2 Spectral observations and data reduction

All results presented below were obtained by observations in a snap-shot mode during two runs with the KPNO 2.1m telescope and the Calar Alto 2.2m telescope (see Table 1).

2.1 Observations with the KPNO 2.1 m telescope

The observations were made with the GoldCam spectrograph used in conjunction with the 3 K $\times$ 1 K CCD detector. We used a 2'' $\times$ 229'' slit with the grating 09 (316 grooves mm^-1) in its first order, and a GG 375 order separation filter. This filter cuts off all second-order contamination for wavelengths blueward of 7400 Å which is the wavelength region of interest here. This instrumental set-up gave a spatial scale along the slit of 0 $.\!\!^{\prime\prime}$75 pixel^-1, a scale perpendicular to the slit of 2.7 Å pixel^-1, a spectral range of 3700-7500 Å and a spectral resolution of $\sim$ 5 Å. These parameters permitted cover simultaneous coverage of the blue and red spectral range with all the lines of interest in a single exposure and with enough spectral resolution to separate important emission lines such as H$\gamma$ $\lambda$4340 and [O III] $\lambda$4363, and H$\alpha$ $\lambda$6563 and [N II] $\lambda$6584. Normally, short exposures were used (5 min) in order to detect strong emission lines, to measure redshifts and make some crude classification.

Reference spectra of an Ar-Ne-He lamp were recorded to provide a wavelength calibration. The spectrophotometric standard star Feige 34 from Massey et al. ([1988]) was observed for the flux calibration at least once a night. No effort was made to orient the slit along the parallactic angle, so line flux ratios could be spectrophotometrically inaccurate. The observations were complemented by dome flats, bias-, and dark frames. The seeing was about 3 $\hbox{$^{\prime\prime}$ }$(FWHM).

2.2 Calar Alto 2.2 m telescope observations

Follow-up spectroscopy with this telescope was conducted as a back-up for a main program which needed photometric conditions. So, the observations presented here were obtained in non-photometric conditions and the absolute flux calibration of the data is unreliable.

The Cassegrain focal reducer CAFOS of the 2.2 m telescope was used with a long slit of 300 $\hbox{$^{\prime\prime}$ }\times$ 3 $\hbox{$^{\prime\prime}$ }$ and a grism of 187 Å mm^-1 linear dispersion. Spectra were recorded on a 2 K $\times$ 2 K Site CCD operated in a 2$\times$1 binned mode (binning only along the dispersion direction), resulting in a spectral resolution of about 20 Å (FWHM), and a wavelength coverage $\lambda = 3700 - 8100$ Å. No order separation filter was installed. The slit orientation was again not aligned with the parallactic angle to keep the duty-cycle high. The exposure times varied between 10 and 15 min depending on the object brightness. The observations were complemented by standard star flux measurements, Hg-He-Cd lamp exposures for wavelength calibration, dome flat-, bias-, and dark-frames. The seeing was between 1.5 and 2.5 $\hbox{$^{\prime\prime}$ }$ (FWHM).

2.3 Data reduction

2.3.1 Reduction of the KPNO 2.1 m telescope data

The KPNO two-dimensional spectra were bias subtracted and flat-field corrected. We then use the IRAF software routines IDENTIFY, REIDENTIFY, FITCOORD, TRANSFORM to do the wavelength calibration and the correction for distortion and tilt for each frame. Then the one-dimensional spectra were extracted from each frame using the APALL routine without weighting. For all objects we extracted the brightest part of the galaxy covering a spatial size of 7 $^{\prime\prime}$. All extracted spectra from the same object were then co-added. Cosmic ray hits have been removed manually. To derive the instrumental response function, we have fitted the observed spectral energy distribution of the standard star Feige 34 with a high-order polynomial.

2.3.2 Reduction of the Calar Alto 2.2 m telescope data

This reduction was fully done at SAO with the standard reduction system MIDAS (Munich Image Data Analysis System, Grosbøl [1989]). We applied the context LONG as follows: bias and dark subtraction, flat-fielding, cosmic-ray removal. After the wavelength mapping, a night sky 2-D background subtraction was performed. 1-D spectra were extracted by adding the consecutive CCD rows centered on the object intensity peak along the slit. Then the corrections for atmospheric extinction and flux calibration were applied. Despite of the non-photometric observing conditions, we corrected the spectra for the instrumental response with a response curve established by observations of the spectrophotometric standard star BD+33$^{\circ}$2642.

2.3.3 Line parameter measurements

In the final spectra, redshifts and line fluxes are measured within MIDAS, applying Gaussian fitting to the emission lines. To determine redshifts for individual galaxies, averages are taken over the prominent individual emission lines (mostly H$\beta$, H$\alpha$, [O III]$\lambda$4959, 5007 Å). The line [O II]$\lambda$3727 Å is not included in the redshift determination since for most of the objects its observed wavelength is determined with significantly larger uncertainties due to the extrapolation of the linear scale below the first line of the reference spectrum (He I$\lambda$3889 Å). However [O II]$\lambda$3727 Å was used to determine the redshift in rare cases where it is the only strong emission line. The errors of the redshift in such cases can be several times larger than the typical one (compare Table 2).

To improve the accuracy of the redshift determination for the Calar Alto spectra, and further, to reduce possible small systematic shifts in the zero point of the wavelength calibration, we additionally checked the wavelengths of night sky emission lines on the 2-D spectra at the position of the object spectrum. If some measurable shift was detected it was incorporated in measurements of emission line positions.

The emission line fluxes are computed by summing up the pixel intensities inside the line region applying standard MIDAS program tools. For all spectra, the individual emission line fluxes of the H$\alpha$, [N II] $\lambda\lambda$6548, 6583 Å and [S II] $\lambda\lambda$6716, 6731 Å line blends are obtained by summing up pixel intensities over the total blend and then modeling the individual line fluxes using Gaussian fitting.