October 2020 STAN
CalCOS Errors Updated
On 2020 October 15, version 3.3.10 of the COS calibration software CalCOS was released. This release includes an improved method for calculating flux errors. The CalCOS ERROR arrays for the _x1d, _x1dsumN, and _x1dsum files had been calculated using the Gehrels (1986) approximation for small-N Poisson distributions. However, analyses showed that CalCOS was not applying this approximation correctly, which could result in larger than expected ERROR arrays when count values were small (less than ~ 20). This larger result in the ERROR array was mainly due to the error propagation, which included an extra "+1" term from the Gehrels approximation for every exposure that was combined in the _x1dsumN and _x1dsum files. Additionally, only one error column was included, and symmetric errors were to be assumed. This is misleading because Poisson distributions in the low-counts regime are strongly asymmetric.
The new release of CalCOS provides similar flux errors to the Gehrels analytic upper/lower confidence limit approximations. Rather than directly propagating the errors, the new release of CalCOS calculates a final error from three new columns that provide information about sources of error for the individual exposures: flat-field, gross counts, and background counts (VARIANCE_FLAT, VARIANCE_COUNTS, and VARIANCE_BKG, respectively). Also in this release, the errors are no longer assumed to be symmetric. For consistency with previous CalCOS reductions, the ERROR column still contains the upper confidence limit, and a new column ERROR_LOWER is added for the lower confidence limit. The new flux error estimates should be a significant improvement for data with moderately low signal-to-noise (S/N ~ 10–30), but users with extremely low S/N data (< 5) may need to use custom error calculations. The errors approach the symmetric Gaussian root-N limit at high counts, so users with higher S/N data (> 50) will see little difference with the improved errors.
The left column of the figure shows sample COS FUV _x1d spectra for low (top) and high (bottom) S/N cases in regions dominated by continuum emission. The red lines illustrate low-order polynomial fits to the data. The dashed blue lines show the upper and lower flux errors from the previous version of CalCOS, and the solid green lines illustrate the newly updated flux error estimates from version 3.3.10. For comparison, the large orange squares show the expected upper/lower 1-sigma confidence limits based on the detected counts in 1 Å bins and the “exact” limits tabulated by Gehrels (1986). Note the significantly improved agreement for low S/N data, especially for the lower confidence limits. The right column of the figure makes the same demonstration for high and low S/N _x1dsum files. Note in particular the reduced magnitudes of the upper/lower confidence limits in the low S/N case and their agreement with the 1-sigma tabulated errors.
This release also includes a minor change to the _x1dsumN files, where the GCOUNTS column did not properly take into account the DQ_WGT value. The new version of CalCOS is available through the astroconda distribution. All COS data were reprocessed with CalCOS 3.3.10, and users wishing to use these new error calculations are encouraged to retrieve their newly recalibrated data from the HST archive.
New COS/FUV Time-Dependent Sensitivity and Photometric Throughput Tables Delivered
The sensitivity of the COS FUV detector declines with time. The time-dependent sensitivity (TDS) for each mode is modeled as a piecewise linear function of time and is assumed to be a smooth function of wavelength. This TDS is characterized in a TDSTAB reference file. After the initial conversion of counts to fluxes, the fluxes are further modified by a factor that changes with time. The TDS models are derived from observations of flux standards, which are monitored every two months for the FUV. The monitoring program uses the following subset of available cenwaves to track the standard modes TDS: 1222, 1291, and 1327 for the G130M grating; 1577 and 1623 for the G160M grating; and 1105 and 1280 for the G140L grating.
Starting in mid 2019, the TDS monitoring program showed systematic deviations from the model. For the FUVA detector, the decline was about 2%/yr steeper than predicted at the short wavelength end and about 2%/yr shallower at the long wavelength end. For the FUVB detector, the decline was slightly steeper than the prediction at all wavelengths. In order to correct these deviations, new parameters for the standard modes were derived. The model required a new temporal breakpoint at 2019.0 to account for changes in the slope since then.
Previously, the slopes for all monitored cenwaves were averaged to derive the slopes for each grating. The monitoring, however, has revealed that the slopes depend strongly on detector position, rather than only on wavelength. To account for this, the TDSTAB is now a hybrid cenwave-dependent file. For monitored cenwaves, the slopes are determined only from monitoring of that cenwave. For unmonitored cenwaves, slopes continue to be averaged as before.
Support of cenwave-dependent TDSTABs was introduced in CalCOS 3.3.9. The new TDS parameters were used to update the sensitivities at the reference time at lifetime positions (LP) 3 and 4, so the corresponding FLUXTAB files were also updated. A new FUV TDSTAB and new FUV FLUXTABs for LP3 and LP4 were delivered on 2020 October 6. Users who have obtained FUV spectra since 2017 October 2 are encouraged to retrieve their newly recalibrated data from the HST archive.