3 Observations

The observations reported in this paper were acquired with the 1.5 m, f/20 TIRGO telescope from March 13 to April 13, 1997, when 32 nights were allocated to the present project. Only 22/32 nights were useful, 16 of which were entirely or partly photometric. The seeing ranged from 1.5 to 3.5 arcsec (FWHM) with a mean of 2.4 arcsec, as shown in Fig. 1. These seeing conditions are mostly due to the large pixels ($\sim$1arcsec) of ARNICA at TIRGO, and as such represent a necessary disadvantage, because they also provide the large field-of-view (4.1$\times$4.1 arcmin²) fundamental for our observations.

Figure 1: The seeing distribution

The folded Cassegrain focus of the telescope is equipped with the Arcetri NIR camera, ARNICA, which relies on a 256² NICMOS3 array detector (Lisi et al. 1993; Lisi et al. 1996; Hunt et al. 1996). With a pixel size of 0.97 arcsec, the field-of-view is 4.1$\times$4.1 arcmin². Obtaining a satisfactory background subtraction is the main difficulty of NIR observations. At 1.65$\mu$m, the sky brightness at the Gornergrat is $\approx$13.0 - 13.5 mag arcsec^-2, and varies on time scales comparable with the typical duration of an observation (e.g., Wainscoat & Cowie 1992). To achieve a brightness limit 8 mag arcsec^-2 fainter than the sky requires an image in which the deviations from flatness do not exceed 0.06%. Thus, data acquisition techniques must be able to monitor the sky fluctuations and data reduction must take these into account.

To this end, we used two types of pointing sequences, or "mosaics'', according to the size of the galaxy. Galaxies with an optical diameter $\ge$ 1 arcmin were observed using a sequence in which half of the time is devoted to the target, and half to the surrounding sky (hereafter denoted as a type "A'' mosaic). Typically eight on-source pointings were alternated with eight on the sky, positioned along a circular path around the galaxy and offset by 4 arcmin from the source position. The on-source positions were "dithered'' by 10 arcsec in order to facilitate the elimination of bad pixels. To save telescope time, small galaxies (with an optical diameter < 1 arcmin were observed with a pointing sequence consisting of nine pointings along a circular path and displaced from one another by 1 arcmin such that the target galaxy is always in the field (hereafter denoted as a type "B'' mosaic). On-chip integration times were set to 6 s to avoid saturation but to ensure background-limited performance. Total integration times on-source were typically 400 s and ranged from 150 to 950 s.

Table 1 gives the parameters relevant to the NIR observations as follows:
Column 10: number of frames $N_{\rm f}$ combined to form the final image (depending on the adopted mosaic).
Column 11: number of elementary observations (coadds) $N_{\rm c}$. The total integration time (in seconds) is the product of the number of coadds $N_{\rm c}$ times the number of combined frames $N_{\rm f}$ times the on-chip integration time $t_{{\rm int}}$ which was set to 6 s.
Column 12: seeing (in pixels, with 0.97 arcsec per pixel).
Column 13: adopted zero point (mag/s).
Columns 14-17: observing dates (day-month-1997).

  

Table 1: The program galaxies. This is a one page sample. The entire table containing 558 entries is only available in electronic format

3.1 Photometric calibration

Observations of the standard stars HD 84800 (H = 7.53 mag) and HD 129653 (H = 6.92 mag) from Elias et al. (1982) were taken hourly throughout the nights for calibration purposes. The calibration stars were observed with a third pointing sequence (mosaic type "C'') which consisted of five positions, starting with the star near the center of the array, followed by positioning the star in each of the four quadrants of the array. The telescope was defocussed to avoid saturation. During the photometric nights (i.e. those of march 20, 21, 25, 27, 29, 31 and of April 1, 2, 6, 7 and 13) the typical uncertainty in the nightly calibration is 0.05 mag. Those nights were mostly used to observe galaxies with unknown H band photometry.

In the remaining non-photometric or marginally photometric nights we did not rely on the photometric calibration; instead we conservatively observed only those galaxies with aperture photometry available from the literature. For these galaxies the calibration was derived from the "virtual aperture photometry'' (see Sect. 4.1 below).

Several (108) galaxies were observed in more than one night (see Cols. 14-17 in Table 1). This was done either to check the photometric consistency, because some observations were taken in non-photometric conditions, or, if the objects were fainter than average, to obtain a longer integration. For these objects we obtained an average frame by combining the various sets of observations, and using the zero point from the frame(s) taken under photometric conditions.

3.2 Data reduction procedures

Since dome exposures cannot be obtained at TIRGO due to the vicinity of the telescope secondary ring to the dome, the multiplicative correction for the system response, or flat-field (FF), was derived daily from the observations. Many tens (typically 30 to 40) sky frames taken in similar conditions throughout the night were combined with a median filter. The combined frame, normalized to its median counts was used as the flat-field frame.

The reduction procedure varied according to the type of mosaic. For type "A'' mosaics, the (usually eight) sky exposures were combined with a median algorithm to form a median sky. For type "B'' and "C'' mosaics, the median sky was determined by combining all the frames in the pointing sequence. The median algorithm is necessary to remove unwanted star and galaxy images in the median sky frames. The median sky was first normalized to its median, then multiplied by the median counts of the individual target frames. Finally this rescaled frame was subtracted from each of the target observations. Such a procedure accounts for temporal variations in the sky level which are on the order of 5% during a pointing sequence, but introduces an additive offset which is subsequently removed (see below). The sky-subtracted target frames are then divided by the FF frame. Each of these corrected frames was then analyzed for low-spatial-frequency gradients, and if necessary, fitted with a two-dimensional 3 degree polynomial which was then subtracted. (We carefully checked that this procedure did not produce artificial features which could disturb the photometry of the target objects). If this process was not effective in removing the spatial gradients, the corresponding frames were rejected from further analysis. Finally, the corrected frames were registered using field stars and combined with a median filter which allows bad pixel removal. Foreground stars were eliminated by manual "editing'' of the target frames. All image reduction and analysis was performed in the IRAF environment and relied on the STSDAS package.

We have assessed the quality of the final images both on small spatial scales, and over the entire array. All of the images are truly background limited, as the noise we measure is the same as that which we would theoretically expect from the statistical fluctuations in the sky background, according to Hunt & Mannucci (1998) (see Fig. 2). The typical pixel to pixel fluctuations are $\sim$ 21 mag arcsec^-2 , i.e. 0.05-0.06% of the sky.

Figure 2: The distribution of the sky rms as a function of integration time. The solid line is the prediction by Hunt & Mannucci (1998)