth approximation

Real Earth

An approximation : X = sec z · (1 – 0.0012 (sec2z – 1))

 Magnitudes and color indices

• Astronomical magnitudes are defined, following

Pogson (19th century), as :

m = – 2.5 log(I)
• The difference between the magnitudes in two
different passbands is called a color index:

B–V = mB – mV = –2.5 log(IB/IV)

in the case of the B and V bands
 Bouguer’s Law
• Absorption by the atmosphere can be described by a standard
exponential law:
I = Io exp (–K·h)
where Io is the intensity outside the atmosphere, I is the value
measured after traversing a thickness h and K is the absorption
coefficient per unit length caracterizing the properties of the
absorbing medium. K·h is called the "optical depth".

• Taking logarithms, we can re-write this in magnitudes as:

m = mo – k ·X
where X is the airmass and k is called the extinction coefficient
per unit airmass. This is called Bouguer’s law (18th century).
Atmospheric extinction
Atmospheric extinction
coefficients depend on
wavelength and on the
altitude of the observing
site.

They can vary with time

(from season to season,
from night to night and
within a single night).

For OHP, the coefficients

are at least 70% larger
than the values for La Silla.
We recommend assuming
they are twice those for La
Silla.
Components of extinction
Effective wavelength
• Measuring objects whose
spectra have different colors
implies that the effective
wavelength of a broad-band
filter changes with the color
of the object
• The atmospheric extinction
coefficient changes
accordingly
• We speak thus of the effective
wavelength of a given filter
for a given star color

λeff = ∫ λ ⋅ e( λ) ⋅ S( λ)⋅ dλ ∫ e( λ)⋅ S( λ)⋅ dλ

€
Bouguer plot
• A plot of apparent
magnitude versus
airmass

• A least-squares fit
through many
observations of the
same star trhough
the night results in a
reliable determination
of the extinction
If extinction varies from night to night
• A simple example shows that
it is useful to combine
information obtained during
different nights in order to
obtain better reductions

• In the top half, two nights are

reduced independently, giving
different extra-atmospheric
magnitudes

• In the bottom half, the same

value is imposed to mout,
resulting in a more accurate
extinction
If extinction varies during the night

Simulation of a night with variable

extinction (kV = 0.26, 0.24, 0.22, 0.20)
decreasing with time
• Star M is before meridian and rising
• Star D is past meridian and setting
Taking the measurements at face value
(dotted lines) we derive incorrect
extinction AND extra-atmospheric
magnitudes
Assuming extinction is the same for both
stars and that it varies slowly and
isotropically, one can solve the necessary
equations to get the extra-atmospheric
values and derive the extinction
Extinction at La Silla

Two volcanic eruptions took

place during this monitoring by
the Geneva group:
• El Chichón (March 23 and
April 4, 1982)
• Pinatubo (June 12 and 13,
1991)
Extinction at La Palma
 How extinction works
• For broad-band filters, extinction coefficients have to take into
account a dependence on object color. For a filter having a
central wavelength λ and an associated color index co:

kλ = kλ' – kΔλ" · co
The first-order term is the atmospheric extinction proper while
the second-order term takes into account the dependence on
object color due to the width of the filter bandpass.

The first term varies with time but the second term usually
remains constant for a given instrumental setup. The
instrumental color should be the ‘extra-atmospheric’ value, but
observed values are used.
 How to determine extinction
• Extinction coefficients can be determined accurately by
observing a set of standard stars spanning a wide range of
colors for several values of the airmass, both before and after
meridian passage.

• A faster, separate determination of the primary and secondary

extinction coefficients is possible :

– For k’λ observe pairs of stars having the same known color at
low and high airmass (ascending, descending) so the second-
order term goes away

– For k”λ observe pairs of stars of different known colors, located in

the same field of view, at different airmasses, so the first-order
term goes away
 What if extinction is unknown ?

• Best way around is to measure your program stars

and your standard stars at the same airmass

• If the extinction is not only unknown but variable,

then pick your standard stars as close as possible in
the sky to your program stars and alternate between
the two to be able to interpolate any extinction
changes taking place
 Reduction process

• Steps taken to minimize the influence of data

acquisition imperfections on the estimation of
the desired astronomical quantity

• At some time, REDUCTION steps are replaced

by ANALYIS steps, which depend on the
research subject being pursued, although the
dividing line is not well defined
 How to reduce data

• The best approach to dealing with a problem is to avoid it to

begin with

• Perform only reduction steps that matter. If application of a

reduction process results in final lower noise, then apply it

• Know when to stop attemps to better reduce your data. Have

you reached the ultimate precision you expect ?

