 |
EUROPEAN SOUTHERN OBSERVATORY
La Silla - Science Operation Department
|
|
Flatfields for the mid infrared: Experiments with TIMMI2
|
| Version: 1.0 2002-08-20 Oliver Schütz |
1. Why are conventional flatfields not possible in
the mid-IR ?
Different from the optical wavelength range and near-IR (1-5 µm),
no flatfields are made for observations at thermal wavelegths (10-20
µm). Techniques used to obtain flatfields
at shorter wavelengths will fail in the mid-IR . A mid-IR "dome
flat" would saturate the detector because the screen is at room
temperature. Simple sky frames in the mid-IR are dominated from fast
fluctuations in the sky background and flatfields obtained from these
will not be constant in time neither in spatial dimensions. Additional
to this, data will not be comparable if taken with different electronical
read-out modes. Altogether, any possible TIMMI2 flatfield must comply
with the following characteristics:
-
be spatially close to the science object
-
be very close in time to the science exposure
-
taken with the same detector read-out mode
-
of course, taken with the same filter settings
Therefore for almost each science observation an
individual flatfield would be required. In first tests however,
even within the shortest possible time between science object and
calibration data, no good flats could be obtained.
2. The current observing situation with TIMMI2
All data at thermal wavelengths are taken with chopping and nodding
corrections to eliminate the thermal background from the sky and
telescope (see the
TIMMI2
web page for more detailed information on this technique). As a
result of this procedure, images reduced by the TIMMI2 pipeline or an
equivalent observer's routine are already relatively flat. However,
certain inhomogenities persist. A point source often appears fainter in
the lower half of the detector and the variations in the count rate over
the detector array may amount up to 20%. These patterns show to some
level a constant trend but are also variable in time, and therefore
cannot be explained only by the individual pixels' response of the
detector. Sky background variations and atmospherical influence have a
major impact on these features. This is the reason
why the question of flatfielding mid-IR data up to now only has been
solved for space based observations (Starck J.L. et al. 1999,
Astron. Astrophys. Suppl. Ser. 134, 135-148).
3. Basic ideas for creating a detector gain map
For optical detectors, variations along the detector array are mainly
caused by dust on the detector or a remaining misalignment within the
light beam, due to which the beam may hit the detector not exactly
perpendicular. In the infrared, the different response of the detector
array results from wide-range variations in the pixel sensitivity.
Different from optical detectors each pixel is individually read out and
electronics are more difficil than in the optical. To
determine the sensitivity of each single pixel, its response is measured
in a sequence of images with different levels of illumination.
For mid-IR observations this varying illumination can be achieved by the
following ways:
-
Observe the sky at different airmass. The background level increases with
the distance from zenit when the telescope moves to thicker atmospherical
layers.
-
Observation of varying, thin clouds. Their water content makes them "hot",
i.e. bright at thermal wavelengths. Natural variations in the clouds cause
different brightnesses. Thick clouds however will saturate the detector
immediately.
-
Create a flatfield from temporal background variations within the science
exposure. This sounds difficil, but turned out to be the most promising
approach.
In the brightness sequence, the median count rate of each single exposure
is compared to the registered count rate of the individual pixel. A linear
fit to all pairs of the values (median of an image, count for a fixed
pixel) creates a response function for the single pixel. Ideally the
slope of this function should be close to 1. Pixel with a lower value react
less on exposure to light than those with a higher value. Bad pixel are
close to 0, hot pixel are much brighter than 1. Applying this to all pixel
creates a gain map for the whole detector array. With this, variations
in the sensitivity of the detector array can be identified. Dividing a
raw image by this gain map results in the corrected image, principally.
This works well for the near-IR, but for the mid-IR
more sophisticated methods are necessary. The
problem is that for TIMMI these gain maps appear to be highly variable with
time.
