ThesisPDF Available

Mid-IR background calibrations for the E-ELT’s METIS instrument


Abstract and Figures

METIS, one of the E-ELT's first instruments, will not offer classical chopping and nodding. If this is not solved then the residual background on reduced observations will be of the same order or larger than potential science targets. For this reason an investigation has been started to understand and quantify the nature and source of the residuals in order to infer whether or not it is possible to reduce the background and reach the shot noise limit. To answer this question we have developed a specialised observing plan that allows us to look at the relation between the background residuals and various variables like the chop throw, chop frequency, chop direction, telescope altitude, filter and rotation angle. This plan has been executed on both the VLT and the GTC using the VISIR and CANARICAM IR instruments respectively. From the analysis of this data we have discovered that contrary to what is believed in the literature the high order residuals are likely to be caused by the telescope and that they are stable over time but not over rotation angle, while the gradient is caused by the atmosphere and is not constant over time. Using this theory we have found a new way of reducing the background without nodding that, for chopping throws below 10", gives nearly identical results when compared to classical chopping and nodding. We believe that with this method we have found a solution for the challenge stated above but conclude that more rigourous follow-up observations are needed.
Content may be subject to copyright.
Mid-IR background calibrations for
the E-ELT’s METIS instrument
submitted in partial fulfillment of the
requirements for the degree of
Author : Alexander G.M. Pietrow
Student ID : 1054678
Supervisor : Prof. Bernhard Brandl
2nd corrector : Dr. Matthew Kenworthy
Leiden, The Netherlands, July 22, 2016
Mid-IR background calibrations for
the E-ELT’s METIS instrument
Alexander G.M. Pietrow
Leiden Observatory
P.O. Box 9500, 2300 RA Leiden, The Netherlands
July 22, 2016
METIS, one of the E-ELT’s first instruments, will not offer classical
chopping and nodding. If this is not solved then the residual
background on reduced observations will be of the same order or
larger than potential science targets. For this reason an investigation
has been started to understand and quantify the nature and source of
the residuals in order to infer whether or not it is possible to reduce the
background and reach the shot noise limit. To answer this question we
have developed a specialised observing plan that allows us to look at
the relation between the background residuals and various variables
like the chop throw, chop frequency, chop direction, telescope altitude,
filter and rotation angle. This plan has been executed on both the VLT
and the GTC using the VISIR and CANARICAM IR instruments
From the analysis of this data we have discovered that contrary to
what is believed in the literature the high order residuals are likely to
be caused by the telescope and that they are stable over time but not
over rotation angle, while the gradient is caused by the atmosphere
and is not constant over time. Using this theory we have found a new
way of reducing the background without nodding that, for chopping
throws below 10”, gives nearly identical results when compared to
classical chopping and nodding. We believe that with this method we
have found a solution for the challenge stated above but conclude that
more rigourous follow-up observations are needed.
Version of July 22, 2016– Created July 22, 2016 - 20:26
1 Introduction 1
1.1 The Mid IR Background 1
1.2 Classical Calibrations 7
1.3 Challenges 11
2 Data 15
2.1 Technical time 15
2.2 Observations 16
3 Methods 21
3.1 Classical Chopping 25
3.2 Chopping without nodding 25
3.2.1 Stability over time 26
3.2.2 Polynomial fitting 28
3.2.3 Principal Component Analysis 29
3.2.4 Multidirectional chopping 31
3.3 Atmosphere VS Telescope 32
3.3.1 Wavelength Dependencies 32
3.3.2 Derotator Test 35
3.3.3 Effects as a function of airmass 35
3.3.4 Closed dome 36
4 Results 37
4.1 Effects as a function of Throw 38
4.2 Residual stability over time 39
4.3 Alternative chop methods 40
4.4 Disentangling Backgrounds 45
4.5 The Derotator test 49
Version of July 22, 2016– Created July 22, 2016 - 20:26
4.6 Effects as a function of airmass 50
4.7 Testing on a source 51
5 Conclusions and Discussion 55
Version of July 22, 2016– Created July 22, 2016 - 20:26
Chapter 1
1.1 The Mid IR Background
The mid-infrared (MIR), usually defined as the region between 5 and 25 µm, is
a relatively underutilized band in the EM spectrum. This is a region that is very
important in certain fields of astronomy, especially those where dust plays a big
role. Possible science topics that would gain from MIR observations could be
the evolution of gas in protoplanetary disks, the study of star and planet evo-
lution, accretion and outflows around protostars and the potential detection of
biomarkers in exoplanet atmospheres. Besides this we can also study thermal
structure, chemistry and dynamics in the thick atmospheres of our local gas gi-
ants and objects at moderate redshift. A more in depth look into these topics
can be found in (Brandl et al., 2008; Otarola, 2014; Lacy et al., 2002).
Observations in this regime are done primarily by space telescopes. This is
because this wavelength regime suffers heavily from our Earth’s atmosphere,
which radiates strongly in the IR wavelengths. Influenced by the vertical pro-
files of temperature, pressure and molecular abundances of, among other things
H2O,CO,CH4. The magnitude of all of these factors in turn depend strongly
on the telescopes location and altitude above sea level. (Kendrew et al., 2010)
The simplest way to understand this is to imagine the atmosphere as a 300K
greybody with varying transparency and fully opaque regions. This effect is es-
pecially bad in the MIR due to the fact that the atmospheres blackbody peaks in
this regime due to Wiens law. In addition to this, the telescope, its dome and its
optics emit as a greybody of 15%that also peaks at this wavelength resulting
in a high background and a high photon shot noise (PSN). An estimate of the
For the E-ELT, this can differ for other telescopes.
Version of July 22, 2016– Created July 22, 2016 - 20:26
brightness of this background as a function of wavelength has been made by
Kendrew et al. (2010), where the publicly available Reference Forward Model
(RFM)is used in combination with the HITRAN 2004 mollecular line database
(Rothman et al., 2004) to create an estimate of the sky brightness at two alti-
tudes for a theoretical telescope. (2600m for Paranal and 5000m for an ideal, yet
undetermined location.) The results of these estimates can be seen in Fig. 1.1
together with a 10% greybody estimate of the telescope at the local ambient tem-
perature. This shows that the amount of atmosphere between you and the star
and the temperature is a big factor, explaining why most people prefer space
telescopes. We can see a similar plot done by Jeff Meisner that simulates the
Figure 1.1: The atmosphere from about 2 to 10 micron (solid line) given by the RFM
with a resolution of R=3000, plotted together with the telescope (dashed line) at 10%
emissivity. In Red the conditions for Paranal and in blue a high and dry location. Im-
agecredit: (Rothman et al., 2004)
E-ELT and its location in Fig. 1.2. To form a better understanding of the back-
ground we have added a 0th magnitude star that is spread out over one arcsec2
as a comparison. We see that the background is above this value in most of the
Version of July 22, 2016– Created July 22, 2016 - 20:26
1.1 The Mid IR Background 3
MIR regime and that it goes up even further with longer wavelengths. If we
would compare the MIR to optical bands, the obtained background would be
similar to what we see when observing during the day while using a telescope
with luminous optics, so not the most ideal setup.
Figure 1.2: The two main background components of the MIR background are given
by the atmosphere (Red line), with a varying transmission over the wavelength range
and the telescope (The black line is the sum of both) that has been modeled as a gray
body of 15%. The combination of the two backgrounds is enough to outshine some of
the brightest stars, in this case compared to a 1 arcsecond, 0th magnitude star. In the
blue regions one can see the five main observing bands that are used in the MIR. Image
was obtained from Jeff Meisner
A bright sky by itself would be fine, if it would be stable. Unfortunately this is
not the case in the MIR and we get to deal with varying sky brightness changes
with very short (second) time scales over spatially small regions, often much
larger than N, as would be expected from poison statistics. (Geballe and Ma-
son, 2006) In Fig. 1.4 we can see the three main components of MIR radiation,
namely; the telescope, atmosphere and detector. The first two will be looked
into now, while the detector will be covered further down.
The biggest effects in the N band are caused by CO2, CH4, H2O and O3. This
means that an increase in water vapour column (WVC) will cause a reduction in
transmission and an increase in sky background, as is shown for 1.0 and 3.0 mm
H2O in Fig. 1.5. This means that cold, high and dry places with low emissivity
and high cleanliness are the best for MIR observations. Besides this, it is possi-
ble to reach the diffraction limit in this regime, but only with good conditions
Version of July 22, 2016– Created July 22, 2016 - 20:26
Figure 1.3: Left: A diagram showing the brightness of the background in the MIR
(red) and the brightness of a diffraction limited star that has the same surface bright-
ness (black). Right: A diagram showing the relative difference in brightness between
the MIR background (red) and a diffraction limited 10th magnitude star. (green) Both
images were obtained from Jeff Meisner.
and guiding.With the main contributors to background fluctuations being the
WVC, unstable weather, thin cloud, wind-borne dust and sometimes bugs and
To get a better feel for the numbers we do some basic S/N calculations to see
how bright the background is. We can determine this brightness quite easily by
following the method as described at the Gemini website (Geballe and Mason,
2006), but with the numbers for the VLT.
So if we start with an 8.2m class telescope, with a temperature of 300K and an
emissivity of 9% (Rupprecht, 2005), we will get a background of 431 Jy/arcsec2
at 8.7 micron from the telescope and its optics. We can add about 23% (Meisner,
2016) more to this to account for the atmosphere ending up with 530 Jy/arcsec2
of background flux. If we compare this to a 0th magnitude star, which shines at
about 46.6 Jy (Cohen et al., 1999) at 8.7 micron in the N band§and we focus the
star on a 1 arcsecond2region, then we will see that 530
35.21+530 =94% of the light in
that region is background, comparable to Fig. 1.2. Of course this number goes
down when the star is focused on a smaller region, but we generally also look at
much fainter sources. The same star focussed on a diffraction limited 0.3 arcsec2
spot would give us a 57% background for example. This is clearly a problem,
VISIR is diffraction limited when optical seeing is below 0.6” (Tristram, 2016)
§This is based on Alpha boo being 633 Jy at -3.14 mag.
Version of July 22, 2016– Created July 22, 2016 - 20:26
1.1 The Mid IR Background 5
Figure 1.4: The three main components that emit MIR radiation are the atmosphere, the
telescope and the detector itself. Looking at the time-scales we can see that the telescope
is the most stable with stabilities in the order of hours, followed by the atmosphere in
minute time scales and the detector which is stable for subsecond timescales.
as we need a much higher S/N to be able to do any form of science. We can
subtract out most of the background, but we eventually we will hit the ultimate
limit. This limit is the shot noise that scales with the square root of the counts
on the detector. This noise level cannot be surpassed and limits the S/N to our
observations. Fortunately there is a lot to gain between the raw background
and the shot noise limit, so getting to this point is already a great success.
