Instrument Science Report ACS 2013-02 Column Dependency in Charge Transfer Efficiency Correction S. Ogaz, J. Anderson A. Maybhate, L. Smith July 2, 2013 ABSTRACT In 2010 Anderson and Bedin devised a pixel-based algorithm to correct the charge transfer efficiency (CTE) losses in the Wide Field Channel (WFC) of the Advanced Camera for Surveys (ACS). There have been several improvements to the CTE code throughout its development. In this ISR we will discuss the modification made to the CTE code to account for the column to column variation in the number of Y charge traps. For WFC we have found that 81% of columns fall within 10% of the average, and 96% fall within 20%. These column-specific measurements have been incorporated into the new version of CALACS. Introduction ACS has seen a significant loss in CTE over its lifetime. This is due to the radiation damage it receives while in orbit, which creates electron traps in the CCD. These traps catch the electrons passed down from pixel to pixel during readout and release them at a later time, creating trails in the final image. An algorithm was created in 2010 by Anderson and Bedin that is able to correct the trailing that occurs in ACS images. This correction was added to the ACS pipeline, CALACS, in 2012. Since the first implementation of the CTE c 2008 The Association of Universities for Research in Astronomy, Inc. All Rights Reserved. Copyright correction several improvements to the code have been made. One such improvement is a column-based adjustment to the level of the CTE correction. Because of the stochastic nature of trap creation in CCDs, we expect that some columns will have more traps, and therefore more CTE losses, than other columns. This idea led to a study of this variation in the hopes that we could get an accurate measure of CTE in each column and feed that back into the correction algorithm. Measuring the difference in CTE between columns was a somewhat delicate process. In order to be sure we were measuring the column sensitivity due to a variation in the number of traps and not the variation in signal, we needed images with a consistent and even signal level. We also needed to be able to separate the counts from the CTE trails from the source counts. We were able to fulfill both these requirements with ACS/WFC flat field images. This also allowed us to analyze the column dependency over each year of ACS operation starting in 2002. Data Table 1 summarizes the flat field data that were used for this study. The high level of counts and low variation in these frames mean that any difference in CTE trails comes from a difference in the number of traps present. The actual pixels we used to measure the CTE trails are located in the virtual overscan of the image. The virtual overscan of any image contains only two sources of counts: the bias and CTE trails from the image area of the flat fields. This makes these pixels an ideal place to measure trails. In addition, the trails in this area are created by the image pixels furthest from the readout in the Y direction. These pixels pass every trap present in the column during readout, making their trails an accurate representation of the total number of traps in that column. Figure 1 is a schematic of the top left corner of WFC2, and shows the exact pixels used in our analysis. The bottom pixel of the virtual overscan region is the first pixel of the trail from the image portion of the chip. As shown in Anderson & Bedin (2010) the trail drops off steeply with distance from the initial flux pixel. For simplicity we take the first pixel of trail as a measurement of the relative CTE trailing in each column. Figure 2 shows a cutout of the virtual overscan portion of a WFC2 flat field image in ds9. The trails can be clearly seen at the edge of the CCDs. 2 Year 2002 2003 2004 2005 2006 2007 2009 2010 2011 Number Median of Flats of Pixels 11 21 18 6 6 3 5 15 6 48,932 49,719 49,456 49,170 50,986 59,636 60,264 60,126 60,058 Standard Deviation of Pixels 7,347 7,409 7,369 7,325 7,370 8,737 9,005 8,982 8,980 Table 1: Number of flat files used for each year, along with the median and standard deviation of all the pixels in all flats from one year. This table shows the low variation in the flat field images. Analysis Figure 3 shows the average distribution about the average of bias-subtracted CTE-trail first pixels six flat field images we have for the year 2009. We estimated the bias level independently for each column by averaging the first 11 pixels of the virtual overscan region (see Figure 1). This plot shows a clear correlation between column and CTE trail value. For each column we took an average of all points in one year, using the standard deviation as the error value. This average was then compared to the boxcar average of the surrounding 50 pixels (25 pixels to the left and right), and a normalized percent difference was taken. Figure 4 shows this for a standard column in the WFC2 CCD, column 2603 for the years 2002 to 2011. For the years 2009 to 2012 the trail in column 2603 consistently falls at about 130% of the average. We then condensed the data from all years into a single relative-CTE estimate for each column. By visual inspection we determined that the trail values for years 2002 2007 had large error bars and a large variation over time. This is expected as the column variations due to CTE loses should not be significant over this time period, resulting in low signal to noise. Therefore, to obtain our final values we took an average of the 2009 to 2011 data points only. 3 Fig. 1.