3. The reliability of observed HI lines
The profiles shown in Fig. 1 need further investigation concerning possible problems associated with instrumentation and data processing. It must be excluded that any spurious intensities would result from the baseline correction or from the stray-radiation correction procedure.
We repeated the reduction of the LDS to check whether the procedures (described by Hartmann 1994) could be responsible for the observed HVD components. In the LDS reduction, third-order baselines had been fitted to emission-free parts of the spectra (Hartmann 1994, Sect. 2.3.4). Additionally, partial baselines had been fitted and subtracted, as had sine-wave ripples (assumed to be caused by standing waves between the telescope dish and receiver).
It seemed conceivable that the removal of partial baselines had produced artifacts which could be interpreted as an HVD component. In repeating the reduction we therefore did not subtract any partial baselines. For the determination of emission-free line channels we introduced additional constraints. For each box of pointed observations within an area of we calculated the variance of the signal. All channels showing fluctuating lines were eliminated from the emission-free regions. Thus any emission which varies noticeably with position (clouds) or time (interference) was suppressed. After elimination of the obvious line emission around 0 we excluded an additional range of 30 at both wings of the line from the emission-free regions. Excluding unreliable channels at the edges of the bandpass, we fitted a third-order baseline over the velocity range -430 380 . Sine-wave ripples were eliminated in a similar way as described by Hartmann (1994). To exclude any possible software problems, the code for the entire reduction procedure was rewritten. The data reduction process was iterated several times to find optimum boundary conditions for the baseline determinations. For the majority of the observations, good baselines could be achieved this way. However, severe interference caused a number of profiles to fail the fitting routine regardless of the boundary conditions.
To identify profiles which were affected by interference we used the recorded temperature of the 21-cm receiver. Abnormally high or low values (by %) were assumed to indicate bad observations. We rejected all observations for which fluctuations in exceeded 10% of the running mean. In addition we rejected all profiles with an rms-noise exceeding the average noise by a factor of 4, as well as those affected by interference spikes with amplitudes exceeding the noise by a factor of 10. Profiles which passed these criteria were found to have well-defined baseline regions free of line emission. On average the baseline was defined by 417 channels, corresponding to 50% of the analyzed velocity range. Due to our restrictions 28% of the observations were excluded from the analysis.
As mentioned in Sect. 2, systematic spurious intensities due to reflections from the ground (Hartmann et al. 1996) limit the accuracy of the LDS. Such lines with dispersions and intensities up to 50 mK affected the analysis of the profiles averaged over predominantly between latitudes (Westphalen, 1997). Based on 2700 spectra showing the typical signature of reflections from the ground, a proper correction for such reflections was calculated and applied to the entire LDS data set. The first attempts to correct profiles for ground reflections by Hartmann et al. (1996) had failed because a significant fraction of the test profiles were affected by interference. This problem could be overcome as described above only after analyzing a large number of affected profiles for this purpose.
We further checked our data for additional systematic errors in the stray radiation correction. Any increase of a side- or back-lobe level did not affect the profile wings at the most extreme velocities, but only introduced significant errors in the velocity range of the main line components. We therefore compared the dispersions of stray-radiation profiles and HVD components. Stray radiation profiles were decomposed into Gaussian components using the same criteria as for the analysis of corrected profiles. Forcing the broadest component to fit the extreme profile wings we found that the dispersion of the stray radiation components is significantly smaller than the HVD components identified in the corrected profiles.
Fig. 3 shows the average velocity dispersions of the broadest stray radiation components which were removed from the spectra, along with the dispersions observed in the averaged profiles. Not only are the stray-radiation velocity dispersions systematically smaller than those of the HVD component, they are uncorrelated with galactic latitude. The latitude dependence of the HVD component is the strongest evidence that this emission is genuine. The systematic uncertainties due to residual stray radiation are estimated to be 5-15 mK, well below the 50 mK amplitude observed for the HVD component.
The data obtained after such a restrictive reduction as described above are incomplete, but in an unbiased manner, and can be used to estimate the uncertainties which may have affected the profiles plotted in Fig. 1. No preference can be given to neither the original dataset nor the revised version (both corrected for reflections from ground). Thus positive or negative deviations have the same probability. The uncertainties estimated this way are plotted in Fig. 2 for . No comparison can be made for latitudes since in this range the reliability of the baselines may be affected by extended wings of the conventional H I emission. We rejected profiles for which less than 300 channels were used to fit the baseline. To allow a comparison with the model calculations given in Sect. 4, we have also plotted the model data in Fig. 2. The amplitudes of the model exceed 50 mK while the baseline-uncertainties of the analyzed averaged profiles are at most 15 mK. We conclude that our data reduction leads essentially to the same results as the reduction by Hartmann (1994). The correction for reflections from ground is essential for an analysis of weak lines. Residual errors in the stray radiation corrections are probably in the range 10 to 20 mK, hence too small to affect the HVD components found in all latitudes.
© European Southern Observatory (ESO) 1998
Online publication: March 30, 1998