Version of July 22, 2016– Created July 22, 2016 - 20:26
Figure 1.5: The background between 6 and 14 µm showing the difference in back-
ground transmission as a function of the vapor column. Lord et al. (1992)
Version of July 22, 2016– Created July 22, 2016 - 20:26
1.2 Classical Calibrations 7
1.2 Classical Calibrations
The classical solution to removing the large MIR background is a combination
of two methods, namely chopping and nodding. Chopping is the act of moving
the secondary mirror of your telescope by a little bit, shifting the image on the
detector, creating an offset in the pointing by a few arcseconds from the detec-
tors point of view. This operation is repeated at a frequency of a few Hz with
exposures being taken every few ms. This frame rate has to be high enough
to prevent over saturation while being as low as possible to avoid excessive
readout noise. Typically we use frame rates that read out around every 20 mil-
After averaging these frames and subtracting the two pointings, we are left with
a so called ’chop difference frame’ from which all constant components have been
removed. Ideally this would only leave the source and remove the entire back-
ground, but because we moved the telescope beam across the primary mirror, il-
luminating a slightly different part of the telescope, we will see slightly different
parts of the telescope. Because the telescope emits strongly in this wavelength
range, these differences result in residuals and we are left with something in
the order of a few percent of the original background. Unfortunately, if we re-
member the numbers calculated above, this is close to, and usually much higher
then, the brightness of the source and therefore not enough to get a proper S/N.
The classical solution to these residuals is to move or nod the telescope by the
same offset and in the same direction as was used for chopping and chop in
the same way as before. (So that the science target and empty space switch
places.) When repeated about once or twice a minute, this process produces
two pairs of images that both contain the same residuals but have inverted signs
for the source. We get one chop difference frame for each nod position and
when we subtract those we will get a ’nod difference frame’. This gets rid of most
aberrations and gets the noise down to the shot noise limit. This process is
illustrated in Fig. 1.7.
In general this method works relatively well, removing almost all of the back-
ground. However it is very important to have a stable sky in the region that is
being observed, otherwise the difference frames will not cancel each other out
properly leaving large residuals on the final image. (Geballe and Mason, 2006)
Looking at Fig. 1.6, we can see the four aforementioned pointings, of which
two overlap. In Fig. 1.8, we can see how such an image would look on the chip.
These nod difference frames have three copies of the source on the image. (If the
chopping throw was shorter than the image size, otherwise we would see only
Version of July 22, 2016– Created July 22, 2016 - 20:26
one image.) The central image is positive and has counts for half the exposure
time and the other two images are negative and correspond to a quarter of the
time each.
Figure 1.6: A diagram showing the four pointings that are used when chopping classi-
cally. In nod position A the telescope is aimed at the source and then offset by chopping
away from it. In nod position B, the telescope is moved as a whole to the position that
it was chopping to, switching the source with the offset position. Now it chops again.
(Volk, 2007)
We can express the above story qualitatively by using the terminology as de-
scribed in Volk (2007). Using Fig. 1.6 as our reference image we can make
a telescope exposure starting with nod position A and define the Flux, Fchop
nod ,
background B and telescope background T. With this we can write the flux for
Nod A and Chop A as,
A=B1+TA+S. (1.1)
Similarly we can express the flux for Nod A, Chop B as,
A=B2+TB. (1.2)
These two frames are identical apart from the fact that chop position B has no
source in it and that TBdiffers from TAbecause the beam has moved over the
primary mirror. However, this effect is small with TBTAand the back-
grounds are of similar magnitude too. Next we can look at the flux for Chop
A and Chop B for Nod B, which are given by,
Version of July 22, 2016– Created July 22, 2016 - 20:26
1.2 Classical Calibrations 9
Figure 1.7: A classical chop/nod reduction shown step by step. We start with two pairs
of chop exposures which we can subtract in order to remove most of the background.
The remaining two images have nearly the same aberrations but have inverted symbols
for the source. This means that subtracting them removes the background and doubles
the source.
B=B4+TB+S. (1.4)
We see that these two equations are nearly identical to the two above, with only
the sky backgrounds being different and the source being on the other chop.
Subtracting the two chop positions for both nods will leave us with two chop
difference frames, one for each nod position. We will call these frames ‘nA’ and
=B12 +TAB +S
Version of July 22, 2016– Created July 22, 2016 - 20:26
10 Introduction
Figure 1.8: The resulting image obtained after executing the plan as described in Fig.
1.6 on observations of the standard star HR 2652.
=B34 +TAB S
Now if the background and telescope terms would have been small enough,
then we would have just the source left, but we have discussed above that this
is not the case, as can be seen in Fig. 1.7. These terms can still be of the order
of magnitude or brighter than the source and we need to find a way to rid our-
selves of them. Therefore we want to subtract the two chop difference frames
to get
nA nB =B12 +TAB +S(B34 +T0
=B12 B34 +2S
Here we see that the telescope background is gone and if taken with a short
enough time in between exposures, we can assume that the background hardly
changes between exposures and thus can say that the two background terms
cancel out, leaving us with a flat background and our source.
The noise of this image is composed of shotnoise from the photon flux itself, 1/f
noise from background variations, readnoise and the detector. 1/f noise is quite
Version of July 22, 2016– Created July 22, 2016 - 20:26
1.3 Challenges 11
hard to model, but if we assume that chopping and nodding takes care of it, this
allows us to reach background limited performance. We can estimate a value
for this noise if we take the example value of 530 Jy/arcsec2for the background
that was given above and apply the flux identity described in Oke (1974) This
will give us a flux of 8.1·1010 photons/m2/s/µm/arcsec2at 8.7 µm. If we then
assume a 0.7 micron bandpass, 0.045”x0.045” pixels, a throughput of 0.7 and a
Quantum efficiency of 70%, we will be left with 8.1·108electrons/s/pixel. For
a single chop frame the pipeline averages 7 exposures of 87.5ms in total, this
gives us an average amount of 10.1·106electrons/pixel, giving us a shot noise
of 3.2·103electrons/pixel. This means that our image is shot noise limited
if the other noise components are below that. Which is true for the 150 elec-
trons/pixel RMS noise introduced by the detector (Ives et al., 2012), and we
assume it to be above the 1/f noise introduced by the atmosphere if the chop-
ping takes care of it. So if we know that we are shot noise limited and if we
know the counts of all the individual flux components and assume that they are
all Poisson distributed, then we can express the noise limit as the square root
of the sum of these components. This is true for all background components
but not for the source, however its contribution to the counts is very small com-
pared to the other components (like the read noise), so we can neglect it and
write the noise for a single frame as.
σpB1+B2+B3+B4+2TA+2TB=N. (1.8)
With N the amount of counts in electrons on our images.
1.3 Challenges
METIS, the ‘Mid-infrared ELT Imager and Spectrograph’, is the dedicated MIR
instrument, working in between 3 and 19 µm, that has been selected by ESO as
a first generation E-ELT instrument. This instrument will be at the forefront of
MIR astronomy, rivalled only by the Spitzer Space Telescope and the upcom-
ing James Webb Space Telescope. However it does have a couple of advantages
over space telescopes, the main ones being the higher angular resolution due to
the larger aperture, a much higher spectral resolution than is currently possible
due to weight limitations for space telescopes and the fact that the instrument
is easy to reach and maintain during its operation. (Brandl et al., 2008)
The challenges that come with this telescope are mostly because of the sheer
size of the 39m primary and the mirrors that come after it. The secondary mir-
ror is too large (4.2m) and heavy to use as chopping mirror and because of that
Version of July 22, 2016– Created July 22, 2016 - 20:26
12 Introduction
it has been decided to make an internal chopper in the pupil plane of the in-
strument.(Paalvast et al., 2014) METIS would be located at the nasmyth focus
(see Fig. 1.9) of the E-ELT and have 6 warm and several cold mirrors before
the chopper.(One of which would be the AO deformable mirror) This will
introduce a larger than usual thermal footprint from the telescope, which will
be modelled as a 15% greybody compared to 9% in the VLT.
Now that we know the challenges that come with observing in the MIR and the
solutions that are used, one might wonder why we bother with this at all. The
answer to this question is simple, the classical solutions will not work for the
European Extremely Large telescope (E-ELT).(Pantin, 2015) Due to the fact that
the E-ELT will be a dynamical structure that continously reajusts to keep up the
alignment and image quality. This makes it impossible to nod properly, as the
symmetry cannot be kept, meaning that using classical methods we will not be
able to get down further than a few percent of the original background, which
is still too high for observing. For this reason we need to find alternative ways
of calibrating the images.
Unfortunately we do not yet properly understand the origin and magnitude
of the observed systematic effects, which makes it very difficult to construct
reliable thermal models of the E-ELT. The currently agreed upon explanation
blames the beam motion on the primary mirror for the residuals that can be
seen in chop difference frames, which is odd as a surface in the pupil plane
should not create such structures. Theories exist about the cryostat window
being the cause of the abberations (Brandl, 2015), but this also has never been
Because of this we want to gain a better quantitative understanding of the ther-
mal background and its uniformity as a function of various variables and cali-
bration techniques. On top of this we want to develop, implement and analyse
new background reduction strategies that do not require nodding.
To do this we will request technical time at two large telescopes capable of ob-
serving in the MIR, the VLT and the Grantecan (GTC). These observations will
be specially developed to test various aspects of the background, as can be read
in Chapter 2. The obtained data will be subjected to various tests and reduction
methods that are described in Chapter 3 and their outcome in Chapter 4. Finally
conclusions will be drawn in Chapter 5.
Clasically there would only be one, namely M1.
Version of July 22, 2016– Created July 22, 2016 - 20:26
1.3 Challenges 13
Figure 1.9: A raytrace of the E-ELT with a beam of light going to the Nasmyth platform.
The METIS chopper will be inside of the instrument, meaning that the light needs to
reflect off 6 warm mirrors before hitting the chopper. This is significantly more than in
classical setups.
Version of July 22, 2016– Created July 22, 2016 - 20:26
Chapter 2
2.1 Technical time
To better understand the origin of the various contributions of the thermal back-
ground we have requested observing time on both the VLT and the GTC. With
the former we would want to observe while using the VISIR instrument, which
is located in UT3 (Melipal) and similarly for the GTC we want to observe with
their IR instrument, the CANARICAM. The aim of the proposal is to obtain a
coherent dataset, tailored especially for this purpose, to better understand the
thermal background and its uniformity under different observing schemes.
For these observations we propose to take a set of ’cheap’ exposures that should
take no more than two hours in total and can be done during twilight and does
not require good seeing conditions. This would make our tests non invasive in
the existing schedule and should be doable in the time frame given for a master
While it is desirable to have an IR point source in the images, we also wish to
have a stable telescope with no movement other than that of the chopper. Track-
ing and derotation should be turned off, which makes it hard to look at a source.
For this reason we have chosen to look at empty sky with a given Alt and Az=0.
We know that most of these observations can be found in the archive, but we
think that it is essential to have a coherent data set with the same boundary con-
ditions and instrument settings, otherwise we will have too many variables and
never be able to pinpoint the exact reason. This dataset taken on both telescopes
should be a good first step in identifying the origins of the thermal background.