• Know how to estimate how well you should do in a set of CCD

reductions
Removing instrumental signature

• Before making any measurement on your CCD frames,

the data are have to be corrected for two kinds of
effects:

• Additive
– Dark counts and "cosmic" rays
– Bias level systematics
– Scattered light or background gradients

• Multiplicative
– Pixel-to-pixel variations in sensitivity
 What's flat-fielding ?
• In order to correct for changes in sensitivity across the detector, you need to
take exposures of a uniformly illuminated source

• Correction for pixel-to-pixel variations in quantum efficiency is usually the most

important reduction process and is one of the most difficult to obtain correct
calibration data to support

• Pixel-to-pixel variations are normally coupled with optical vignetting corrections

although they differ in nature

• Be prepared to spend substantial time at the telescope and a large part of the
total effort in the reductions to arrive at good flat-fields

• Difficulties:
• Provide uniform illumination to 10-3 or better
• QE variations may have a dependence on wavelength. Thus the
correction is valid only if the spectral distribution of the light over the
filter bandpass matches that of the object being observed
The first is difficult enough to satisfy, the second is essentially impossible to
satisfy for broad-band photometry
 Flat-fielding methods

• Flats using the dawn or dusk sky

– Advantages: Really flat fields for small fields of view; High illumination levels
– Disadvantages: Color not well matched to night sky or stars; Possible
polarization; Rapidly changing illumination levels; Stars may appear in frame

• Flats using dome illumination

– Advantages: Controllable illumination; Can be done in daytime; Possibility of
averaging many such frames
– Disadvantages: Color not well matched by lamps; Possible shadows, diffuse
light or gradients

• Flats using "empty" star fields

– Advantages: Good color match for sky; Same conditions as data frames
– Disadvantages: low-illumination levels; lots of reduction effort; Wastes dark
observing time
 How to prepare your flats
• Since you are going to divide your science frame by the flat-field
frame, it should have the highest S/N possible

• Obtain several frames (minimum 5) for illumination levels close

to 1/2 of saturation

• Compute a median frame to reject "spikes"

• Subtract well-averaged bias frame or bias value

• Divide by the average value of the frame to obtain a

normalized flat

• Obtain flats every day or two since dust specks (on filters or on
dewar window) might come and go
 Averaging calibration frames

• Many calibration frames have to be averaged to reduce the

noise and the final result of the reduction will be:

Final_result = (Science_frame - Av_Dark)/(Av_Flat - Av_Dark)

• The error in the final result frame can be expressed as:

σ2FinRes = {(F-D)2σ2S + (S-F)2σ2D + (S-D)2σ2F}/(F-D)4

Where F, D, S and FinRes stand for Flat, Dark, Science_frame

and Final_result

σ2FinRes will be small if σS , σD and σF are small

 Examples of chromatic effects
TK 1024

TK 512 • In the top row are flats in B, V, R

and I for the current TK1024 chip at
the 1.2-m with little or no visible
chromatic effect
• In the bottom row are flats in R
and I for the old TK512 with an
extreme chromatic effect
 The case of the moving spots

This illustrates the pitfalls when using flat-fields taken at other times:
the dust specks on the dewar window have moved !
Flats taken on : 29 February, 28 March and 2 September
Sometimes they even move during a week !
 Image sampling
Different cases of
image sampling
- horizontally: as a
function of the projected
image size (in pixels)
- vertically: as a function
of image centering in the
array

Critical sampling (Nyquist criterion) ≈ width of PSF

For a Gaussian PSF this corresponds to the FWHM = 2.355 pixels
What’s the real shape of a PSF ?

The actual point spread function

of a star image is complex:
• A central core
• An exponential drop
• An extended inverse-square
aureole

King (1971)
Red "halo" effect
• The PSF profiles obtained at the
1.2-m OHP telescope with the B,V
or R filters agree with the King
profile
• However, the I filter profile
exhibits a "halo" due to light
passing through the CCD, being
diffused in the glass header,
I-filter reflected back into the CCD from
the rear metallized surface and
finally being detected.
V-filter •This is a well-known effect of
thinned CCDs.

Michard (2002)
 Instrumental magnitudes
The photons (electrons) registered by the CCD in a time t coming from a given star can be
written as:

λ2
S = ∫ φ (λ )⋅( λ )⋅ A⋅ε (λ )⋅T (λ )⋅E(λ )⋅Q(λ )⋅t ⋅dλ
λ1
hc

where φ(λ) is the€ flux outside the atmosphere (in erg/cm2 s Å), A is the effective collecting area,
ε(λ) is the efficiency of the telescope optical system, T(λ) is the filter transmission, E(λ) is the
atmospheric extinction and Q(λ) is the quantum efficiency of the CCD.

