The RADHIT_SAT module (“Radiation Hit detection prior to Saturation Correction”) detects and flags cosmic ray events in FITS data cubes and updates the status bits in the dmask.fits file. . The module looks for events in time series data (ramps) and works on a pixel by pixel basis; no spatial information is used. Cosmic ray hits are assumed to cause discontinuous signal jumps (either positive or negative) in the ramps. The samples with detected radhits are tracked by setting bit # 9 in the output updated dmask. RADHIT_SAT first finds the sample number in the ramp with the highest likelihood of having a radhit. This likelihood is computed from Bayesian statistics. In general, Bayesian statistics combine probabilities of events (such as the actual occurrence of a radhit), with other independent probabilities (such as how often false positives are detected by a measuring process). In determining whether a particular sample has suffered from a cosmic ray event, RADHIT_SAT compares the height of the jump (divided by the sample uncertainty) with an input radhit detection threshold. The signal-to-noise threshold is one of the controlling parameters of RADHIT_SAT, and is referred to as “NominalRHMag”; its default value is 80. The sample uncertainty is the quadrature combination of sample read noise and photon noise in electrons. The sample read noise is another controlling parameter of RADHIT_SAT, and is referred to as “Readnoise Radhit.” From experience, its default value has been set to 26 electrons for SL and SH, and 46 electrons for LL and LH. If the Bayesian probability that the jump is a cosmic ray event exceeds a threshold probability (“RHPriorProb,” another controlling parameter with default value 1%), the sample is declared a radhit, and its corresponding dmask sample is flagged as indicated above. Whenever a radhit is found, the remainder of the ramp is subdivided into segments before and after the radhit. The algorithm is run again on these segments to search for additional radhits. The process is repeated until either no radhits are found or there are no segments long enough (>3 samples) to run the algorithm.
In this invocation of RADHIT, strong radhits are detected and flagged to improve subsequent ramp fitting and droop correction (see SATCOR and DROOPOP in Sections 5.1.5 and 5.1.6 below). A second pass of RADHIT after data linearization is used to pick up the more subtle radhits prior to final slope computation; see Section 5.1.12.