Status: offline
Registered: 09/30/2005
Posts: 4040
|
[Moderator's Note: Posted from submitted story on behalf of user.]
I am analyzing high-resolution near-infrared spectra retrived with our original spectrograph WINERED installed on a 1.3-m telescope at Kyoto Sangyo University.
In order to correct telluric absorption lines lying on a target star spectrum, I am using the IRAF task "telluric". The "xcorr" and "tweakrms" options are set YES, so that the RMS is to be minimized by varying the shift and scale values automatically.
However, the result of the automatic fitting, which can be seen from the link below, was not satisfactory.
http://flic.kr/p/KBvebn
The sample region is shown as the horizontal bar. As can be clearly seen, the automatic fitting left a lot of spiky residuals. IRAF returned an RMS value of 0.0136.
Thereafter, I manually changed both the shift and scale values by visual inspection. Through a trial and error process, I achieved a satisfactory fitting, which can be seen from the link below. At this time, IRAF returned an RMS value of 0.018, which is larger than the automatic result.
http://flic.kr/p/KyzMRJ
I felt strange about those values that IRAF returned. Therefore, I calculated by myself the RMS values for the above two sets. Specifically, I calculated (sigma_i^n{(y_i - )^2 / n})^0.5 where x_i is in the sample region. The results were ~0.025 for the automatic fitting, while ~0.012 for the manual fitting.
Note that I scaled the spectra created from the "telluric" task so that the continuum level is unity. Thus it is not strange that the values between IRAF and my calculation agrees perfectly. However, it is quite strange that RMS(auto) \$this->_split2($m[0]) RMS(manual) in my calculation, while RMS(auto)
|
Status: offline
Registered: 11/11/2005
Posts: 728
|
Hello,
I'm sorry for the late reply. I would need to reproduce the result with your spectra and parameters to give you a precise answer. However, my guess is that the task is finding a local minimum in the process but not the best minimum. What the task does is a kind of binary search, much like your trial-and-error, and tries to follow the direction that minimizes the RMS on each iteration. It is possible that it can get a wrong answer. The reason the task was written with the interactive options was exactly so someone could use their better intelligence than the computer to get the absolute best result.
Frank Valdes
|