Transients are the non-linear departures from the instantaneous
response of the detector when the light source is either placed on or
moved away from the detector. Results so far suggest that the shapes of the
transient and the time-scales on which transient effects persist in
the data depend on the input source flux, the prior history of the
detector, the amount of background light and, additionally, show
pixel-to-pixel variations.
Transients affect: (1) the flux calibration, and (2) transients
introduce artifacts (such as ghosts). The CIA procedure
stabilize offers 6 options for removing the effects of
transients, ranging from simply masking the unstabilized data to
fitting complex, non-linear models of transients to the data. Each
method has limitations and you are strongly encouraged to read the
list of recommended readings to learn more before selecting
one. However, you should note that:
No ISOCAM Transient removal method corrects the so-called ``hook'' effect
(see recommended reading).
Transient removal methods may over-correct the data.
So, which method should you use? There is no one method which works for
all data. We can, however, offer the following general guide-lines.
Always make a copy of the data before applying transient correction.
Recognize the limitations of each model. For example, the Fouk-Schubert
treatment (method='FS') has been shown to work well for sources with diffuse,
low-contrast, extended emission but fails for point-sources.
Similarly, the 'vision' option which uses redundancy (eg in a rastered
observation with large overlapping regions) provides unsatisfactory results
for observation without redundancies (eg staring mode observations).
Decide if you need to apply any complex fitting methods. If the
on-source integration time (not the total time, but the duration for
which the source remained on a particular region of the detector) is
~150 seconds or longer, the signal has likely to have stabilized (see
FLUX CALIBRATION) for "intermediate" to "high" flux sources and
relatively "high/medium" background. For more details see the
'recommended readings section'. For on-source integration times
between 100-200 seconds the source is likely to be between 80-100
percent stabilized. You should weigh the effect of this uncertainty on
your science goals vs both the additional time-commitment required and
of any uncertainties that the models themselves may introduce.
Compare several transient methods with particular attention to how each
changes the data cube.
Be wary of any transient models which change the data cube by more
than ~66 percent from the original levels. The initial few readouts
(on-source integration times less than ~30 seconds may indeed require
drastic corrections to the observed signal. In such cases a higher
correction than 66% can occur. However, when the signal
appears to be stabilized large corrections usually mean
over-corrections.
