Appendix A: Reduction of the LWS data

The photometric calibration of the LWS is primarily based on observations and models of the planet Uranus. The S/N in these observations is rather modest, which will affect the relative spectral response function (RSRF) derived from these data. When calibrating the "science data'', the registered photocurrent is divided by the RSRF, propagating any uncertainty in the RSRF, which will ultimately lead to errors in the derived flux density, $F_{\lambda}$.

The derived photocurrent after standard processing, resulting in an SPD file and where identified instrumental peculiarities have been removed, is shown for the LW 5 LWS detector in the upper panel of Fig. A.1, together with the relative spectral response function scaled to the same mean value. From the figure, it is evident that many features seen in the photocurrent are due to the detector and grating response.

Figure A.1: Upper panel: the observed wavelength dependence of the photocurrent (thick line) for the detector LW 5of the LWS. The thin line refers to the grating response function (RSRF) of the detector, scaled to the same mean value as the photocurrent. Middle panel: the lower curves are the same as in the upper panel, after subtracting boxcar fits. The division of the photocurrent by the RSRF results in the upper curve, which is the observed flux density (for clarity a median fit to the continuum has been subtracted). Lower panel: the same as in the middle panel, but with a small wavelength shift (<$0.15~\mu$m) of the observed photocurrent, before division. Evidently, the "noise'' is much reduced: compare, e.g., the spurious "absorption'' feature near 172 $\mu$m in the middle and lower panels.

The middle panel of Fig. A.1 shows again the photocurrent and the RSRF, but now after the subtraction of a wide boxcar smoothing function (continuum subtraction). The upper curve shows the result of the division of the photocurrent by the response function. This allows us to gauge the effects on the spectra from the narrow features of the RSRF. In the long wavelength regime of the LWS, and in LW 5 in particular, there are spurious absorption features at positions, where the RSRF is steep. In regions, where the RSRF has steep gradients, can already very small wavelength errors create large features that could be mistaken for spectral lines. A wavelength mismatch between the photocurrent and the RSRF will also introduce an overall lowered S/N.

In the lower panel of Fig. A.1, a relative shift by less than a quarter of an resolution element (the original sampling rate was at four times the spectral resolution) of the photocurrent resulted in the considerable reduction of an apparent broad "absorption'' feature near 172 $\mu$m, an overall better S/N and therefore a better definition of the spectral lines.

All individual spectra for all ten detectors have been carefully monitored for obviously anomalous features which (most likely) were introduced by the RSRFs. This was done for the RSRFs of both OLP 8 and OLP 10 and any such spurious features were, of course, corrected for. This does not guarantee, however, that no such false spectral features do still exist in our data, as we were very restrictive in our application of any wavelength shifts.