This FLAT is an improved version of the Investigation Definition Team (IDT) ???? flat, e3??????, where a correction has been applied to spatial scales between ~7 pixels and 800 pixels. The resulting flat is effectively the F606W preflight thermal-vaccuum flat, with an improved on-orbit illumination pattern correction applied to scales >7 pixels. On the very largest scales, the chip-to-chip normalization of the IDT flat, e3??????, has been preserved. The improved illumination pattern correction was derived using F502N Earthflats, and the previous F502N reference flat, e3809354u, as follows: The following F502N EARTHFLATs taken after 1 May 1994 were used (i.e. post-cooldown). These are all 1.2 sec. exposures taken on shutter "A" in gain 15. They have been corrected for shutter shading. They were selected to have count levels in the range 600 to 3000 DN. These have been carefully selected to have less that 1.5% peak-to-peak streaks. This was done by an iterative process: The individual EARTHFLATs were flattened with the previous F502N refernence flat, and examined. Ones with >3% peak-to-peak streaks were discarded, and the remaining ones were used to generate a flat. The new flat was used to flatten the EARTHCALs again. Ones with streaks >1.5% were discarded, and the remaining ones are used here. Dataset Dateobs. Days from nearest decon. u2a30s03t u2a31304t u2a31a04p u2a31s03t u2a31u03t u2a31w03t u2a31z04t 94 Jun 17 +4 u2a32504t u2a32804t u2a32e04t u2a32m03t u2a31504t 94 May 4 +11 u2a31903t u2a31k04t u2a31s04t u2a31z03t u2a32004t u2a32404t u2a32703t u2a32k03t u2a32o03t These were combined using the STSDAS/IRAF task XSTREAKFLAT. The PC pixels of the resulting streakflat were then scaled by the ratio of the WF/PC pixel (4.7835) areas to bring the PC up to the intensity of the WF CCD's. This streakflat was then normalised and inverted by running it through the STSDAS/IRAF task NORMCLIP, with the maximum threshold set to 10. Bad columns and pixels were then removed by interpolation using the WFIXUP task. This image was divided by the IDT F502N flatfield, e3809354u. Single pixels in the correction image which differed from surrounding pixels by more than ~2% were reset to the local mean. This was an effort to minimize the impact of bad pixels on surrounding data. (These discrepant pixels were also marked in the data quality file for the output flat.) The result was then smoothed by convolving it with sigma=3 pixel Gaussian function over most of the image; near the pyramid vertices smoothing was done in only one dimension (parallel to the vertice). Each chip was then divided by the average pixel value in it's central 400x400 region, (PC1=0.9863, WF2=0.9792, WF3=0.9772, WF4=0.9780), so as to set the mean correction to unity over the center region of each CCD. The final step was to multiply this illumination pattern correction image into the current F606W reference flat, e391433bu. Regions of the output flat exceeding 2.0 were also marked in the data quality file as being suspect (regions in pyramid shadows). The RMS correction (over central 600x600 pixels) and maximum corrections (over entire CCD) applied to the ITD flat are given in table below. The largest corrections are near the outer edges of CCDs WF2 and WF3. The RMS corrections are several times the expected photon noise (recall the correction image is smoothed). Group RMS Correction Max. Correction ------ -------------- ---------------- 1 0.26% 1% 2 0.28% 4% 3 0.21% 8% 4 0.21% 1% To test the stability of this correction image, we derived two independent versions, each using half of the above listed F502N EARTHFLATs. The two final correction images were in excellent agreement. The table below gives their difference in percent over the central approx. 400x400 pixels of each group. The pixel-to-pixel differences are consistent with photon noise. Group ( Test1 - Test2 ) in percent ------ -------------------------------------------- RMS (unsmoothed) RMS (10 pixel smooth) 1 0.27% 0.05% 2 0.29% 0.03% 3 0.23% 0.03% 4 0.23% 0.02%