Astron. Astrophys. 350, 447-456 (1999)

4. Data reduction

The longitudes examined here lie between , and the latitudes between . This longitude range was divided into 4 "blocks", each of which was approximately in length. The preliminary data reduction was performed using standard software, based on the NOD 2 reduction system (Haslam 1974).

At this point the scans had been assembled into several hundred crude maps, each of which was of the order of in size and which had been appropriately calibrated. These maps were then sorted and assembled into both l- and b-scanned maps (l-maps and b-maps) of all 4 blocks. Each of the blocks was examined for interference, bad data points or baseline problems; any affected areas were either corrected or flagged.

During the assembling process, it was carefully checked that no baseline discontinuities existed near the boundaries of each of the small maps. If a discontinuity was detected, the baseline was corrected by the subtraction of an appropriate baselevel from the scan. For Stokes Q and U data, these baselevels were determined by fitting a zeroth- or first-order polynomial function to the scan data, and subtracting this fit. This is equivalent to requiring that the sum of all pixel values in the scan is equal to zero (and also, in the case of a first-order fit, that there is no linear slope along the length of the scan). Whilst the requirement that the sum of pixel values equates to zero will never be exactly satisfied in practice, this condition is met to a good approximation in this instance. (Indeed, this is usually an acceptable approximation whenever the scan length is appreciably greater than the largest structures present in the data.)

Next, the baselevels of each column in the b-maps ( in length) and each row in the l-maps (approximately in length) were adjusted, by using a similar "cross baselining" procedure to that described by Duncan et al. (1995). Briefly, this process involves subtracting each scan in each l-map (b-map) from its corresponding b-map (l-map) pixels, and fitting a low-order polynomial (third order or less) to the residual. Any points differing from the fit by more than were flagged as being associated with sources, and the fit repeated. This second fit was then subtracted from the scan. Such a two-pass approach to the polynomial fitting greatly improved the fit quality in the vicinity of bright sources, which exhibit significant amounts of instrumental polarisation.

4.1. Instrumental polarisation

The instrumentally polarised component of the emission was estimated by Junkes et al. (1987a) to be approximately 0.7%. As with the Parkes 2.4 GHz survey data (Duncan et al. 1997), the instrumental component was detected only in the vicinity of bright, discrete sources, usually appearing in the Stokes Q and U images with the distinctive quadrupole pattern.

Examination of the images revealed that only sources with flux densities exceeding some tens of Jy per beam area produced prominent instrumental polarisation. The strongest such sources appearing in the data can be identified with HII regions such as W31, W33, M17, W43 and W51. As in the Junkes et al. (1987a) work, we have chosen not to remove this instrumental component.

4.2. Combining

Each pair of l- and b-scanned maps was then combined, using the "Plait" software of Emerson & Gräve (1988). Note that, over some isolated regions of the data, we could not obtain both l- and b-scanned observations. Over these areas, some scanning effects may be evident.

The overlap regions of adjacent blocks were examined and it was confirmed that no discontinuities existed at block boundaries; hence, no extra baselevels were added. All the blocks were then combined to form the survey "strip".

At this point, the data were combined with the existing, low-latitude maps of Junkes et al. (1987a). Note that these latter maps consisted of b-scans only. Discontinuities between the low- and high-latitude regions were removed as much as possible, although some isolated regions of discontinuity are still evident.

To improve the signal to noise ratio, both the Stokes Q and the Stokes U surveys were then smoothed with a Gaussian which increased the effective beamwidth to 1.2 times the nominal telescope beamwidth (to ; see Table 1) and appropriately rescaled.

The final polarised intensity images were then formed from the corresponding Stokes Q and U data, in the usual manner. Bias in the polarised intensity images was removed to first order (Wardle & Kronberg 1974; Killeen et al. 1986).

© European Southern Observatory (ESO) 1999

Online publication: October 4, 1999
helpdesk.link@springer.de