The instrumental magnitude of a star is obtained from the number of electrons generated in the
CCD by the light from the star :
minst = –2.5 log(S*)

S* is obtained by suming is over the npix pixels included in the digital aperture centered on the
star:
n pix n pix

S* = ∑ a j ⋅ s j − ∑ aj ⋅b
j =1 j=1

where aj is the area of the jth pixel, sj is the electron count in the jth pixel (star+sky) and b is the
sky background count per pixel.
Determining the background ?
EASY HARD
Where is the background here ?

Same field, but shown with different cut values

Spectrum of the background ?

Foltz & Massey (2000 Publ. Astr. Soc. Pacific, 112, 566)
 Characterizing the background
The optimal radius of the
aperture is close to the
FWHM of the stellar profile

The optimal inner radius of

the sky annulus is close to
3 times the FWHM

The background is computed from the histogram of the pixels in the annulus. If
noise spikes or other objects are present in the sky, they should be rejected before
computing its value (sigma-clipping).

The mode is the best estimate of the sky. It is defined as the maximum of the
histogram (the most probable value). Assumptions: (1) the histogram is unimodal
(only one peak) and (2) there are enough pixels to get meaningful statistics.

The sky annulus should contain at least 100 pixels. The outer radius should not be
too large since the background may be different far away.
Starlight within a given aperture
 Signal-to-Noise
CCD Equation: N* is the total number of
S N* photons (electrons) from the star, NSky and
= Ndark are the number of electrons/pixel from
N 2
N * + npix ( N Sky + N dark + N RON ) the sky and dark signal, and NRON is the
read-out noise per pixel. npix is the number of
pixels within the aperture.

€ S N*
Bright star : N
= ≈ N* Poisson noise dominates !
N*

S N*
Faint star : =
€ N npix 2
N * + npix (1+ )( N sky + N dark + N RON + G 2σ f2 )
nB

nB is the number of pixels used in the background estimation, G is the gain and σf2 is an estimate
of the error introduced in the A/D converter (the round-off error of ±0.5 ADU)
€
S/N vs. aperture radius
• The signal to noise ratio
obtained for the measurement of
a point source is not constant as
a function of aperture radius

• There is an optimum
radius at which the S/N will be a
maximum. This is because the
signal from the sky background

• This optimum radius the FWHM

of the stellar profile.
(In this illustration the pixel size
is 0.4" and the FWHM is 1.2")
Classical aperture vs. Optimum aperture

In the left-hand image one would expect to In the right-hand image one gets the best
get best results with a radius large enough S/N with an optimal radius equal to the
to enclose all the light for the brightest stars FWHM of the star profile (0.68"/pixel and 2"
in the frame FWHM)
Actual growth curves
• Here are actual growth curves measured
from five different stars in a given CCD
frame.

• The three brightest stars follow the

theoretical expectation

• The two faint stars start out in a similar

manner, but eventually the background level
is high enough to overtake their PSF in the
wings

• Corrections, based on the bright stars,

can be applied to these curves to obtain
good estimates of their true brightness

(Remember, in this illustration the FWHM is

3 pixels)
CCD Aperture Photometry
• Basic assumption :
• well-isolated stars
• Determine centers of the stars :
• use digital centering algorithms
• Determine sky background :
• software annulus starting at 3 FWHM
• reject stars and noise spikes
• Determine stellar magnitude :
• circular aperture
• partial pixels handled or not ?
• What aperture size ?
• Optimal radius ~ FWHM
• Curve of growth correction
 PSF-fitting photometry
• Basic assumption • The fitting process
– all point sources can be represented – analytic function (Gaussian,
by a point-spread function Lorentzian, Moffat)
– use only pixels near star center
(fitting radius ~ FWHM)
• Finding stars – fit several stars simultaneously
– use finding algorithm or input when too close to each other
existing table from aperture
photometry
• Different implementations
– DAOPHOT, ROMAPHOT, SPS ,
• Finding the sky level DoPHOT, etc.
– start with results from aperture
photometry • Complications
– Extreme crowding
• Building the PSF – Variable PSF
– use stars with high S/N ratio
– eliminate close neighbors • Follow the CookBook !
 PSF CookBook
• Input existing position table or make initial pass with star-finding algorithm
• Determine sky level for each star and perform aperture photometry on all objects
• Generate an initial estimate of the PSF using the "best" stars
• Use the preliminary PSF to fit all stars to obtain more accurate magnitudes
• Subtract the stars that have been measured and run the finding algorithm on the
residual image
• Fit the newly "discovered" stars and any stars from the original frame whose
magnitudes have changed
• Subtract all the stars from the frame and redetermine the sky levels for each star
using the new residual frame
• Refine the estimate of the PSF. There should be more usable stars since close
companions have been removed
• Refit all stars in the list
• Repeat the sky estimation and PSF refinements if needed. Third-generation PSF
usually give the best results
 You can discover new stars !
By constructing a PSF from well-isolated stars, one can fit this
PSF to stars in crowded areas and proceed to subtract them
out of the frame.
New stars (star "e" below) may appear underneath (here under
star "a" !
 Transforming observations into a
standard system
In order to correct your measurements for atmospheric
extinction and for color-dependent terms, you must observe
standard stars :
• having a wide range in color
• for a large range in airmass

