3.2. Bias Subtraction
As discussed in Section 2 above, we use the strength of the solar Fraunhofer lines in the ZL spectrum to measure the ZL flux. This is a differential measurement in the dispersion direction; uniform, additive offsets due to bias or dark current will not affect our results beyond the small effect on flux calibration. Structure in either bias or dark current, however, will increase random errors in the results. For example, sharp features in the dispersion direction will increase the rms errors in the average strength of the Fraunhofer absorption lines. Also, because we extract a single, one dimensional spectrum from every two dimensional image by averaging over the full spatial extent of the spectrum, any fluctuations in bias level in the spatial direction add random errors to the averaged flux found at any wavelength.
To remove spatial variations in both directions, the bias correction was done in three steps. Variations in the dispersion direction were subtracted using the 150 column overscan region. The overscan was fitted with a Savitsky-Golay routine (Press et al. 1992), which follows rapid jumps in the mean level, and the fit was then subtracted from each column. The mean bias level was then removed by subtracting a single mean bias value, which is the average of the bias row (row 1024) in each frame. Finally, bias variations over the chip were found to be very stable after the overscan column and mean bias level were subtracted from every image. These variations were thus removed by subtracting a "superbias" image, which is the average of 250, overscan- and bias-subtracted bias frames. To verify the accuracy of this procedure, we test-reduced 50 bias frames which had not been included in the superbias. After bias subtraction, these test frames had an average value of 0.005 DN with an rms scatter of 0.03 DN and no residual systematic structure.