Sky Subtraction

The final CCD frame consists of two sets of spectra, one set taken at the object position and one set taken at the sky position (figure

). The object and sky spectra are separated on the CCD image by the number of rows equal to the pixel offset used for the shuffle. We obtained data using a 700 pixel shuffle (roughly 1/3 of the total size of the array). Simply shifting the frame by 700 pixels and subtracting the shifted frame from the original data results in excellent sky subtraction, as shown in figure

Figures

and

show the effect of the N&S subtraction on a single row in two different wavelength regions. It can be seen that subtraction is possible to the shot-noise limit of the sky, and that the residuals of the strong emission lines do not show strong `derivatives' that generally result from standard background model subtraction techniques.

Experimentation with individual nod exposures of 15 sec, 30 sec, 60 sec, 120 sec, and 180 sec, suggests (as one might expect) that the shorter times sample sky variations better and result in better sky subtraction (figure

). In general, we found that individual nod exposures shorter than 1 min are preferable; nod exposures as long as 3 min work, but the OH removal leaves some residuals, suggesting that the lines vary on timescales 1min. Table

shows sample statistics for two BRACKET mode observations: a `short' nod exposure (30 sec 30 nods) and a `long' nod exposure (180 sec 5 nods). Although the statistics are limited, it can be seen that the mean sky value is closer to zero for the `short' nod compared to the `long' nod. However, the RMS values appear to be larger for the `short' nod exposure. The reason for this is unclear at present: it could be due to noise introduced by the shuffling (the `short' nod exposure has 6 times as many nods as the `long' nod exposure) and partly due to the longer total integration time (i.e., the time between the start of the observation and the readout) for the `short' nod exposure (2759 sec for `short' versus 2094 sec for `long'). A comparison of the histogram of counts in two frames:

	storage region	inbetween obj&sky	bottom of frame
number of pixels	419211	209500	90400
a0077 (180s x 5 nods):	0.45+/-1.87	2.93+/-2.29	3.05+/-2.25
a0078 (30s x 30 nods):	0.49+/-1.81	4.38+/-2.56	4.09+/-2.51

**Figure:** Histograms of the bias-subtracted counts in three regions of the charge-shuffled CCD frame. The top, middle and bottom panels show the count distribution in the top (storage) region, middle (inbetween the object and sky spectra) region and bottom (below the sky spectra) region. The black and red lines show the count distributions in the frames a0077.fits (which had 180 sec 5 nods) and a0078.fits (which had 30 sec 30 nods) respectively.

Comparison of Short and Long Nod Exposures

Region	Mean	RMS	Median	Mode	Mean	RMS	Median	Mode
900:1180,225:245	0.721	10.60	0.565	0.448	1.522	9.23	0.717	2.636
700:1000,225:245	0.620	10.57	0.332	0.348	1.402	9.66	0.800	-0.982
900:1100,18:38	0.332	11.38	-0.252	-3.818	1.256	10.12	1.198	-0.111
500:900,18:38	-0.067	11.16	-0.286	4.544	0.870	10.07	0.591	-1.014
850:1200,137:156	1.628	11.05	1.195	-3.791	1.878	9.80	1.551	0.285
1000:1350,137:156	1.322	10.32	1.366	1.620	1.846	8.97	1.053	0.177