Throughout the observing procedure we have tried to keep doing the same set
Version of July 22, 2016– Created July 22, 2016 - 20:26
16 Data
Chop Amplitude 2,5,10,15,30 arcseconds
Chop Frequency 0.5,1,2 Hz
Chop Direction N-S, E-W
Filters J8.9, J7.9, SIV
Altitude 25, 30 ,45, 90 degrees ALT
Derotator angle 0, 90 degrees
Table 2.1: All variables that we want to vary during our tests with the values that we
wish to vary.
of observations and control what we vary as strictly as possible. For this rea-
son we have chosen to use a certain nodding pattern that shall henceforth be
referred to as the ’nodding cross’. This is a set of eight science observations as
displayed in Fig. 2.1 that is designed to allow us to make different nodding
and chopping patterns in different directions, allowing us to get the maximum
amount of information from a minimal amount of observations. The cross is
defined as a collection of standard chop/nod operations where we start chop-
ping at pointing A in direction B, then nod to B and continue nodding in the
same direction. Then we repeat the block but nod to position C in the direction
of C, etc. Most of the observations in this observing run will follow this ’nod-
ding cross’ template. During these tests we will keep the directions the same
but vary chopping throw and frequency.
Furthermore we want to add some observations in other filters and with differ-
ent rotation positions on the derotator to see if the residuals respond to that.
2.2 Observations
The technical time observations for both the VLT as the GRANTECAN have
resulted in multiple datasets with different variations. The original observing
plan proposed to make cross shaped nodding patterns with three different chop
throws (5”, 10”, 30”) and two different frequencies (1Hz, 4Hz). All of these tests
were done while using the J8.9 filter for the VLT and the Si2-8.7 filter for the
GTC, which both are similar and have relatively broad filter on the lower end
of the N-band. These filters were chosen because they were the closest filters
that have significant contributions from both the telescope and the atmosphere.
These tests were taken during twilight with reasonable conditions and were also
partially repeated with the dome closed, as a comparison without atmosphere.
Version of July 22, 2016– Created July 22, 2016 - 20:26
2.2 Observations 17
Figure 2.1: The so called ’nodding cross’, showing how the nods will be set. The chop-
ping is always parallel to the nodding direction. The cross starts with chopping from
A to B, then nodding to B and chopping in the same direction. We call this the AB arm.
After this we nod back to the A position and begin chopping in the AC direction. In
this way we work trough all four arms.
Our field is 38.0 x 38.0 arcsec2, but will be windowed down to a region of in-
terest of the central 31.5 x 31.5 arcsec2. In all cases we will chop parallel to the
nodding direction, have no random jitter, use only 1 nodding cycle and have a
60s total integration time in high gain mode.The only things that will change
are the chopping position angle, the frequency and the chopping amplitude. In
addition we have turned off the telescope tracking and observe an empty patch
of sky during most observations.
If everything would have gone well we would have four sets of data, two from
each telescope. This would be the entire observing plan taken with an open and
closed dome. However it is not as straightforward to obtain and retain observ-
ing time as it is with science observations. For both the VLT and the GTC the
observing date has been pushed back many times due to a myriad of reasons
including weather and priority going to more important technical tests. For
the VLT we eventually set an observing date at the end of January where three
hours were allotted to this project, however when the observing began we got
the message that the time was cut down to 1 hour. For this reason we had to
make on the spot decisions on what tests to do and what tests not to do. In the
This means 20 e/ADU
Version of July 22, 2016– Created July 22, 2016 - 20:26
18 Data
end we split the time into two sets, one with a low (88.9 degrees ALT) airmass
and one with high (25 degrees ALT) airmass. The first set consisted of three
nodding crosses with a throw of 5,10,30 arcsec at 4Hz, one nodding cross with
a 30 arcsec throw at 1Hz and a filter test. The latter consisted of two nodding
crosses with a throw of 5 and 30 arcsec at 4Hz and one cross with a 30 arcsec
throw at 1Hz. We also did a filter and derotator test at this ALT. Unfortunately
the exposure times of the observations were not altered, so all the high airmass
observations have passed their maximum saturation level and are therefore not
usable. Because the derotator test was an important test for us, we have re-
quested for it to be done once more and it has been. We have received the data
plus a half nodding cross on the 16th of june. (Taken with the same settings
and position.) We also got a nodding cross taken at two airmasses of 30 and 45
degrees ALT with a random pointing and rotation. These images are not ideal,
but better than nothing. Finally we also got a nodding cross centred on a source,
namely the star HR 2652. This was done with tracking but at a high altitude.
With the Grantecan we had even less luck. We got a closed dome dataset with
three crosses with throws of 5,10,30 arcsec at 3.6 Hz. We got the same data for
1Hz, a derotator test and a time stability test. The rest of the observations were
scheduled to be taken later that month, but due to faults in the instrumentation
the CANARICAM needed to be removed from the telescope for at least a year,
making further observations impossible within the time frame of this project.
A summary of all of these observations can be found in table 2.2.
The GCT choppers maximum frequency.
Version of July 22, 2016– Created July 22, 2016 - 20:26
2.2 Observations 19
Summary of observations made for the METIS calibration
Date Telescope Type Amplitude [”] Freq [Hz] Filter tint ALT [Deg] Burst
3-11-’15 VLT [C] Cross 8.0 3.0 NeII 0.0278 88.9 N
28-12-’15 VLT [O] Cross 5.0 4.0 J8.9 0.0125 45 Y
29-12-’15 VLT [O] Cross 5.0 4.0 J8.9 0.0125 30 Y
23-01-’16 GTC [C] Cross 5.0 3.6 Si2-8.7 0.07 80.0 N
23-01-’16 GTC [C] Cross 10.0 3.6 Si2-8.7 0.07 80.0 N
23-01-’16 GTC [C] Cross 30.0 3.6 Si2-8.7 0.07 80.0 N
23-01-’16 GTC [C] Cross 5.0 1.0 Si2-8.7 0.07 80.0 N
23-01-’16 GTC [C] Cross 10.0 1.0 Si2-8.7 0.07 80.0 N
23-01-’16 GTC [C] Cross 30.0 1.0 Si2-8.7 0.07 80.0 N
23-01-’16 GTC [C] Cross 5.0 3.6 Si2-8.7 0.07 80.0 N
23-01-’16 GTC [C] Derot 30 3.6 Si2-8.7 0.07 80.0 N
22-03-’16 VLT [O] Cross 5.0 4.0 J8.9 0.0125 88.9 Y
22-03-’16 VLT [O] Cross 10.0 4.0 J8.9 0.0125 88.9 Y
22-03-’16 VLT [O] Cross 30.0 4.0 J8.9 0.0125 88.9 Y
22-03-’16 VLT [O] Cross 30.0 1.0 J8.9 0.0125 88.9 Y
22-03-’16 VLT [O] Derot 30.0 4.0 J8.9 0.0125 25 Y
22-03-’16 VLT [O] Filter 30.0 4.0 J7.9, SIV 0.0125 25 Y
22-03-’16 VLT [O] Cross 5.0 4.0 J8.9 0.0125 25 Y
22-03-’16 VLT [O] Cross 30.0 4.0 J8.9 0.0125 25 Y
22-03-’16 VLT [O] Filter 30.0 4.0 J7.9, SIV 0.0125 88.9 Y
24-03-’16 VLT [O] Star 10.0 4.0 J8.9 0.0125 Y
16-06-’16 VLT [O] Derot 10.0 4.0 J8.9 0.0125 88.9 Y
16-06-’16 VLT [O] Time 10.0 4.0 J8.9 0.0125 88.9 Y
Table 2.2: A table showing all obtained observations for both the VLT and the GTC. All files are grouped together into
relevant blocks to avoid clutter, these blocks can be seen in the ’Type’ column. With ’Cross’ being the standard nodding
cross, ’Derot’ the derotator test, ’Filter’ the filter test, Time a set of identical chopping patterns with set times in between
and ’Star’ being a standard chopping cross executed on a source. In the ’telescope’ column, we can see the telescope that
has been used and whether the dome is [O]pen or [C]losed. Observations with striked out dates are oversaturated and not
Version of July 22, 2016– Created July 22, 2016 - 20:26
Chapter 3
Figure 3.1: In this illustration we see four chop difference frames corresponding to
the nodding cross of Fig. 2.1. The background left after the chop subtraction is still
significant compared to average sources and very high compared to a nod subtracted
frame. Such a frame can be seen in the lower right corner of the image.
Since we will not be able to nod with the E-ELT, we will have to find ways to
deal with the chopped difference residuals that are left after a chop subtraction.
These residuals are often in the order of a few percent of the original back-
Version of July 22, 2016– Created July 22, 2016 - 20:26
22 Methods
ground, but since the source is also of this magnitude and often less, we cannot
just ignore it. In Fig. 3.1 we can see the residuals of the chopped difference
frames for the four directions of the nodding cross. Here it is clear that the
background residuals are far above the shot noise limit and that these images
cannot be used without further reduction due to the gradients and small scale
structure inside them. For comparison we have added a nod difference frame
in the lower right corner, showing what we are lacking without proper nodding
and what we want to achieve with our alternative methods.
In this chapter we introduce different background reduction techniques and
compare them to classical chopping and nodding. We begin by studying the
residuals as a function of four variables. Namely the chop throw, direction, fre-
quency and telescope elevation. Afterwards we test the residual stability over
time and then we choose a subset of these variables and explore new back-
ground reduction methods.
After obtaining this data we have to define methods to quantify and display
background quality of the reduced frames after different background reduction
techniques. Primarily we want to see if the background is flat and limited by
shot noise. This means that we want the mean to be zero, without any gradients
or higher order structures and that the noise should have a standard deviation
of around a theoretical minimum of
σ=electrons (3.1)
But since we have our counts in ADU, we should convert them by using the con-
version factor found in the image header. This value is given to be 20 e/ADU.
σ=220 ·ADU
Because we subtract two frames the noise rises with 2 and because we average
over n chop pairs, the noise goes down with that number. Filling in the ADUs
from the images and the amount of frames we get σ=2.16 ADU.
However when comparing these numbers to our calculations from before, we
find that there we would expect to have 1262 e/ADU rather than 20. Repeat-
ing the calculation with this number gives us σ=0.27 ADU. Neither of these
numbers seem to be correct, as our chop and nod subtracted frames have a STD
that is lower than 2.16 and the e/ADU rate from our calculations is too high.
Therefore we assume that the images are shotnoise limited and that the shot-
noise limit is 1.1 ADU.
Version of July 22, 2016– Created July 22, 2016 - 20:26
Figure 3.2: With the Quadrant test we test the flatness of the reduced images. To do
this we cut them into four smaller squares and calculate the mean and STD of every
quadrant. The variation in these values tells us about the flatness on both large and
small spatial scales.
We attempt to quantify the gradient and higher order structures by cutting the
images into four quadrants and calculating the mean and standard deviation of
every quadrant. (See Fig. 3.2.) Large scale variations (like a gradient) will be
shown by a large spread in the four means while small scale structures will be
described by a larger standard deviation. To illustrate the method we prepared
two test images, one with a large overall gradient and one with smaller higher
order structures. Both images can be seen in Fig. 3.3 and in Table 3.1 as the
gradient test and the structure test.
