It was established during the operational lifetime of ISOCAM that observations of the same source have a very good repeatability. Indeed, repeated photometric observations on HIC89474 throughout the whole ISO lifetime give a standard deviation of around 2% for both the LW and SW detectors.
Observations of a larger set of standard stars which were used to determine the absolute flux calibration give a RMS of around 5% for both detectors. It was also found that the LW detector is very linear over a very large range of flux levels from a few mJy up to the saturation level.
The important finding here is that the signal needs to be stabilized to improve on the photometric accuracy. Current routines in CIA are capable of correcting the observed number of digital counts to within 1-2 % of the expected number of counts had the detector responded instantaneously to the incident light. The transient algorithm has now been introduced in OLP.
The CVF Spectral Response Function was established based on several observations of standard stars. The analysis shows that these are highly reproducible (per observation) and accurate, i.e. the known fluxes of the standard stars are correctly estimated from the observations. The RMS of repeated measurements is 3% for the SW part and 4% for the 2 LW CVF segments. A systematic comparison of fluxes derived by OLP for the calibration standard stars reveals that OLP fluxes are systematically lower than the expected star model fluxes. This is puzzling as those very model stars were used to establish the calibration of the CVFs. A very likely explanation is that the calibration curves were established by aperture photometry whereas OLP point source detection is based upon PSF fitting algorithms. It would appear then that aperture photometry should yield more accurate results than currently found by OLP.
One other aspect which one has to handle with great care is to exclude the ghost image. Including it when applying aperture photometry may lead to errors of up to 25%.
It was also mentioned in section 2.1 that the FlatField is very accurate in the central part of the array but can be off by a factor of 2 or more at the edges. This effect cannot, for the time being, be corrected for staring observations. However in the case of rasters with sufficient background and redundancy one can build a FF from the very observations being analyzed.
The pin-cushion effect was mentioned in Section 2.2. There are tools in CIA to correct for this effect. Note that in the case of rasters with redundancy this correction has two beneficial results: