iraf.net

hi there,this should be obvious, but i confess i'm confused. i extracted a longslit spectrum using apall, and did variance weighting, extracting the extras spectra (raw, background and sigma).i need to obtain a kind of "S/N per pixel spectrum as a function of wavelength". i need it to analyze the program spectrum in a code that needs to know either the error or the S/N per pixel of the spectrum, in order to know how to weight each pixel of the program spectrum.is there a way to combine the extras spectra extracted by apall so that i have this "S/N per pixel spectrum"?thx a lot
paulabtw, when looking at the extras i noticed that the blue part of my background spectrum has negative counts. isn't weird? i indeed see using imexamine that there are pixels with values down to -3 (does this comes from a problem in bias correction?), but the background spectrum extracted by apall shows values down to -35.

Hi Paula,You could extract the error band, say with SCOPY and format=onedspec, and then use image arithmetic to convert it to a SNR.Yes it is possible that after basic CCD processing some pixels could be negative. Other than that I would need to understand what has been done to the data. Some cautions for you. The optimal weighting can sometimes produce artifacts if there are strong cosmic rays or low signal. Also the error computation does something approximate with negative pixels since the usual readnoise/gain noise model for the detected counts fails.I hope this is useful.Yours,
Frank Valdes

Hi,I need the same thing, "S/N per pixel spectrum as a function of wavelength" and i don't know any other task from splot to do that. I opened a discussion under "S/N for spectra" topic. Mike told me Frank is the best person to answer my question. My problem is that, i can find S/N for every single point on any spectrum or in batch mode as Mike suggested, for many points by bplot. Is there any other way to do that? Should I get S/N values from continuum points on the normalized spectrum by splot and then average them? By the way what exactly is an error band?Thanks,
Ozgur Basturk
Ankara University, Turkey
Astronomy and Space Science Department

Hello,In order to compute the SNR for a pixel in a 1D spectrum you need an estimate of the uncertainty in the measured flux. The can be hard, particularly if the spectrum has been dispersion and flux calibrated. There are two typical ways to estimate this.One is for data taken on a 2D detector with known noise characteristics. For example CCDs characterized by a readout noise and a gain. This is a challenge because most 2D calibration operations (dark subtraction, flat fielding, etc) modify the noise characteristics and the uncertainties are not propagated by the software (such as CCDPROC). So what is done in this case is estimate an uncertainty using a noise model during the extraction of a 1D spectrum from the 2D format. This is what APEXTRACT does in the "sigma band". This is a spectrum recorded with the extracted spectrum based on assuming a simple readout noise and gain with Poisson statistics is correct for the each 2D pixel and then propagating this during the 2D extraction. In summary, the uncertainty is estimated from the 2D statistics and all pixels that contribute to the 1D spectrum value.The second method is to estimate the uncertainty from the statistics of the 1D pixels around the region of each pixel. The problem here is that this only makes sense with continuum regions since lines should not contribute to the standard deviation of pixels. In this case, which is conceptually easier and more independent of modeling and previous calibration operations, usually the best you can do is a typical noise value for all pixels and the SNR is then the ratio of the measured spectrum divided by this noise value at all wavelengths.The latter is what is done with SPLOT and the 'm' key. The former depends on using APEXTRACT as briefly mentioned in previous postings.This discussion only gives you ideas for how to possibly determine the SNR. To make a SNR spectrum there are a number of ways, though these require several steps and possible some scripting to automate. The tasks that might prove useful areCONTINUUM
SARITH or IMEXPR
MIMSTATThe goal here is to determine either a single value or spectrum for the standard deviation with lines and any continuum slopes removed. Once your have that you use SARITH to compute the SNR spectrum.I realize this is a recipe but hopefully you can devise an image processing recipe or script appropriate for your data and needs.Yours,
Frank ValdesAn error band is an associated spectrum produced by the APEXTRACT package tasks when extracting a 1D spectrum from a 2D format. It is obtained when the parameter "extras" is yes and variance weighting or cleaning is used. The spectrum is stored in a "multispec" format which is a 3D image where the lines in the first "band" [*,*,1] are the 1D spectra and the associated data is in the higher dimensional bands. One of these would be the sigma spectra.

Frank,Thanks for your clear and detailed explanation, I'll try what you have suggested and get back to the topic if i have a problem.Thanks again,
Ozgur Basturk
Ankara University, Turkey
Astronomy and Space Science Department