3 Data reduction

  The initial reductions of all CCD images (bias subtraction, flatfield correction, removal of bad columns) were done using the standard tools in IRAF.

3.1 Standard calibrations

The instrumental magnitudes for the standard stars were obtained on the CCD images using the phot task in DAOPHOT (Stetson 1987) with an aperture radius of 20 pixels. The extinction correction was then applied directly to the raw instrumental magnitudes. The standard calibrations were carried out using the photcal package in IRAF with the transformation equations assumed to be of the linear form Here the magnitudes written in capital letters correspond to the standard system, and magnitudes written with small letters are in the instrumental system, corrected for extinction only.

  

Table 1: The standard transformation coefficients for each of the four observing runs. N is the total number of observations of individual stars used for the transformations. The number of stars $N_{\rm s}$ actually involved is always smaller, some of the fields being observed several times. The rms values refer to the scatter around the standard transformations

The coefficients of the standard transformation equations are given in Table 1 for each of the four observing runs. NOT1 and NOT2 refer to observing runs at the Nordic Optical Telescope in May 1997 and October 1997, respectively, and DK1 and DK2 refer to runs at the Danish 1.54 m telescope in September 1997 and February 1998. The CCD at the NOT was interchanged between the two runs which may explain the changes in the instrumental system from one run to the other, while the quantum efficiency of the CCD at the 1.54 m telescope was notoriously unstable during the summer 1997/1998, and therefore the changes in the zero-point constants are hardly surprising. The rms scatter of the standard star observations relative to the standard system after the transformation is generally of the order of a few times 0.01 mag, with a slightly larger scatter in the U-B transformation. Thus, the accuracy of the standard transformations is sufficient for our purpose.

3.2 Reduction of science data

Following the initial reductions, the three exposures in each filter were aligned and combined into a single image, yielding an effective integration time of 15 minutes in BVRI and H$\alpha$ and 60 minutes in U. As both the DFOSC and ALFOSC instruments can be rotated around the optical axis it was sometimes necessary to counter-rotate the individual exposures before they were combined (using the IRAF task rotate), even if attempts were made to position the instruments in the same way from night to night. This is due to the limited angular resolution of the encoders on the rotators. The necessary offsets and rotations were calculated by manually inspecting each frame and measuring the x-y coordinates of two selected stars using the IRAF task imexamine. By this method it was possible to align the images to a precision of about 0.1 pixel, accurate enough for our purpose. Flux conservation during the rotation was checked by comparing photometry on a synthetically generated image and on a rotated version of the same image. No deviations larger than 0.003 mag were found.

The actual combination of the individual images was done using the task imcombine, with the reject option set to crreject in order to remove cosmic ray events. It was necessary to be particularly careful if the images to be combined had slightly different seeing, because the centres of bright stars could then be "rejected'' by imcombine. We checked that this did not happen by also combining the images using no rejection at all and subtracting the two versions of the combined images from each other, making sure that no residuals were seen at the positions of stars. In general it was possible to avoid this problem by adjusting the snoise ("sensitivity noise'') parameter of imcombine.

As the first step towards detecting point-like objects we then created a smoothed version of a V-band image, using the median filtering task in IRAF and a $5 \times 5$ pixels box median filter. The median-filtered image was then subtracted from the original version, resulting in a residual image that contained no large-scale structure but only point sources superimposed on a uniform background. Point sources were finally located in the residual image using the daofind task in DAOPHOT.

Because star clusters are sometimes seen as slightly extended sources even at the distance of the galaxies in our sample, the photometry had to be obtained as aperture photometry rather than by PSF fitting (see also Östlin et al. 1998). The photometry was carried out directly on the combined images, and not on background subtracted images. In Paper I it was argued that the accuracy obtained by this method was quite satisfactory. An aperture radius of 4 pixels was chosen for the determination of colour indices, while an aperture radius of 8 pixels was used to measure V band magnitudes. It was shown in Paper I (see also e.g. Holtzman et al. 1996) that aperture corrections tend to cancel out for colour indices, thus making it possible to measure accurate colours even with a quite small aperture.