— Schematic of CCD 2 of WFC/ACS. The purple area is the science image and the top blue area is the virtual overscan area. Each trail pixel has been bias subtracted using the average of the bias pixels shown. Also shown are the range of pixels used to take a boxcar average. Fig. 2.— Cutout of a raw flat image, taken from the virtual overscan region of columns 2874-3020. White represents high intensity, black represents low intensity. At this stretch it is possible to see the various amounts of trailing in the virtual overscan caused by the varying number of traps in each column of the CCD. Figure 5 gives our final results, with Figure 6 showing a close up of columns 2650 to 2700. 96% of the columns lie between 1.2 and 0.8 for both CCD chips on WFC, meaning a typical column does not vary from the average by more than 20%. These final values scale the level of CTE correction for each column with the number of traps in each column. The extreme outliers (i.e. above 200% difference) have been confirmed as hot columns. 4 Fig. 3.— Column number versus the ratio of the trail in column 2603 to the boxcar average. Data for each column is taken from the six 2009 flat images, so that each column is composed of six data points. In order to test the column correction, we ran the CTE correction, with and without the column adjustment, on flats taken in 2012. These flats were not part of the original column dependency measurement. The following files were used: jbwa07p6q, jbwa07p8q, jbwa07paq, jbwa14efq, jbwa14ehq, jbwa14ejq. We processed each flat in the same way as the previous flats, performing our own bias subtraction. Then each file was run through CALACS with all processing steps set to OMIT except DQICORR and PCTECORR. The CTE correction code has been adjusted so that it will run successfully on the virtual overscan regions of an image. The virtual overscans were left in the image (since BLEVCORR was set to OMIT). 5 Fig. 4.— CTE trail strength for column 2603 for the years 2002 to 2012. The x-axis represents the year and the y-axis represents the percent difference of the trail in column 2603 and the boxcar average, normalized to one. As the CCDs receive radiation over time, the number of traps change, resulting in a change in the CTE trails. Once we had two versions of the processed flats (one with the column correction enabled, one with it disabled), we compared the first pixel of trail in the virtual overscan. This comparison was done by fitting a line to the first pixel of trail for each column in all six images and looking at the residuals of the fit. As the original column adjustments were done with a boxcar average, we chose to do a linear fit to each amp individually. In theory, if the CTE correction and bias correction were each done perfectly, we would be left with only the read noise and the Poisson noise from the trails. In practice, the CTE correction, with or without the column correction, is designed to err on the side of under-correcting, rather then over-correction. As such there will always be a small percentage of flux left in the trail (Úbeda and Anderson, 2012). Table 2 shows the results of the fits. Unfortunately, we did not have enough data to get significant statistics. The column and non column corrected files have approximately the same residuals. 6 Fig. 5.— Final column to column variation for WFC2. These points are the average of the values calculated from 2009 2011. The error bars represent the standard deviation of the measurements taken from each flat field added in quadrature. All outlier values have been confirmed as hot columns. 7 Fig. 6.— An enlargement of Figure 5 showing columns 2650 to 2700. Amp Linear Fit w/o Column Corr Residuals w/o Column Corr Linear Fit w/ Column Corr Residuals w/ Column Corr A B C D −1.55×10−2 x + 155 1.54×10−2 x + 171 −0.97×10−2 x + 192 4.94×10−2 x + 108 2.13×109 1.52×109 7.85×107 1.09×1010 −1.67×10−2 x + 158 1.50×10−2 x + 172 −0.90×10−2 x + 192 5.02×10−2 x + 107 2.16×109 1.51×109 8.02×107 1.09×1010 Table 2: The linear fits and the sum of the residuals for each amp post CTE correction. One set of files has the column correction enabled, one had the column correction disabled. The fits and residuals are in units of electrons. 8 Conclusions This CTE trail column dependencies described here have been implemented in CALACS as a look up table in the pctetab pcte.fits file. The final values get applied as multiplier to the CTE correction, with 1.0 producing no change in the strength of the correction. 96% of the columns lie between 1.2 and 0.8 for both CCD chips on WFC. Values higher than 2.0 were considered to be outliers and set to 1.0. Overall, this study has shown that the variation in the number of traps in each column has a small overall effect on the amount of CTE trailing in an image. The ACS team will continue to monitor the column dependency of the CTE trails and improve the accuracy of the CTE correction. Acknowledgments We would like to thank the ACS Team for their help and input. We would also like to thank Matt Davis, Pey-Lian Lim, and Warren Hack for their help with our CALACS questions. References Anderson, J., & Bedin, L. 2010, PASP., 122, 1035 Úbeda, L., & Anderson, J. 2012, Instrument Science Report ACS 2012-03 (Baltimore: STScI) 9