COBE FIRAS Products

Gradient Kernal

The FIRAS team compared observations grouped into the same sky pixel. Because of the finite sizes of the pixel and the FIRAS beam (Explanatory Supplement Section 7.9.4), spatial gradients in the source spectra in the vicinity of the pixel will produce variations which are not intrinsic to the FIRAS instrument. Crosstalk between these variations and the model functions will introduce errors to the offset spectra, hereafter referred to as "stripes" . Previously, this source of error was handled by excluding data from pixels within 10° of the Galactic plane, under the assumption that spatial gradients at higher Galactic latitudes were negligibly small. For the final release, the FIRAS team added model functions to account for the spatial gradients, enabling us to include more data in the destriper, and assuring that gradient errors at higher Galactic latitudes truly are suppressed.
The gradient functions were derived from the DIRBE data. First the high resolution DIRBE data were convolved with the FIRAS beam (Section 7.9.4) centered on each of the 6144 skycube pixels. Then a six parameter quadratic model was used to fit each pixel and its nearest eight neighbors. That model was used to estimate the DIRBE intensity for each coadd. Finally the weighted average of the coadd estimates for each pixel were subtracted from the estimates for that pixel. This was done so that there is no net signal in the model. Functions were formed from DIRBE bands 8, 9, and 10 (100, 140 and 240 ¯m) although only the functions for bands 9 and 10 were used in the released destriper models. F­statistic analysis indicated that two DIRBE gradient stripes were optimal. The FIRAS team selected the Band 10 and Band 9 functions.