Version of July 22, 2016– Created July 22, 2016 - 20:26
24 Methods
Figure 3.3: Two test images designed to illustrate the working of the quadrant test. The
first image shows a gradient moving smoothly from top to bottom, while the second
image shows a high order repeating pattern. The gradient will be picked up by the
mean while the STD is sensitive to small scale variations.
Data counts µ1µ2µ3µ4µ±σµ σ
Gradient Test 2.7 1.36 1.36 4.08 4.08 2.72 ±1.36 0.79
Structure Test 20.02 20.01 20.01 20.01 20.02 20.02 ±0.007 7.08
Table 3.1: Results of the Quadrant test for the two test datasets that are illustrated in
Fig. 3.3. This shows the strength and weakness of looking at the mean and STD of
the quadrants and how using both can help as a first look. The first column shows the
average amount of counts in ADU of the image. The next four columns show the mean
value of the four quadrants, followed by the mean of these four means ±the spread on
the values. The final value is the mean STD of the four quadrants.
For the gradient test we can see that the means are different for the top two
and bottom two quadrants, suggesting a vertical gradient in the data. Because
the gradient goes up very smoothly the standard deviation will be small and not
vary across the four quadrants. In the structure test we can see that the opposite
is true, the means are very close to each other, suggesting a flat image, while
the standard deviation is very big. This test is not perfect, but it can indicate
complete outliers and methods that won’t work at all. Once a sufficiently flat
image has been produced we can compare the standard deviation of the entire
image with the theoretical value that we derived above. Besides this it is also
important to do a visual inspection of the resulting image and to repeat the
process on a binned down image.
Version of July 22, 2016– Created July 22, 2016 - 20:26
3.1 Classical Chopping 25
3.1 Classical Chopping
To get a good feeling for the numbers and a baseline as reference, we want to
calculate the means and STDs for a large array of classical nodding patterns.
Trying to extract the maximum amount of data from our observations we com-
pare classical nod patterns with three different chop throws, four directions,
two frequencies and two airmasses. With this we hope to find a pattern in the
brightness of the background residuals and with that an optimal chopping strat-
egy and reduction method. We have taken the nod difference frames of obser-
vations with one of these variables being varied every time and carried out the
quadrant test on these frames. The results can be seen in Table 3.2.
3.2 Chopping without nodding
Figure 3.4: Chop difference frames for the AC nodding direction for 5”, 10” and 30”
chop throws with the same scaling.
Since we cannot nod and have to suppress the background by testing new re-
duction strategies, we should first observe the individual difference frames to
understand them better. In Fig. 3.1 we can see the chop difference frames for
all of the four nodding directions. It is clear that the pattern, both high and
low order, is different in every direction, but similarities can be seen in oppo-
site facing chops. This raises the question of whether or not we can remove the
background by using these symmetries. When looking at a single chop direc-
tion as a function of chop throw. (Fig. 3.4) We can see that both the high and
low order aberrations seem to stay the same, but change in amplitude. This
brings up the question of stability. Do the aberrations that gain in brightness
as a function of throw grow linearly and stay identical? We can easily find the
scale factors between the frames by comparing two regions where the gradient
is largest and plotting the difference in counts between the two regions. This
has been done in Fig. 3.5.
Version of July 22, 2016– Created July 22, 2016 - 20:26
26 Methods
Figure 3.5: To test the linearity of the background we will compare the difference in
counts in the two black boxes for three chop throws of 5”,10”,30”.
3.2.1 Stability over time
This brings up the question of stability of time, as there is time between the
frames. We know that the sky changes in the order of minutes and the telescope
stays stable for a longer time. There is a time difference of about 15 minutes
between each of the three exposures and therefore 30 minutes between the first
and last exposure. If the images scale linearly this either means that both back-
grounds did not change in this timespan or perhaps that several effects cancel
On top of that stability of the residuals over time is very important if we want to
stack multiple exposures and lower the noise that way. In Pantin (2015) we see
that while the S/N of nod subtracted frames improves with longer integrations,
the S/N of chop subtracted frames stays roughly the same for integrations over
1 minute. It is also mentioned that the aberrations introduced by the telescope
don’t change rapidly for about 2-3 hours, potentially, allowing us to make some
sort of flat that we can take out of our observations. However, when put to the
test, we can see in Fig. 3.6 that the background changes very strongly over a
period of two hours, going from lines at about 30oto nearly horizontal patterns.
This would suggest that our frames which have a t of 30 minutes at most
should be fine, but it can never hurt to repeat this test for ourselves. This can be
done by using observations with identical chop throw, frequency and direction.
Version of July 22, 2016– Created July 22, 2016 - 20:26
3.2 Chopping without nodding 27
Figure 3.6: The two top images show a chop difference frame for a VISIR standard star.
The two images are taken one hour apart, illustrating the strong changes in background
residuals. The lower two images are nod difference images. The left one is taken with
minimal time between the two nod frames and the second nod difference frame has 1
hour between the two nod frames. A clear difference is visible in effectiveness of the
subtraction. Pantin (2015).
From our data we can find three identical datasets with 6 minutes between each
set and one that has been taken about 3 months later. Therefore we can explore
the stability of the patterns over time within the range covered by Pantin and
far beyond it. This will be done in the same way as before, namely by use of the
quadrant test on the nod subtracted frames. However instead of subtracting the
datasets corresponding nod frame, we take the corresponding nod frame of the
dataset that was taken at a later time. (see Fig. 3.7a.) Following the theory of
Pantin (2015), this should give us four flat images at dt=0,6,12,18 min and one
completely mismatched frame at dt=3 months.
A different way to test this linearity is to use asymmetrical nodding patterns
Version of July 22, 2016– Created July 22, 2016 - 20:26
28 Methods
Figure 3.7: a) An illustration of classical chop/nod reductions with a certain time in
between observations. We simulate this by matching two different pairs of observa-
tions with the same chop throw and chop direction. b) The same process as seen in (a)
but here we match frames with a different chop throw. We attempt to correct for the
residual amplitude by multiplying the first term by a scalefactor.
where we pair two nod frames with different chop throws. So for example we
take nod position A in the AC direction with a 5” throw and subtract the linearly
scaled nod position B with a 10” throw in the AC direction. (see Fig. 3.7b.) These
frames are taken 15 to 30 minutes apart from one another so temporal effects,
if negligible, will not play a role here. The results of both tests can be found in
table 3.1.
3.2.2 Polynomial fitting
Looking at the structure of the aberrations as seen in Fig. 3.1 and Fig. 3.4,
one could imagine that a simple polynomial fit might be enough to remove all
large structures from the background, leaving only the higher order aberration.
This would not be a good method for images with large structures, but could
perhaps be useful for observations with small chop amplitudes. A first attempt
would be to fit a 1st order polynomial to take care of the gradient and leave the
sources intact. Therefore it is interesting to see if the fringes on the images are
still dominant over the poisson noise. To test this theory we have fitted a first
order polynomial to three chop difference frames with a chop throw of 5, 10
Version of July 22, 2016– Created July 22, 2016 - 20:26
3.2 Chopping without nodding 29
and 30”. These images have then been subjected to the Quadrant test and the
obtained results can be seen in 3.2. This method by itself will probably not be the
best method of background subtraction, but could prove useful in combination
with other methods.
3.2.3 Principal Component Analysis
Principal component analysis (PCA) is a statistical procedure that orthogonalises
a dataset of observations into a set of linearly uncorrelated variables that we call
principal components. We always get the same number of variables as the num-
ber of datasets that are put in, with the first few components being dominant
and the rest negligible. The transformation also sorts the component on their
magnitude, meaning that the component that introduces the biggest variability
to the data will be first, the biggest contribution after that will be second and
so on. Using this method we end up with a vector of uncorrelated orthogonal
components which are eigenvectors of the symmetric covariance matrix. This
method is sensitive to the scaling of the original variables. (Pearson, 1901)
This method is not useful for reducing the background, as it will subtract out
the source, but we can plug in all 24 frames from the three nod crosses of 5”,10”
and 30” throw to see if the images share certain structures.
Version of July 22, 2016– Created July 22, 2016 - 20:26
30 Methods
Data counts µ1µ2µ3µ4µ±σµ σ
Theoretical 8000 0 0 0 0 1.1 ±0 0
Classical 5” AC 7990 0.5 0.5 0.60 0.60 0.60 ±0.04 1.19
Classical 10” AC 8076 0.2 0.16 0.17 0.17 0.18 ±0.02 1.21
Classical 30” AC 8160 0.71 0.72 0.68 0.62 0.68 ±0.04 1.21
Classical 30” AD 8189 0.32 0.30 0.30 0.38 0.33 ±0.03 1.20
Classical 30” AE 8033 0.31 0.26 0.19 0.15 0.23±0.06 1.19
Classical 30” AC 1Hz 7173 4.12 4.02 4.24 4.14 4.13 ±0.08 1.64
Classical 30” Airmass 7305 1.32 1.39 1.41 1.39 1.38±0.03 1.13
Classical 30” dt=0 8160 0.71 0.72 0.68 0.62 0.68 ±0.04 1.21
Classical 30” dt=6 min 6766 0.17 0.14 0.18 0.08 0.14 ±0.04 1.10
Classical 30” dt=15 min 6670 0.33 0.27 0.29 0.18 0.27±0.05 1.10
Classical 30” dt=84 days 6670 0.15 0.38 0. 0.28 0.2±0.14 1.17
Asymetric 5-10” AC 7996 1.56 2.04 1.33 1.78 1.68 ±0.26 1.22
Asymetric 10-30” AC 7992 3.65 5.71 2.29 4.61 4.07 ±1.26 1.5
Asymetric 5-30” AC 7998 5.41 7.91 3.80 6.56 5.92±1.51 1.64
Polynomial fit 5” 7990 0.01 0.02 0.02 0.01 0.01 ±0.01 0.85
Polynomial fit 10” 8123 0.03 0.05 0.02 0.05 0. ±0.04 0.88
Polynomial fit 30” 8076 0.12 0.06 0.15 0.09 0. ±0.11 1.17
ICA 30” AC-AE 8160 -9.38 -15.82 -7.47 -12.74 -11.35 ±3.2 2.5
ICA 30” AC-AE - polyfit 8160 0.02 0.03 0.07 -0.08 0.01 ±0.05 1.48
ICA 10” AC-AE - polyfit 8076 0.01 -0. 0.02 0.03 0.02 ±0.01 1.2
ICA 5” AC-AE -polyfit 7990 0.01 0. 0.01 0.02 0.01 ±0.01 1.19
Table 3.2: In the first section we see the results of the Quadrant test for varying chop
throw, direction, frequency and airmass. In the second part we show the Quadrant
test for different times in between nods and asymmetrical chopping/nodding. In the
last part we can see the Quadrant test results for the other reduction methods. The
first column shows the average amount of counts in ADU of the image. The next four
columns show the mean value of the four quadrants, followed by the mean of these
four means ±the spread on the values. The final value is the mean STD of the four
Version of July 22, 2016– Created July 22, 2016 - 20:26
3.2 Chopping without nodding 31
3.2.4 Multidirectional chopping
Figure 3.8: Instead of subtracting the chop difference frames obtained from
NodABChopAfrom NodABChopBwe now add the chop difference frames obtained
from NodAB ChopAto NodADChopA.
