Jun 262014

Analysis of the IR channel observations of Frontier Fields data has revealed that some of the observations contain a time-variable background signal.

We know that the background signal rate is changing with time thanks to the way in which the IR channel data are collected. Each IR observation is produced using multiple non-destructive readouts of the detector. In the case of Frontier Fields, we usually have 14 – 16 readouts (or “reads”, in WFC3 jargon) of the detector for each exposure. With this method, we can watch the signal in a given pixel grow with time throughout the observation.

On the Frontier Fields project, one of the first data quality checks we perform is to examine the signal level versus time for each IR exposure. In order to produce high signal-to-noise plots, we typically will calculate the mean signal across the detector for each of the 14-16 reads, and plot these signals versus time. We expect to see nice, straight lines with a constant slope, such as that in Figure 1.

An example of a Frontier Fields WFC3/IR exposure without time-variable background.

Figure 1:  Examples of two Frontier Fields WFC3/IR exposures with nearly linear signal accumulation.

But instead, in some cases, we see something like that shown in Figure 2.  In this case, the curves indicate that the mean signal rate was higher for the first half of the exposure than during the second half. Comparison with other, non-varying exposures indicates that the signal rate the in the second half of the exposure is the expected signal rate. This implies that there is some amount of “extra” signal present during the early reads of the exposure, and that this extra signal slowly turns off as the exposure progresses.

Figure 2:  Bad ramps

Figure 2: An example of the mean signal rate changing during the exposure.

Further investigation has revealed that two sources are responsible for this variable background.

The first, seen in Figure 3, is a scattered light which falls along the left side of the detector at times when HST is pointed close to the bright limb of the Earth. A more detailed explanation of this scattered light is given in Section 6.10 of the WFC3 Data Handbook. This glint appears in only a small number of the Frontier Fields observations, and then usually only in the final ~100 seconds of a particular observation.

Figure 3: Scattered light on the left side of the image is due to HST pointing at the bright limb of the Earth.

Figure 3: Scattered light on the left side of the image is due to HST pointing at the bright limb of the Earth.

The second source of time variable background has been traced to atmospheric helium emission (at 10,830 angstroms) occurring along the line of sight, only when HST is above the bright side of the Earth. For more details on the orbital geometry and the emission itself, see Instrument Science Report WFC3-2014-03 by Gabe Brammer.  Given the limited wavelength range of the emission, this excess flux only affects the Frontier Field observations taken with the F105W filter. All of our other filters have a short-wavelength cutoff that is redward of the emission.

So, how do we deal with these two types of “extra” signal in our data? We have come up with a method that we believe does a good job of removing both. For each pixel, we fit a function to the signal versus time, using all of the reads.  This function models how a pixel should behave in the presence of a constant signal throughout the exposure. We then compare this model fit to the actual signal values in the exposure. In the case of a pixel contaminated by one of the two types of time-varying background signal, the resulting fit will not be very good. In this case, we then ignore the first read of the exposure, and try re-fitting the function to the signal in reads 2 through the end of the exposure. If the fit is still not good, then we ignore reads 1 and 2, and fit our function from read 3 to the end. This process is repeated until we arrive at a good fit.

With a good fit produced using only the good reads of the exposure, we can now predict what the signal should be in all of the reads. At each read then, the difference between the measured signal and our modeled prediction represents the amount of “extra” signal in that read. At this point, we could simply remove this excess signal from our pixel, but this would make for a very noisy result since most of the pixels on the detector see only the background signal and therefore have a relatively low signal to noise ratio. In order to produce a higher quality correction, we perform one more step. Once we have determined the amount of extra signal present in all of the pixels in all of the reads, we go read-by-read and smooth the extra signal across the detector. This smoothed signal is then subtracted from the original exposure, resulting in an observation which has been cleaned of the excess signal. Figure 4 shows the mean signal for the same two exposures as Figure 2, after our correction has been performed.

Figure 4:  Corrected ramps

Figure 4: The mean signal after all ramps are corrected for the variable sky in the same exposures used in Figure 2.

The corrected signal is not perfectly linear with time, but the majority of the extra background signal has been removed.

Bryan Hilbert – Frontier Fields Science Data Products Team Member – WFC3/IR

 Posted by at 12:21 pm