This effect can be seen as an offset between the odd and even columns of the array. It is a function of the flux of the array and it evolves with time. The amplitude also depends on the quadrant and can be as large as 6%.
Before March 30th, 2001, the effect was only visible when the flux on the array was above 10,000 ADUs. It was possible to avoid it by ensuring that the flux was never above this. However, after this date and for reasons which are unknown, the effect became visible at much lower flux levels. On July 24th 2001, the read speed of the array was slowed down. The odd-even column effect was reduced to less than 1% over the flux range 0-20000 ADU. As a direct consequence of this, the minimum integration time of the array was set 3.55 seconds.
On August 24th, 2001, the effect re-appeared, even with the slower readspeed.
On October 20th, 2001, during an instrument intervention, the problem disappeared. On December 15th, 2001 the effect re-appeared.
Since the effect depends on the quadrant, each quadrant has to be treated separately. A very effective way of removing the effect is to take a Fourier Transform of individual quadrants and mask the one pixel in Fourier Space that corresponds to a spatial frequency of one pixel. The effect of applying the correction on the photometry is less that a hundredth of a magnitude. This works very well for the lower right, upper left and upper right quadrants. It works less well for the lower right quadrant where there appears to be a y dependence to the effect. The y dependence can be removed by masking additional pixels.
An eclipse routine to remove the effect in imaging data has been created. It is called isaacp odd-even. The routine works less well for spectroscipic data.
The effect is harder to remove from spectroscopic data, and this is especially true for the flat fields. We have retaken all flat fields for all the spectroscopic set ups that were used between May 12th and July 24th 2001. If you suspect that your spectroscopic flats are contaminated by the odd-even column effect, then you should search the archive for data that was taken with the same instrument setup after July 24th.