An instrumental magnitude at a given airmass X, mλ(X), is

related to its value outside the atmosphere, mλο by the relation
(Bouguer ’s law) :

mλ(X) = mλo + (k'λ + k"Δλ·cΔλ)X

where cΔλ is a color index and k' and k" are the first and second-
order extinction coefficients
 Transformation equations
Here we want to relate extra-atmospheric instrumental values to well-known
catalogued values

• For a color index : Co = α + β · co

• For a magnitude : Mo = mo + γ + δ · co
Where the subscript "o" denotes extra-atmospheric values, lower-case
symbols represent instrumental values and upper-case symbols represent
catalogued values. α and γ are the "zero-points" while β and δ are
proportionality constants.

These proportionality constants should be close to 1 if the instrumental

system is well-matched to the standard system

A least-squares fit to mo and co determined for a good number of stars of

different colors and magnitudes should yield the unknown constants α, β, γ
and δ
 Differential photometry
• The difference in magnitude between two objects depends on the
color difference and the airmasses. In the case of B and V
measurements taken at two different airmasses X1 and X2:

ΔV = Δv - k'vΔX + δ·Δ(B–V)
Δ(B-V) = β·Δ(b–v) – β·k'bvΔX – β·k"bvΔ(b–v)<X>

where ΔX = X1-X2 and <X>=(X1+X2)/2. Lower-case denotes

instrumental magnitudes at a given airmass while upper-case
denote extra-atmospheric values in the standard system.
 High-precision photometry
• In the case of variable objects, the best approach is to select several
comparison stars within the same frame and measure the variable object
relative to them. In this manner very high accuracy can be reached

• As an example, here is the light curve of the exoplanet transit discovered last
year at OHP in the very bright (V=7.7) star HD189733. The errors are 2-3
mmag
 Bibliography
• Methods of Experimental Physics, Vol.12, Part 1: Astrophysics -Optical and Infrared,
N.Carleton (Ed), Academic Press 1974
• Handbook of CCD Astronomy, Steve Howell, Cambridge University Press 2000
• Astronomical Techniques, W.A.Hiltner (Ed), University of Chicago Press 1962
• Introduction to Astronomical Photometry, M.Golay, D.Reidel Publ. Co. 1974
• Astronomical Photometry: A Guide, C.Sterken & J.Manfroid, Kluwer Academic
Publishers 1992
• Astronomical CCD Observing and Reduction Techniques, S.B.Howell (Ed.) ASP
Conference Series Vol 23, 1992
• CCD Precision Photometry Workshop, E.R.Craine et al. (Eds.), ASP Conference Series
Vol. 189, 1999
• Stellar Photometry -Current Techniques and Future Developments, C.J.Butler &
I.Elliott (Eds.) IAU Coll. 136, Cambridge Univ. Press 1993
• CCD Astronomy, C.Buil, Willmann-Bell 1991
• A Practical Guide to CCD Astronomy, P.Martinez & A.Klotz, Cambridge Univ. Press 1998
 Bibliography (cont'd)
Filters
• M. Bessell (1990) PASP 102, 1181

Aperture Photometry
• S.Howell (1989) PASP 101, 616
• K.Mighell (1999) ASP Conf. Proc. Vol. 189, p.50

PSF-fitting Photometry
• P.Stetson (1987) PASP 99, 191
• K.Janes & J.Heasley (1993) PAPS, 105, 527

Standard Stars
• A.U.Landolt (1992) AJ 104, 340
• C.Christian et al. (1985) PASP 97, 363