Improvements to the image processing of Hubble Space Telescope NICMOS observations with multiple readouts

We report on improvements made to the standard NICMOS processing pipeline. The calculation of the uncertainties on the signal accumulation rate has been modified to include the statistical correlations between the consecutive readouts. This leads to a similar to 30% correction in the photometric weight of individual pixels containing faint objects. In order to correct a problem with the existing cosmic-ray rejection algorithm, we have developed and implemented a joint fit procedure, where the accumulating signal is fit as linear functions of time with the same rate both before and after the cosmic-ray (CR) impact. This procedure leads to significantly smaller uncertainty in the count rate for pixels affected by CRs. We also show that regions neighboring CRs found by the standard NICMOS pipeline are systematically brighter. This interpixel correlation substantially increases the footprint of CR impacts and is treated for the first time by our new pipeline. The new processing is most relevant for photometry from deep observations of faint targets, for which accurate, optimal, and unbiased uncertainty estimates are important. We present examples of these improvements for deep NIC2 images of a high-redshift supernova from the Supernova Cosmology Project. The net improvement is a factor of 2 reduction in the number of 3 sigma photometric outliers.