P -- photon statistics

This is used to choose between the different ways in which the errors are calculated. There are three possible choices selected by the integers 1 to 4 which have the following meanings :

1. Errors from photon statistics.
2. Errors from variations in the sky aperture.
3. Errors from data variance.
4. Gaussian errors from variations in the sky aperture.

The first works out the errors from photon statistics in the sky and signal apertures. This requires you to know and set-up the parameters PADU and BIASLE which convert the data values to numbers of photons. The message `Errors from photon statistics' will signify that this has been chosen.

The second method of calculating the errors is from the measured variance in the sky aperture. This method assumes that the measured variance is due to photon statistics and scales the measurement in the object aperture accordingly. This method still requires the parameter PADU to be known, but does not need the BIASLE parameter to be known. The message `Errors from sky variance' will signify that this has been chosen. If neither the PADU or BIASLE parameters are known, then it is best to use this method to indicate the reliability of the measurements, but not to take the quoted error values as absolute since this method will be wrong by a factor $\sqrt{PADU}$, where $PADU$ is the unknown conversion factor.

The third method of calculating the errors is from the data errors that are stored with the image (one per pixel). This method of calculating the errors also requires the parameter PADU to be known. The message `Errors from data variance' will signify that this has been chosen. A variance component may not always be present in data file along with the data array (indeed this can only be true if you are storing your images in NDFs, see here), and if this is the case then PHOTOM will issue the warning `Data does not have a variance component' if this method is selected.

The fourth method of calculating the errors is like the second and uses the measured variance in the sky aperture. This method assumes that the measured variance is due to some gaussian source and doesn't require any knowledge of the PADU and BIASLE values (which are unknown when dealing with data that has been combined using a mean, say from a CCD Mosiac dithered on the sky), but clearly this can only measure an upper limit as the actual noise in the object will be (fractionally) less than in the sky. The best way to avoid such uncertainty is by propagating data variances through all the stages that produced the combined data and using method three.

A full discussion of the calculation of the errors, assuming photon statistics, is given elsewhere.