Removing Cosmic-Ray Hits from Multi-Orbit HST Wide Field Camera Images

Abstract

We present an optimized algorithm that removes cosmic rays ("CRs") from multi-orbit HST Wide Field/Planetary Camera ("WF/PC") images. It computes the image noise in every iteration from the WF/PC CCD equation. This includes all known sources of random and systematic calibration errors. We test this algorithm on WF/PC stacks of 2-12 orbits as a function of the number of available orbits and the formal Poissonian sigma-clipping level. We find that the algorithm needs >= 4 WF/PC exposures to locate the minimal sky-signal (which is noticably affected by CRs), with an optimal clipping level at 2-2.5 x sigma_Poisson. We analyze the CR flux detected on multi-orbit "CR stacks", which are constructed by subtracting the best CR filter images from the unfiletered 8-12 orbit average. We use an automated object finder to determine the surface density of CRs as a function of the apparent magnitude (or ADU flux) they would have generated in the images had they not been removed. The power law slope of the CR "counts" (gamma ~= 0.6 for N(m) approaching m^gamma) is steeper than that of the faint galaxy counts down to V ~= 28 mag. The CR counts show a drop off between 28 <~ 30 mag (the latter is our formal 2-sigma point source sensitivity without spherical aberration). This prevents the CR sky integral from diverging, and is likely due to a real cut-off in the CR energy distribution below ~11 ADU per orbit. The integral CR surface denisty is <~ 10^8 deg^-2 and their sky-signal is V~= 25.5-27.0 mag arcsec^-2 or 3-13% of our NEP sky background (V=23.3 mag arcsec^-2), and well above the EBL integral of the deepest galaxy counts (B_J ~= 28.0 mag arcsec^-2). We conclude that faint CRs will always contribute to the sky-signal in the deepest WF/PC images. Since WFPC2 has _2.7 x lower read-noise and a thicker CCDs, this will result in more CR detections than in WF/PC, potentially affecting ~10-20% of the pixels in multi-orbit WFPC2 data cubes. (SECTION: Instrumentation and Data Analysis)