The final method requires us to chop in different directions to simulate nodding
without actually moving the telescope. This method uses the fact that opposing
chop directions seem to be eachothers inverse and can remove close to all low
order aberrations by simply adding the two difference frames. So we would
be adding the chop difference frame obtained from Nod position A in the AB
direction to the chop difference frame obtained from nod position A in the AD
direction. (see Fig. 3.8) This is not just a hunch based on symmetries seen by
eye, as mathematically this also makes sense, as shown below.
We start in the much same way as we did with Eq. 1.1 and define two chop
observations Aand Bfor our two nod directions AB and AD. We change the
naming convention slightly to preserve clarity of which chop and nod frame we
are working with. Therefore the flux F will have a subscript with a certain nod
nfollowed by the chop position in the superscript. Besides this we introduce an
offset term Othat is defined as the difference between the telescope background
of two chops. Now we can define the following four pointings.
Here we can assume OB≈ −OAbecause we expect them to be roughly equal
and that the reversed chopping direction also inverts the sign of the aberra-
tions. Knowing this we can combine the four chops into two chopped difference
frames, which we name P1 and P2.
AC =B3+TA+SB4TAOB=P2(3.4)
Next instead of subtracting them, we have to add these two frames in order to
counter the minus sign introduced by reversing the chop direction. This gives
Version of July 22, 2016– Created July 22, 2016 - 20:26
32 Methods
us our source and six background terms that can be paired and subtracted for
each other, leaving the difference flux.
P1+P2=B12 +B34 OAB +2S(3.5)
The noise is given in the same way as for the classical chopping, namely by
the square root of the sum of all the components, again with the source being
neglected due to its amplitude.
Just like with the offset term we can assume dB34 ≈ −dB12, meaning that we
end up with a smaller sky and telescope background than in the separate chop
difference frames, but a higher one than when applying classical chopping and
In table 3.2 we have placed the results of the quadrant test completed on a frame
resulting from inverse chop addition and three frames resulting from a combi-
nation of this method and a polynomial fit.
3.3 Atmosphere VS Telescope
3.3.1 Wavelength Dependencies
To find out which components are contributed by the sky and which by the tele-
scope we need to find a way to vary the amount of light given by either of the
two components. While this is not possible with our currently used data, we
could remedy this by observing in different filters. Therefore we have made ob-
servations in two more filters to see if this can be confirmed. (See Fig. 3.10.) The
first filter was the J7.9 filter which was chosen because most of the background
comes from the atmosphere, while the SIV filter is chosen for the opposite rea-
son of having a background dominated by the telescope. With this test and
the closed dome observations we hope to be able to discover what effects are
caused by the atmosphere and what effects are caused by the telescope. If the
difference is big enough, it should even be possible to see the variations by eye.
If this is not the case, then we can use a more theoretical approach and write
down these two filters as the sum of the two backgrounds. Now assuming that
one of the backgrounds dominates completely in these images, we can say that,
=B1Sky +B1Telescope
B2Sky +B2Telescope B1Sky
. (3.7)
Version of July 22, 2016– Created July 22, 2016 - 20:26
3.3 Atmosphere VS Telescope 33
Figure 3.9: Inverse chop addition starts similarly to a classical chop/nod reduction by
chopping on and off from the source in a given direction. Then instead of nodding the
telescope, we repeat the same observations but with the chopping angle +180o. This
gives us two frames that apart from the source are each others mirror image. Adding
these two images removes the higher order background residuals and leaves us with a
twice as bright source and a gradient. After fitting and subtracting this gradient we are
left with a flat image comparable to a classically reduced chop/nod image.
Version of July 22, 2016– Created July 22, 2016 - 20:26
34 Methods
Figure 3.10: Three filters plotted over the sky transparancy and a blackbody curve at
288K. The filters are chosen in such a way that the first (FAtm) sees mostly the atmo-
sphere and the third (FTel) sees mostly the telescope radiation while the second is in
between.(FMid) These filter are named J7.9, SIV, J8.9 respectively.
Here the first filter is dominated by sky background and is approximated by
assuming B1Sky B1Telescope while the second filter is quite the opposite and
it is assumed that B2Sky B2Telescope. With these ideal filters we would see
just a gradient on one of the two filters and just the higher order pattern on the
other. Dividing these two filters will result in an image that has certain charac-
teristics. If we divide the pattern by the gradient we will see the pattern with
an overlayed gradient but if we flip the fraction we will divide the gradient by
the pattern and see a gradient inside of the pattern with a smooth background.
This is illustrated in Fig. 4.8, where we took an arbitrary gradient and a higher
order structure in the form of three bars of uniform brightness. Looking at both
of these fractions we should be able to distinguish whether or not the relation is
true for the two filters where B2Sky B2Telescope and B1Sky B1Telescope
Unfortunately, we do not posses these ideal filters and all exposures have non
negligible contributions of both backgrounds. However the amplitude of both
components varies, so we should still be able to distinguish a pattern as shown
in Fig. 4.8.
Version of July 22, 2016– Created July 22, 2016 - 20:26
3.3 Atmosphere VS Telescope 35
Figure 3.11: The derotator test does a two direction nod and then two nods in the same
direction but with a 90oturn in the derotator. On the detector this will seem the same
as doing the same motion twice.
3.3.2 Derotator Test
The second test is the so called derotator test, where we try to distinguish be-
tween the effects caused by the telescope and everything that happens inside
of the instrument. This is achieved by making two standard chop observations
in an L shaped pattern, as seen in Fig. 3.11, followed by a similar observations
in which we chop up, turn the instrument by 90 degrees and chop up again.
From the detectors point of view these two manoeuvres were both the same,
namely an L shaped movement. This means that if the cryostat window or any-
thing inside of the instrument is not rotationally symmetrical, we should see a
difference between the two observations. Any instrument related asymmetries
should show up as variations between the AB and
B frames, while telescope
related variations would be the same in all observations taken in the same chop
3.3.3 Effects as a function of airmass
Once we know that our residuals are stable over time and assume that the
higher order aberrations are caused by the telescope angle, the question might
come up of how this scales when the telescope moves and looks through differ-
ent airmass. First we would assume that as there is more atmosphere to look
through, the gradient will become stronger with respect to the higher order
structures, but it would not be strange to assume that the higher order patterns
stay the same. Due to the fact that most of our low airmass data is useless we
Version of July 22, 2016– Created July 22, 2016 - 20:26
36 Methods
are forced to make do with one dataset taken at 45 and 30 degrees ALT with a
random telescope rotation angle of 164 degrees. Due to this the patterns will
probably differ from the ones in the other tests, but we should be able to com-
pare the frames within this dataset to see the stability as a function of airmass.
3.3.4 Closed dome
Observing the inside of a closed dome should help us disentangle the back-
grounds even further, as we will not see the atmosphere and only the telescope.
Therefore doing these tests should give us more insight in these components.
Version of July 22, 2016– Created July 22, 2016 - 20:26
Chapter 4
In this chapter we will discuss the perceived effects that occur when changing
the variables discussed above, analyse the outcome of the quadrant test and
other methods to find the optimal chopping strategies and pinpoint potential
causes of aberrations.
We start by looking at table 3.2 to weed out all reduction methods that clearly do
not pass the test. The first candidate for this is the classical chop/nod reduction
at 1Hz, this frame has a significantly higher noise than its 4Hz counterpart and
suggests that any other tests at lower chopping frequencies would only raise the
noise floor, but also might mean that a higher frequency could lower it further.
Second is the polynomial fit, these reductions have taken out the overall gra-
dient but left the higher order structure, leaving an image with a low STD but
lacking in flatness for larger chops, as can be seen in Fig. 4.1. This could be a
viable solution for chops with a throw of 5” or less.
Figure 4.1: Background reduction done by a simple first order polynomial fit, removing
only the gradient and leaving the rest.
Version of July 22, 2016– Created July 22, 2016 - 20:26
38 Results
Now that we know what data to not pay attention too, we can focus on the data
that is left. The first question that is in want of an answer is the linearity of the
background as a function of throw.
4.1 Effects as a function of Throw
Figure 4.2: We compare two corners of a chop difference frame by summing the counts
inside of the black squares and subtracting these two values to see if the patterns scale
linearly. In the left image we see a chop difference frame in the AC direction with the
two regions marked and in the right image we see the obtained values plotted as a
function of throw.
In the last chapter we plotted the difference in counts between the maximum
and minimum values of the gradient. When looking at Fig. 4.2, we see that the
gradient seems to be stable over both time and throw. (As there is 15 minutes
between each of the four frames.) However, we cannot assume that the increase
is linear without also looking at the chop difference frames taken in other direc-
tions. Using the corner test on these data that we plotted in Fig. 4.3. The errors
are given as the STD of the two used regions with a slightly larger samplebox,
however they are so small compared to the data points that we cannot see them,
meaning that there is no linear relation between chop throw and or time.
To learn more about which of the two are the main contributor to the non-
linearity we can look into the asymmetrical chop reduction. In Tab. 3.2 we
can see that these frames had by far the highest STD and the biggest variation
in mean quadrant value. We still cannot say if this is due to time variations or
One of the ways of seeing this is by using PCA on all 8 frames of a nodding cross
and the combined frames of the three crosses that we have for 5”,10”,30”. Un-
Version of July 22, 2016– Created July 22, 2016 - 20:26
4.2 Residual stability over time 39
Figure 4.3: A test of linearity for all four directions of the nodding cross as a function
of chop throw. The top left image shows the AC direction and the top right images
shows the AD direction. The lower left and right images correspond to the AE and AB
directions respectively
fortunately we need more than 5 principal components to reconstruct an image
to within 10% for both cases and therefore we do not think that we cannot add
anything useful by using this method with the limited amount of data that we
have. For this reasone we need to look for another route and make the residual
stability as a function of time our next subject of interest.
4.2 Residual stability over time
Stability of the residuals over time is very important to understand properly if
we wish to successfully reduce the background. We discussed in the chapter
above how the theory from Pantin (2015) suggested that the background is sta-
ble for only 2 hours before it changes dramatically. This seems to be confirmed
by our short dt data, where the background hardly changes at all in 15 minutes.
But when comparing a frame to one taken nearly 3 months later, we see that
this has also hardly changed at all. (See Fig. 4.4) This suggests that the change
in high order aberrations is not dominated by the temperature fluctuations but
something more systematic and is a big discrepancy compared to what is sug-
gested in Pantin (2015).
It is hard to compare the data from Pantin with our own as the detector used for
Version of July 22, 2016– Created July 22, 2016 - 20:26
40 Results
this study has recently been replaced by the new Aquarius detector.(Ives et al.,
2012) Therefore we will not be able to exactly replicate the effects seen in that
paper. In spite of this we can still compare the images and find that the most
striking difference is the fact that both the tracking and de-rotator have been
turned off in our dataset.
