The double-peaked feature in SW1 is an example of a recurring spurious feature in L01 spectra. Any recurring feature is generally expected to be due to an error in the RSRF derivation, and this one is no exception. Although considerable time and effort has been spent on fine tuning the RSRF, some unexplained features remain even in OLP 10.1. As this feature is the most obvious one remaining, a detailed investigation into it was thought to be worthwhile. This was done in several stages.
First, the 15 observations used in the RSRF derivation (see LWS Handbook S5.2) were checked against the Uranus model. This was done at LSAN level, comparing the measured and predicted flux at each wavelength. If the calibration is perfect, the flux ratio will always be equal to 1. It was found that a deviation exists that is consistent with the double-peaked shape seen in many spectra, which implies that the feature is caused by an error in the RSRF itself.
The second stage was to look back at the RSRF derivation procedure. This uses a complex scaling process, by which each scan (out of a total of 104 used) is scaled to a reference scan (taken as the first scan in TDT 32103705). By examining the pattern of scaling factors in each different observation, four observations were found to contain potential inconsistencies. Subsequently, over-plotting suspect scans with normal-looking scans showed a variety of unusual features, some of which were consistent throughout observations. If an individual observation contains a consistent problem, it is likely that the averaging process used in deriving the RSRF will not exclude such data, and hence the overall average becomes skewed.
To test whether the four TDTs identified were responsible for the RSRF error, a new RSRF was derived using the remaining 11 TDTs. When this was compared to the original RSRF it showed a deviation similar in shape to the double-peaked feature in the LSAN data. This implies that the original set of 15 TDTs is contaminated with a least some bad data in SW1, and that this caused the error in the RSRF. As it is difficult to find all the bad scans and features responsible in the LSPD files, it was decided that it would be better to derive a correction from the LSAN data, using the remaining 11 TDTs only.
In the final stage, a process was designed to combine the good data from the 11 LSAN files, and average it together across the nominal wavelength range. This was done so as to give a resolution approximately equal to the nominal spectral sampling, which was used to correct the Uranus data and most other L01 observations. The data were corrected for the distance of ISO from Uranus for each observation, and an average measured brightness was obtained (following a sigma-clip and mean) at each wavelength. The Uranus model was then used to generate a set of theoretical brightness values. The measured values were then divided by the theoretical values to get a correction factor at each wavelength.
To apply the correction, a routine was constructed to interpolate the correction profile with the wavelengths from an observation. The factors obtained are then applied to the flux values, to give a new set of fluxes. Testing of this procedure on a variety of different sources has shown that good improvements are made to the spectral shape in all cases. Little change is seen at low flux levels, as the noise level is generally too high to make out the double-peaked shape.