The pixel size on the NOT images is half of that on the 1.54 m. images, but by coincidence the seeing on the NOT data was also, in general, twice as good as on the 1.54 m. data, so we used the same aperture radii in pixels for the NOT and 1.54 m. observations. Aperture corrections relative to the 20-pixels aperture radius used for the standard star observations were applied to the photometry, based on the assumption that the objects were point sources. The aperture corrections were derived from photometry of a few bright, isolated stars using the mkapfile task in DAOPHOT, and amounted in general to a few tenths of a magnitude. The photometry was corrected for Galactic foreground reddening using the extinction values given in the RC3 catalogue (de Vaucouleurs et al. 1991).

3.3 Identification of star clusters

 

  

Figure 1: Colour-magnitude diagrams for objects in the NGC 2997 field. Top: Objects at distance larger than 600 pixels (4 arcmin) from the centre of the galaxy. Bottom: Objects at distance closer than 600 pixels to the centre of the galaxy

In short, young star cluster candidates were identified as compact bright, blue objects without any H$\alpha$ emission. The specific criteria invoked to select cluster candidates were the following:

The first criterion was that the objects should have B-V < 0.45. In this way most foreground stars are eliminated, as demonstrated in Fig. 1 which shows colour-magnitude diagrams for objects in the field containing the galaxy NGC 2997. The NGC 2997 field is relatively rich in foreground stars, while the galaxy itself occupies only the central part of the frame, and therefore this field serves as an illustrative example. The difference between the colour-magnitude diagrams of the central part of the field (where the galaxy is located) and the outer parts (only foreground) is striking. In fact some of the foreground stars are slightly bluer than B-V = 0.45, so the limit is not completely rigorous and was adjusted slightly from case to case by inspecting diagrams like the one in Fig. 1. A B-V colour of 0.45 for a star cluster corresponds to an age of about 500 Myr (Girardi et al. 1995), so we are really just sampling young objects.

Secondly, the cluster candidates were divided into two groups, a "red'' group with U-B > -0.4 and a "blue'' group with $U-B \le -0.4$,and an absolute magnitude limit was applied. For a given mass, clusters in the "blue'' group will be more luminous than clusters in the "red'' group, and therefore a brighter magnitude cut-off could be applied to the "blue'' clusters. Specifically, the limits were set to M_V < -8.5 for the "red'' group and to M_V < -9.5 for the "blue'' group (distance moduli for the galaxies are given in Paper I).

  

Figure 2: H$\alpha-R$ vs. U-B for objects brighter than MV = -9.5 in the NGC 5236 field

Third, it was necessary to exclude HII regions. In Fig. 2 we show H$\alpha-R$ vs. U-B for all objects brighter than M_V = -9.5 in the NGC 5236 field, chosen as an example because of the large number of sources in this galaxy. The H$\alpha$ photometry was not standard calibrated, so the magnitude scale on the y-axis in Fig. 2 is purely instrumental. It is clear from Fig. 2 that the objects with an excess in H$\alpha$ appear around U-B = -0.4, while H$\alpha-R$ is rather constant for redder objects. Denoting the average H$\alpha-R$ value at U-B > -0.4 by $({\rm H}\alpha-R)_{\rm ref}$, we chose to cut away objects with ${\rm H}\alpha-R < ({\rm H}\alpha-R)_{\rm ref} - 0.5$.The limit could have been placed closer to $({\rm H}\alpha-R)_{\rm ref}$, but by choosing a cut-off at 0.5 magnitudes below $({\rm H}\alpha-R)_{\rm ref}$ we allow for some scatter in the photometry as well.

Finally we did a visual inspection of the cluster candidates. In this way objects which were too extended or "fuzzy'' to be star clusters were eliminated (typically star clouds or associations in the spiral arms of the galaxies).

  

Figure 3: V-band images of a few clusters. The first image in each row is the PSF, followed by 4 clusters

Photometric data are given for each cluster in Table 4, along with the effective radius $R_{\rm e}$ (in pc) as derived by ishape (Sect. 4). A few examples of clusters are shown in Fig. 3.

A table showing the total number of star clusters identified in each of the galaxies in our sample was given in Paper I, and we refer to that paper for a detailed discussion of the properties of the cluster systems.

  

Figure 4: V-band images of the galaxies. The bar in the upper left corner of each image corresponds to 1 arcminute. North is up and east to the left in all images

  

Figure 5: See the caption to Fig. 4