The two observations in the Pantin data are one hour apart, which would trans-
late to a 15orotation on both the telescope and derotator. Comparing the differ-
ence in the angle of the pattern, we see that they compare to the 0 and 90 degree
exposures displayed in Fig. 3.1. If we look at these images then we see that
the high order aberrations have a stronger magnitude when chopping in the 0
degrees direction than when chopping in the 90 degrees direction. This means
that if we subtract these two frames, we will be left with mainly the patterns
on the 0 degree frame with a bit of the 90 degree patterns, as shown in Fig. 4.5
and also visible in the Pantin data on Fig. 3.6. Therefore we think that despite
the difference in detector, the datasets are comparable and the effects causing
the aberrations are the same. This suggests that the background is virtually
stable with respect to the time when not moving the telescope or detector. On
top of this it also suggests that the chopping residuals do not scale linearly as a
function of throw as was theorised in the last section.
4.3 Alternative chop methods
Now that we have learned that the chop residuals do not scale with throw but
are stable over time, we can investigate the effectiveness of our new chop reduc-
tion method. If we compare the values from the quadrant test of the classical
method and those of the inverse chop addition with polynomial subtraction
that are given in table 3.2, we will see that the quadrant test gives values that
are identical within the given error apart from the largest throw of 30” where
we have an offset of 18 times the STD. This suggests that the method does not
work well for very large chop throws, but works well for smaller ones. In Fig.
4.6 we have plotted the fully reduced difference frames for both classical chop-
ping and nodding and the inverse chop addition method. All the images have
the overall STD given as a means of comparison of the overall flatness. From
both visual inspection and this metric we can see that the methods seem to give
the same result for 5” and 10” throws but break up for 30 arcsec. Given the non-
linearity that we have seen in the background as a function of chop throw, this
might mean that the balance between the two backgrounds breaks up or that
the remaining contribution of the high order aberrations is no longer negligible.
This effect would be more apparent in this method than classical chopping due
Version of July 22, 2016– Created July 22, 2016 - 20:26
4.3 Alternative chop methods 41
to the fact that the reduced terms still have a telescope offset term. (See Eq. 3.5.)
If we take a look at the inverse chop addition without subtracting a polynomial
we will see that we get an image with just a gradient and no higher order aber-
rations. This means that the higher order structures and the gradient have a
different source, which according to Eq. 3.5 can only be the atmosphere or tele-
scope contributions. We may not precisely know which component contributes
what, but we now do have a means of splitting the higher and lower order com-
ponents. The gradient can be removed by a polynomial fit of first order and the
other structure by the inverse chop addition. Combining these methods gives
us a flat image comparable to classical chopping, as can be seen in Fig. 4.7.
Now that we have looked different methods of reducing the background in
the chop difference frames, we should also realize that using the inverse chop
method, even though we do not know which component introduces which ef-
fect, we could find out if we can vary the amount of power that is put into either
of the two contributors.
Version of July 22, 2016– Created July 22, 2016 - 20:26
42 Results
Figure 4.4: To test the stability over time we have taken the same observation three
times at different times. Two frames are 15 minutes apart and the third is almost 3
months earlier. Top Row: The choped difference frame for the three exposures. Middle
Row: We see three classically chop/nod reduced images but with three different times
in between the nods. Bottom Row: The same images as in the middle row but binned
down 5x5 times.
Version of July 22, 2016– Created July 22, 2016 - 20:26
4.3 Alternative chop methods 43
Figure 4.5: Left and Center:An illustration of the background obtained from chop dif-
ference frames at the 0 and 90 degree arm (AC and AB) and the result that is obtained
when these two are used for classical chop/nod reduction. Right: The resulting nod
difference frame is comparable to the one obtained by Pantin in Fig. 3.6.
Figure 4.6: We have tested the inverse chop addition with polynomial subtraction
method on three different throws and compared the results to classical chop/nod re-
ductions. Top row: Classically reduced chop/nod images taken in the AC direction
with a throw of 5”,10”,30” respectively. Bottom row: Inverse chop addition with poly-
nomial subtraction on the same three frames as in the top row.
Version of July 22, 2016– Created July 22, 2016 - 20:26
44 Results
Figure 4.7: Top Left:We can see a strong gradient and higher order aberrations in the
chop difference frame. (To be able to use the same colorbar, we have changed the sign
of this frame.) Top Center: We can easily remove the gradient by fitting a first order
polynomial, thus leaving the higher order structure. Top Right: We can also use the in-
verted chopping technique and add two difference frames to get rid of all the higher or-
der structure but retain the gradient. Bottom Center: Combining both methods leaves
us with a relatively flat image, comparable to that of classical chopping.
Version of July 22, 2016– Created July 22, 2016 - 20:26
4.4 Disentangling Backgrounds 45
4.4 Disentangling Backgrounds
Figure 4.8: If we have a gradient (top left) and higher order structure (top right) we
can divide the two to get the structures as seen in the lower row. Gradient/structure
(bottom left) gives an overall gradient with the structure being dampened while struc-
ture/gradient (bottom right) will give us a flat image with the structure overlayed but
with an internal gradient.
In the last section we reasoned that it should be possible to disentangle the two
background components if we could vary the amount of power in either of the
two contributors. The most straightforward way of doing this is by by observ-
ing trough different filters. In Fig. 4.9 we see a plot of the three filters that have
been observed during our technical run at 0 and 90 degrees. Namely J7.9, SIV
and the standard J8.9, which will be called the ’Sky’, ’Telescope’ and ’Middle’
filters respectively for sake of simplicity. Looking by eye at the top row it is
possible to see a difference in amplitude of the two components. the Sky filter
seems to have less ripples and a stable gradient, while the Telescope filter has
much stronger features. The bottom row seems to have less of a gradient alto-
Version of July 22, 2016– Created July 22, 2016 - 20:26
46 Results
gether, but the amount of higher order structure varies a lot. However since the
filters also have different bandwidths and throughputs while having the same
exposure time, the differences could be caused by that. Therefore we should
take a look at the divided result of the two and see what comes out of that, since
divisions are less sensitive to scalar differences.
We have done just that by taking the fractions between the Telescope, Sky and
Mid filters. However due to the fact that the Sky filter was underexposed com-
pared to the other two we could not get any useful results out of fractions con-
taining this filter. The remaining two filters were much more similar in the
amount of ADUs and therefore comparable. In Fig. 4.10 we have show the frac-
tions between the Mid and Telescope filters for both directions.
Looking at the images obtained from the 0 degrees chop direction we have a
difficult time saying anything based on the pattern as it seems to disappear on
its own in the upper right corner, which is where we would like to see the effects
of the division. Fortunately this is not the case for the images taken in the 90 de-
grees chop direction, where we can the strong features that go across the entire
image get dampened strongly in the upper right corner and being brighter in
the lower left corner. When compared to its counterpart in Fig.4.9, we have rea-
son to believe that the gradient in the Mid filter image is stronger than the gra-
dient in the Telescope filter image and therefore we see the structure/gradient
effect when dividing telescope/Mid. This Suggests that the gradient is caused
by the atmosphere and the structure by the telescope.
We could compare this data to the closed dome observations, but the chop dif-
ference frames are all perfectly flat with no residuals or gradients to be seen.
(See Fig. 4.11) We think that this is due to the fact that the dome is significantly
warmer than the atmosphere and completely out of focus and therefore smooth.
If the dome background is much brighter than the telescope background then
the latter would disappear in the shotnoise of the former, which seems to be
confirmed by the high STD. This would give us identical images when chop-
ping and thus the chop difference frames would already be flat.
Version of July 22, 2016– Created July 22, 2016 - 20:26
4.4 Disentangling Backgrounds 47
Figure 4.9: We make observations in three filters, namely the J7.9 as a filter that is
dominated by the sky background, SIV as a filter that is dominated by the telescope
background and J8.9 as something in between. Top Row: We see the the 0 degrees, AC
chop arm in the three respective filters. Bottom Row: We see the same but for the 90
degrees, AB arm.
Version of July 22, 2016– Created July 22, 2016 - 20:26
48 Results
Figure 4.10: Results of the filter test for the two chop arms. All six possible permuta-
tions for divisions have been done to attempt to see the predicted patterns, but only two
gave an usable result. Top Row: The filter test used on the AC arm with the Middle
filter and the Telescope filter. Left we see Middle/Telescope and right we see Tele-
scope/Middle. Bottom Row: The same as in the top row but for the AB arm.
Figure 4.11: The chop difference frames for all four directions of the nodding cross
taken with a closed dome. All images are identical with a STD that is about three times
higher than the open dome images. No higher order residuals can be seen on these
chop only images.
Version of July 22, 2016– Created July 22, 2016 - 20:26
4.5 The Derotator test 49
4.5 The Derotator test
Figure 4.12: The derotator test with a 30” chop throw in the AC and AB direction. Top
Row: The control image where we have chopped in the AC and AB direction for the
two images respectively. Bottom Row: The derotator test images with both images
chopping in the AC direction and with the detector rotating 90oin the second image,
simulating an AB chop from the detectors point of view. See Fig. 3.11 for a diagram of
the test.
Our data suggests that the higher order residuals come from the telescope. We
could take it a step further by trying to determine which part of the telescope is
the main contributor of these residuals. Using our data from the derotator test
can do just that and see whether the main factor is the telescope or the instru-
ment. In Fig. 4.12 we see the results of the test with a 30” throw. The pattern
in the lower right panel is clearly more similar to the upper right panel than
the left one. The higher order structure changed implying that this is caused
by the instrument while the gradient direction flipped by 90 degrees, suggest-
Version of July 22, 2016– Created July 22, 2016 - 20:26
50 Results
ing that its direction is determined purely by the telescope chopping direction.
This suggests that we might be able to remove the higher order structures by
removing the defect in the instrument that causes them.
4.6 Effects as a function of airmass
It might also be interesting to see if the gradient and higher order structures
change as a function of airmass. To test this we have plotted the four arms of
the nodding cross at the two altitudes of 45 and 30 degrees ALT in Fig. 4.13.
Here we can see that the patterns are indeed not the same as in Fig. 3.1 due to
the rotation, but we can see some similarities. If we subtract the two frames we
see that the background disappears quite well and we are left with flat images
that have a similar spread to classically chopped and nodded images. It is not
sure how this scales with airmasses with a larger difference than 22.5%
Figure 4.13: Top Row: A standard nodding cross executed at an elevation of 30o.Center
Row: A standard nodding cross executed at an elevation of 45o.Bottom Row: The
result obtained when the two frames above frames are subtracted from one another.
The flatness of the image, comparable to classical chopping and nodding, implies that
the patterns are not dependant on airmass.
Version of July 22, 2016– Created July 22, 2016 - 20:26
4.7 Testing on a source 51
4.7 Testing on a source
Figure 4.14: Top: HR 2652 as seen in the DSS survey. Bottom: The same object as seen
by VISIR with a 0 (AC) and 90 (AB) degree chopping direction.
