Pixel area variation in CCDs and implications for precise photometric calibration

Images with smooth and moderately flat illumination are commonly used to calibrate pixel to pixel sensitivity variation without consideration that some structure on short spatial scales may be due to reallocation of area between pixels. Errors in the position of pixel boundaries have the effect of repartitioning charge between pixels but do not affect the total charge collected. Since the resulting errors tend to cancel when combining signal from adjacent pixels, this effect has gone largely unnoticed. However, proposed astronomical surveys from space such as SNAP, which strive to achieve high photometric precision with coarse sampling of the point spread function, must take this pixel area variation into account. We present simple analysis techniques to identify how much flat field structure is due to variations in pixel area rather then sensitivity as a function of row/column direction and spatial frequency. We present results for CCDs made with radically different technologies and pixel sizes, which suggest that pixel size variation in the column direction may dominate QE variations on short spatial scales for all CCDs. We posit that a refinement to flat field calibration methods is required wherein large amplitude QE drop outs would be corrected at full spatial resolution, while the low amplitude structure would be spatially filtered along columns to allow mid frequency sensitivity variations to be well corrected without erroneously correcting for high spatial frequencies where pixel area variation exceeds sensitivity variation.