4. Applications to test data
4.1. Observations at different airmasses
Measuring the sky or a target at two or more airmasses leads to a sequence
of exposures with different brightness due to the increasing sky contribution
with distance from zenit. A flatfield can be created as shown above. Its
quality however will strongly depend on temporal and spatial atmospherical
fluctuations during the observations. In our tests
this flatfield could not successfully be applied to a science exposure
made within these observations.
4.2. Clouds
Thin, varying clouds result in a perfect illumination sequence. However,
another detector read-out mode has to be used during these observations.
A
flat obtained from these data is not compatible to the detector mode for
science observations.
4.3. Create the flatfield from the science exposure
Due to background fluctuations during a science exposure, also from these
data an illumination sequence can be achieved. Long-time
exposures or those with larger background fluctuation have better statistics
in the fit of the pixel response. Only these exposures will result in a
reliable flatfield.
The procedure is complex due to the spatial offsets for chopping and
nodding. In the following we describe an example for a stellar point source
observed with four nodding positions, i.e. two full nod cycles. For each
nod position, the first chopping position is called the ON position
(sometimes also called position 0). With OFF (or position 1) we
specify the second chop. The number of videoframes per nod is 1 in this
case. Each chop and nod position results in an individual flatfield, in
our example these are eight totally different flats. We found no correlation
of flats from corresponding positions, modifications are statistical and
time-dependant.
Dividing the frame for each chop and nod position by the corresponding
flatfield appears to be the nearest approach, but will result in corrupted
data. The information of the star at its position on the detector is still
included in these flatfields, even though not directly visible. When dividing
the frame of an individual chop and nod position by the corresponding flat,
the resulting image is destroyed at the position where it actually should
be corrected, i.e. the stellar surrounding. The following procedure avoids
this and is the best approximation up to now for a TIMMI2 flatfield:
-
Take the arithmetic middle of the two flats from the same nod position
in order to eliminate the star's position in the flatfields. Flats from
different nod positions can be highly variable and may not be mixed:
gm_1 = (GAIN_NOD1_OFF+GAIN_NOD1_ON)
/ 2
gm_2 = (GAIN_NOD2_OFF+GAIN_NOD2_ON)
/ 2
gm_3 = (GAIN_NOD3_OFF+GAIN_NOD3_ON)
/ 2
gm_4 = (GAIN_NOD4_OFF+GAIN_NOD4_ON)
/ 2
-
The following is the standard way for reducing one nod cycle (this is also
done by the pipeline). Note that the index number of nod positions does
not correspond to AB AB AB AB.., but is instead AB BA AB BA..
image_original_12 =
(IMA_NOD1_ON-IMA_NOD1_OFF) - (IMA_NOD2_ON-IMA_NOD2_OFF)
image_original_34 =
(IMA_NOD3_ON-IMA_NOD3_OFF) - (IMA_NOD4_ON-IMA_NOD4_OFF)
image_original_final = image_original_12
+ image_original_34
-
To obtain the best approach of a flat-corrected TIMMI2 image, now perform
the following calculations:
image_flatfielded_12 =
(IMA_NOD1_ON-IMA_NOD1_OFF) / gm1 - (IMA_NOD2_ON-IMA_NOD2_OFF)
/ gm2
image_flatfielded_34 =
(IMA_NOD3_ON-IMA_NOD3_OFF) / gm3 - (IMA_NOD4_ON-IMA_NOD4_OFF)
/ gm4
image_flatfielded_final = image_flatfielded_12
+ image_flatfielded_34
This approach can produce a kind of flatfield correction for data taken
with high atmospherical variations or fair weather conditions. The calculations
can be easily included in a personal pipeline routine. On
the other hand this method completely fails to flatfield good data.
The worse statistic when calculating the pixel response is the reason for
this.
Here are some illustrations of applying
the technique from Chapter 4.3 to science data.
5. Conclusions
For all ground-based mid-IR instruments no possibility of a reliable flatfield
correction exists up to now. We have shown a method how TIMMI2 data in
prinicple could be flatfielded. The resulting final image is not always
of better quality, depending on meteorological conditions. No recommendation
to observers can be given at the moment to apply the techniques presented
here.