Now that we have tried and tested several methods on an empty patch of sky,
it would also be interesting to see how these methods behave when looking at a
source. For this reason we have made an observation of the bright standard star
HR 2652 which has a relatively dim companion named TYC 8130-1560-1. This
star was observed at 63 degrees ALT with a rotation angle of nearly 270 degrees
and a chop throw of 10”. In Fig. 4.14 we can see an image from the Simbad
database and the AC and AB arm of the nodding cross, with patterns that cor-
respond to the patterns that we can see in Fig. 3.1 but with a 90 degree rotation.
Now we can test the effectiveness of the reduction method by comparing the
end result with that of classical chopping and nodding, especially by whether
or not we can see TYC 8130-1560-1.
In Fig. 4.15 we can see four sets of reduction methods applied on the two dif-
ference frames that were given in Fig. 4.14. We start with classical chopping,
Version of July 22, 2016– Created July 22, 2016 - 20:26
52 Results
Figure 4.15: A comparison of three different background reduction methods. A simple
photometric count has been made on the small companion of the central star and is
displayed below each individual frame. Left Column: Classical chopping for both AC
and AB directions that were shown in Fig. 4.14. Middle Column: The same sources
after a simple reduction of only a first order polynomial subtraction. Right Column:
The same sources reduced by use of inverse chop addition with polynomial fitting.
where we see the familiar tripple pattern with two negative stars and one posi-
tive star, and for both directions we can see a small point source near the upper
left of the central star. We have done some basic photometry on this object by
using the aper function from the Leiden ’Sterrenkunde Practicum’. This gave us
the background subtracted counts that are found in a region around this object.
These numbers and a basic visual inspection will be our baseline for evaluation
of the methods.
In the second column we can see the polynomial fit method, in which we only
fit a polynomial to the data as a means of reduction. From both the counts and
the image it is clear that this method will not work in practice, as the source is
hardly visible at all and suffers from over subtraction.
Version of July 22, 2016– Created July 22, 2016 - 20:26
4.7 Testing on a source 53
The third column is the most new promising method, namely the inverse chop
addition, where we add two opposing chop directions and then remove the
resulting gradient. As can be seen this method gives us a very comparable
result both visually and when looking at the numbers.
Version of July 22, 2016– Created July 22, 2016 - 20:26
Chapter 5
Conclusions and Discussion
We have observed an empty patch of sky for a long time and under different
conditions. The Derotator test has given us a number of insights about the na-
ture of the MIR background, its stability and cause. From this research we have
drawn 5 conclusions which are listed below.
1. We have looked at the stability of the background as a function of throw
and found that the background does not scale linearly and that the re-
duced images deteriorate in quality with throw. Therefore we suggest us-
ing the smallest possible throw for future observations. This will not pose
a problem for point sources but means lesser performance for extended
2. There is an alternative to getting a background comparable to classical
chopping and nodding. The method is provided in the form of inverse
chop subtraction combined with a polynomial fit. This is because the in-
verse chop subtraction rids us of the higher order structures and leaves a
simple gradient that we can remove with a first order polynomial fit. We
believe that due to the non-linearity of the background as a function of
chop throw, this method is not stable for large (>10”) throws. This is a
good alternative for classical chopping and nodding.
3. It is shown that the amplitude of the gradient can vary heavily for differ-
ent chop directions and over time while the higher order patterns stay the
same if all other variables (e.g. AZ, derotator angle) stay the same. This
implies that the gradient depends on something with a time-scale longer
than 15 minutes while the high order aberrations depend on the chop di-
rection and filter and have no time dependency.
Version of July 22, 2016– Created July 22, 2016 - 20:26
56 Conclusions and Discussion
4. Using observations from three different filters we have found that the
background contributions from the telescope and sky are different. We
have found that the overall gradient is due to the sky and the higher order
structures are due to the telescope. Taken together with the last conclu-
sion this implies that the higher order aberrations are introduced by the
telescope and depend on the angle of the detector with respect to the ALT
direction. Chopping in the ALT direction gives much stronger residuals
than the two perpendicular directions.
5. From the derotator test we have learned that the lower order structures are
caused by something inside of the instrument. This is either the window,
the detector or some kind of interaction between the two. Finding the
exact source of this effect could give us a method of solving the higher
order aberrations, making it a lot easier to reduce the MIR background.
However this is not doable with the data that we have and follow-up tests
are needed.
While doing this project we have come up with some new methods of subtract-
ing the background from MIR observations where nodding is not a possibility.
We believe that when having a small enough chop throw it will most certainly
be possible to properly reduce the image without nodding, however we have
only uncovered the tip of the iceberg here. Most of our tests were very superfi-
cial and based on a handful of data points taken during one hour of observing
at the VLT. To confirm these results we believe that it is crucial to make more
observations where we can focus on the conclusions made above. For these
observations we suggest the following:
1. Take a couple of nodding crosses in various filters with the proper expo-
sure time to better understand the scaling of the gradient and the higher
order structure. We suggest to use a chop throw of 10” to maximize the
amplitude of both backgrounds while also staying in the regime where the
images become truly flat after reduction. Hopefully this will shed more
light on our theory of the atmosphere introducing the gradient and the
telescope the other structures.
2. Similarly we need to test the background as a function of airmass. We
have lost many good data points due to overexposure and the remaining
dataset was far from perfect, although still indicating that the background
does not depend on altitude. We need a few extra measurements with the
same rotation and large airmass variations to confirm this.
Version of July 22, 2016– Created July 22, 2016 - 20:26
3. To get definitive proof of whether the instability after two hours described
by Pantin is truly due to the tracking and derotator, we have to do a sim-
ilar test where we make a nodding cross with the tracking and derotator
running with observations after 1, 2 and 3 hours. We can then compare the
background data after three hours with the 90oarm of the nodding cross.
4. Once we have a better understanding of the above points, it is important
to try inverse chop addition with polynomial fit methods out on point
sources and extended sources. Comparing this to classical chopping should
give a definitive answer to the question of whether they work or not.
5. Because we did not get open dome data for the GTC we have not worked
with the data. Once the CANARICAM is operational again it would be
interesting to finish the observations and redo the analysis that has been
done in this thesis. If this proves to be impossible, a different telescope
should be considered.
Version of July 22, 2016– Created July 22, 2016 - 20:26
58 Conclusions and Discussion
After spending a year on this research project I can honestly say that I have
learned a lot about preparing an observing plan and working with MIR data.
I very much enjoyed the balance between theory and instrumentation and the
chance to work with my own data obtained from the VLT. Because of this I
would like to express my deep gratitude to my supervisor, Proffessor Bernhard
Brandl, for his guidance, help with ESO and constructive criticism. I would
also like to thank Jeff Meisner, Roy van Boekel, Rudolf le Poole and Eric Pantin
for their advice and suggestions concerning the project, as well as Konrad Tris-
tram, Daniel Asmus and Julien Girard from ESO and Antonio Cabrera Lavers
from the ICA for helping me obtain my data at the VLT and GTC. My grateful
thanks are also extended to Steve Hammond, Dominique Petit dit de la Roche
for extensively proofreading my work to remove all the typos.
Finally I would like to thank my parents and Dominique for their continuing
support and encouragement throughout my study.
Version of July 22, 2016– Created July 22, 2016 - 20:26
B. Brandl. VLT Technical Time Proposal, A Study to develop Chop-
ping/Nodding Strategies for E-ELT/METIS. 2015.
B. R. Brandl, R. Lenzen, E. Pantin, A. Glasse, J. Blommaert, L. Venema, F. Mol-
ster, R. Siebenmorgen, H. Boehnhardt, E. van Dishoeck, P. van der Werf,
T. Henning, W. Brandner, P.-O. Lagage, T. J. T. Moore, M. Baes, C. Waelkens,
C. Wright, H. U. K¨
aufl, S. Kendrew, R. Stuik, and L. Jolissaint. METIS: the
Mid-infrared E-ELT Imager and Spectrograph. In Ground-based and Airborne
Instrumentation for Astronomy II, volume 7014, page 70141N, July 2008. doi:
M. Cohen, R. G. Walker, B. Carter, P. Hammersley, M. Kidger, and K. Noguchi.
Spectral Irradiance Calibration in the Infrared. X. A Self-Consistent Radio-
metric All-Sky Network of Absolutely Calibrated Stellar Spectra. 117:1864–
1889, April 1999. doi: 10.1086/300813.
T. Geballe and R. Mason. Introduction to Ground-based Mid-IR Observ-
mid-ir-observing, 2006. [Online; accessed 2016-06-15].
D. Ives, G. Finger, G. Jakob, S. Eschbaumer, L. Mehrgan, M. Meyer, and
J. Steigmeier. AQUARIUS, the next generation mid-IR detector for ground-
based astronomy. In High Energy, Optical, and Infrared Detectors for Astronomy
V, volume 8453, page 845312, July 2012. doi: 10.1117/12.925844.
S. Kendrew, L. Jolissaint, B. Brandl, R. Lenzen, E. Pantin, A. Glasse, J. Blom-
maert, L. Venema, R. Siebenmorgen, and F. Molster. Mid-infrared astron-
omy with the E-ELT: performance of METIS. In Ground-based and Airborne
Instrumentation for Astronomy III, volume 7735, page 77355F, July 2010. doi:
Version of July 22, 2016– Created July 22, 2016 - 20:26
J. H. Lacy, M. J. Richter, T. K. Greathouse, D. T. Jaffe, and Q. Zhu. TEXES: A
Sensitive High-Resolution Grating Spectrograph for the Mid-Infrared. 114:
153–168, February 2002. doi: 10.1086/338730.
S. D. Lord, D. J. Hollenbach, S. W. J. Colgan, M. R. Haas, R. H. Rubin, E. F.
Erickson, P. Carral, and P. Maloney. A Far-Infrared Spectral Line Survey of 23
Disk Galaxies. In American Astronomical Society Meeting Abstracts, volume 24
of Bulletin of the American Astronomical Society, page 1182, December 1992.
Jeff Meisner. personal communication, May 2016.
J. B. Oke. Absolute Spectral Energy Distributions for White Dwarfs. 27:21,
February 1974. doi: 10.1086/190287.
Richter M Packham C Chun M Otarola, A. Atmospheric transmission and ther-
mal background emission in the mid-infrared at Mauna Kea. Adapting to the
Atmosphere Conference 2014, 2014.
Sander Paalvast, Robert Huisman, Bernhard Brandl, Huub Jassen, Bayu
Jayawardhana, Frank Molster, Maurice Teuwen, and Lars Venema. Devel-
opment and characterization of a 2d precision cryogenic chopper for metis,
2014. URL
E. Pantin. Study of chopping only performances. 2015.
K Pearson. Liii. on lines and planes of closest fit to systems of points in
space. Philosophical Magazine Series 6, 2(11):559–572, 1901. doi: 10.1080/
14786440109462720. URL
G. S. Rothman, T. E. Gill, and C. Chang. Saharan Air Layer Interaction with
Hurricane Claudette (2003). AGU Fall Meeting Abstracts, December 2004.
G Rupprecht. Very Large Telescope, Requirements for Sceintific Instruments on
the VLT Unit Telescopes. ESO Internal, March 2005.
K. Tristram. Very Large Telescope: VISIR User Manual v98.1. 2016.
K Volk. Chopping and Nodding for Mid-Infrared Astronomy. 2007.
Version of July 22, 2016– Created July 22, 2016 - 20:26
... Recently, using test data on VLT/VISIR, we showed that a similar technique can achieve essentially the same signal/noise ratio as traditional chopping/nodding on the VLT. 7,16 Due to limitations in the VLT M2 mirror control, we could not test a three-point chopping method as Landau's, 15 but we could alternate on ≈ minute timescales between standard observations chopped North and observations chopped South. Combined, this "Inverse Chop Addition" resulted in a background subtraction that is on par with the standard chop/nod method, at least for short observations. ...
Full-text available
Ground-based thermal-infrared observations have a unique scientific potential, but are also extremely challenging due to the need to accurately subtract the high thermal background. Since the established techniques of chopping and nodding need to be modified for observations with the future mid-infrared ELT imager and spectrograph (METIS), we investigate the sources of thermal background subtraction residuals. Our aim is to either remove or at least minimise the need for nodding in order to increase the observing efficiency for METIS. To this end we need to improve our knowledge about the origin of chop residuals and devise observing methods to remove them most efficiently, i.e. with the slowest possible nodding frequency. Thanks to dedicated observations with VLT/VISIR and GranTeCan/CanariCam, we have successfully traced the origin of three kinds of chopping residuals to (1) the entrance window, (2) the spiders and (3) other warm emitters in the pupil, in particular the VLT M3 mirror cell in its parking position. We conclude that, in order to keep chopping residuals stable over a long time (and therefore allow for slower nodding cycles), the pupil illumination needs to be kept constant, i.e. (imaging) observations should be performed in pupil-stabilised, rather than field-stabilised mode, with image de-rotation in the post-processing pipeline. This is now foreseen as the default observing concept for all METIS imaging modes.
Full-text available
Due to the large size and mass of the secondary mirror on next generation extremely large telescopes it will not be possible to provide classical chopping and nodding as is used during mid-IR observations today. As a solution to this we propose an alternative approach to thermal background reduction called `inverse chop addition`. Here we use the symmetries of the thermal background to replace nodding, which allows us to get nearly identical background reductions while only using a special chopping pattern. The performance of this method was tested during technical time observations on VLT/VISIR. With this method, a higher observational efficiency can be obtained than with `classical chopping and nodding`, while achieving equally good reduction results. These results suggest that `inverse chop addition` could be a good alternative for classical chopping and nodding on both current and next generation ground-based facilities.
Full-text available
We present results of a preliminary study intended to quantitatively estimate the atmospheric transmission and thermal background emission in the mid-infrared (MIR), 7 μm - 26 μm, at the 13N TMT site in Mauna Kea. This is in the interest of supporting the planning of MIR instrumentation for the posible second-generation of astronomical instruments for the Thirty Meter Telescope (TMT) project. Mauna Kea, located at high altitude (4,050 m above sea level), enjoys natural conditions that make it an outstanding location for astronomical observations in the mid-infrared. The goal of this work is to produce a dataset and model that shows the atmospheric transmission and thermal emission for two cases of precipitable water vapor (PWV), a low value of 0.3 mm, and at 1.5 mm which represent near median conditions at the site. Besides, and driven by the interest of the MIR community to exploit the daily twilight times, we look at the specific atmospheric conditions around twilight as a function of season. The best conditions are found for cold and dry winter days, and in particular the morning twilight offers the best conditions. The analysis of PWV data, shows the median value for the site (all year conditions between 6:00 PM and 7:30AM) is 1.8 mm and that periods of water vapor lower than 1.0 mm are common, these supports the opportunity and discovery potential of the TMT project in the mid-infrared bands.
Full-text available
ESO has recently funded the development of the AQUARIUS detector at Raytheon Vision Systems, a new mega-pixel Si:As Impurity Band Conduction array for use in ground based astronomical applications at wavelengths between 3 - 28 μm. The array has been designed to have low noise, low dark current, switchable gain and be read out at very high frame rates. It has 64 individual outputs capable of pixel read rates of 3MHz, implying continuous data-rates in excess of 300 Mbytes/second. It is scheduled for deployment into the VISIR instrument at the VLT in 2012, for next generation VLTI instruments and base-lined for METIS, the mid-IR candidate instrument for the E-ELT. A new mid-IR test facility has been developed for AQUARIUS detector development which includes a low thermal background cryostat, high speed cryogenic pre-amplification and high speed data acquisition and detector operation at 5K. We report on all the major performance aspects of this new detector including conversion gain, read noise, dark generation rate, linearity, well capacity, pixel operability, low frequency noise, persistence and electrical cross-talk. We describe the many possible readout modes of this detector and their application. We also report on external issues with the operation of these detectors at such low temperatures. Finally we report on the electronic developments required to operate such a detector at the required high data rates and in a typical mid-IR instrument.
Full-text available
We present results of performance modelling for METIS, the Mid-infrared European Extremely Large Telescope (E-ELT) Imager and Spectrograph. Designed by a consortium of NOVA (Netherlands), UK Astronomy Technology Centre (UK), MPIA Heidelberg (Germany), CEA Saclay (France) and KU Leuven (Belgium), METIS will cover the atmospheric windows in L, M and N-band and will offer imaging, medium-resolution slit spectroscopy (R~1000-3000) and high-resolution integral field spectroscopy (R~100,000). Our model uses a detailed set of input parameters for site characteristics and atmospheric profiles, optical design, thermal background and the most up-to-date IR detector specifications. We show that METIS will bring an orders-of-magnitude level improvement in sensitivity and resolution over current ground-based IR facilities, bringing mid-IR sensitivities to the micro-Jansky regime. As the only proposed E-ELT instrument to cover this entire spectral region, and the only mid-IR high-resolution integral field unit planned on the ground or in space, METIS will open up a huge discovery space in IR astronomy in the next decade. Comment: 13 pages, submitted to SPIE Proceedings vol. 7735, Ground-based and Airborne Instrumentation for Astronomy III (2010). Simulation code available at
Full-text available
We discuss the design and performance of TEXES, the Texas Echelon Cross Echelle Spectrograph. TEXES is a mid-infrared (5-25 um) spectrograph with several operating modes: high-resolution, cross-dispersed with a resolving power of R=100,000, 0.5% spectral coverage, and a ~1.5 by 8" slit; medium-resolution long-slit with R~15,000, 0.5% coverage, and a ~1.5 by 45" slit; low-resolution long-slit with (delta lambda)~0.004 um, 0.25 um coverage, and a ~1.5 by 45" slit; and source acquisition imaging with 0.33" pixels and a 25 by 25" field of view on a 3m telescope. TEXES has been used at the McDonald Observatory 2.7m and the IRTF 3m telescopes, and has proven to be both sensitive and versatile. Comment: 33 pages, 11 figures. Accepted to PASP. For slightly better, color versions of Figs 1-3, see
Full-text available
METIS, the Mid-infrared ELT Imager and Spectrograph (formerly called MIDIR), is a proposed instrument for the European Extremely Large Telescope (E-ELT), currently undergoing a phase-A study. The study is carried out within the framework of the ESO-sponsored E-ELT instrumentation studies. METIS will be designed to cover the E-ELT science needs at wavelengths longward of 3um, where the thermal background requires different operating schemes. In this paper we discuss the main science drivers from which the instrument baseline has been derived. Specific emphasis has been given to observations that require very high spatial and spectral resolution, which can only be achieved with a ground-based ELT. We also discuss the challenging aspects of background suppression techniques, adaptive optics in the mid-IR, and telescope site considerations. The METIS instrument baseline includes imaging and spectroscopy at the atmospheric L, M, and N bands with a possible extension to Q band imaging. Both coronagraphy and polarimetry are also being considered. However, we note that the concept is still not yet fully consolidated. The METIS studies are being performed by an international consortium with institutes from the Netherlands, Germany, France, United Kingdom, and Belgium. Comment: 15 pages, to be published in Proc SPIE 7014: Ground-based & Airborne Instrumentation for Astronomy II
Conference Paper
The Mid-infrared E-ELT Imager and Spectrograph, or METIS, is foreseen as an early instrument for the European Extremely Large Telescope (E-ELT). A key part of METIS is the Cold Chopper (MCC) which switches the optical beam between the target and a nearby reference sky during observation for characterization of the fluctuating IR background signal in post-processing. This paper discusses the development and characterization of the realized MCC demonstrator. The chopper mirror (Ø64mm) should tip/tilt in 2D with a combined angle of up to 13.6mrad with 1.7μrad stability and repeatability within 5ms (95% duty cycle at 5Hz) at 80K. As these requirements cannot be met in the presence of friction or backlash, the mirror is guided by a monolithically integrated flexure mechanism. The angular position is actuated by three linear actuators and measured by three linear position sensors, resulting in a fast tip, tilt, and focus mirror. Using the third actuator introduces symmetry, and thus homogeneity in forces and heat flux. In an earlier paper, Ref. [1], the design of the chopper and the breadboard level testing of the key components were discussed. Since then, the chopper design has been revised to implement the lessons learned from the breadboard test and a demonstrator has been realized. This demonstrator has undergone an elaborate test program for characterization and performance validation in a cryogenic environment, as discussed in this paper.
We start from our six absolutely calibrated continuous stellar spectra from 1.2 to 35 μm for K0, K1.5, K3, K5, and M0 giants. These were constructed as far as possible from actual observed spectral fragments taken from the ground, the Kuiper Airborne Observatory, and the IRAS Low Resolution Spectrometer, and all have a common calibration pedigree. From these we spawn 422 calibrated "spectral templates" for stars with spectral types in the ranges G9.5–K3.5 III and K4.5–M0.5 III. We normalize each template by photometry for the individual stars using published and/or newly secured near- and mid-infrared photometry obtained through fully characterized, absolutely calibrated, combinations of filter passband, detector radiance response, and mean terrestrial atmospheric transmission. These templates continue our ongoing effort to provide an all-sky network of absolutely calibrated, spectrally continuous, stellar standards for general infrared usage, all with a common, traceable calibration heritage. The wavelength coverage is ideal for calibration of many existing and proposed ground-based, airborne, and satellite sensors, particularly low- to moderate-resolution spectrometers. We analyze the statistics of probable uncertainties, in the normalization of these templates to actual photometry, that quantify the confidence with which we can assert that these templates truly represent the individual stars. Each calibrated template provides an angular diameter for that star. These radiometric angular diameters compare very favorably with those directly observed across the range from 1.6 to 21 mas.
VLT Technical Time Proposal, A Study to develop Chopping/Nodding Strategies for E-ELT/METIS
  • B Brandl
B. Brandl. VLT Technical Time Proposal, A Study to develop Chopping/Nodding Strategies for E-ELT/METIS. 2015.