smoothing XAS data
Hello, In what case smoothing of the data is OK? My data have a lot of noise and smoothing really helps (visually). I compared fit done in Artemis for smoothed and non-smoothed data sets and the result is very similar, however when I searched about it online, it looks like it is not recommended. What is a good way to justify it? Thank you! -Daria
Hi Daria: I smooth (or boxcar average) the data which it is oversampled, such as in a continuous scan where the data in the exafs region is spaced closer than the usual delta k of 0.05. I generally don't like to smooth XANES data since smoothing does tend to attenuate sharp features such as the ones which exist near the absorption edge. Carlo On Wed, 22 Jan 2020, Daria Boglaienko wrote:
Hello,
In what case smoothing of the data is OK? My data have a lot of noise and smoothing really helps (visually). I compared fit done in Artemis for smoothed and non-smoothed data sets and the result is very similar, however when I searched about it online, it looks like it is not recommended. What is a good way to justify it?
Thank you!
-Daria
-- Carlo U. Segre -- Duchossois Leadership Professor of Physics Directory, Center for Synchrotron Radiation Research and Instrumentation Illinois Institute of Technology Voice: 312.567.3498 Fax: 312.567.3494 segre@iit.edu http://phys.iit.edu/~segre segre@debian.org
Not only can smoothing XANES data attenuate sharp features, it can also shift them, since many XANES features, particularly in the “pre-edge” region, are significantly asymmetric about their peak. That can make analysis quite confusing, since, e.g., the size of the boxcar ends up affecting the position of the peak! Best, Scott Calvin Lehman College of the City University of New York
On Jan 22, 2020, at 8:44 PM, Carlo Segre
wrote: I generally don't like to smooth XANES data since smoothing does tend to attenuate sharp features such as the ones which exist near the absorption edge.
At this juncture, I'd like to bring up a "rule" I've made for myself regarding data quality and smoothing: I'd rather walk away from a beam run with one good spectrum than a hundred poor ones. Meaning, for me anyway, if the data aren't smooth enough on their own, I'd rather spend more precious time counting than move on. I'd count for three days on a sample if I had to (I've never had to). While it has been proven that, contrary to the popular saying, one _can_ polish a turd, I'd personally rather not try. I realize this approach was more difficult when I was a grad student and everyone in the research group was clamoring for data, but even in that situation, good data > not-good data. Apologies for the tangent, but I hope someone out there might find it useful. Cheers, Mike
On Jan 22, 2020, at 5:44 PM, Carlo Segre
wrote: Hi Daria:
I smooth (or boxcar average) the data which it is oversampled, such as in a continuous scan where the data in the exafs region is spaced closer than the usual delta k of 0.05.
I generally don't like to smooth XANES data since smoothing does tend to attenuate sharp features such as the ones which exist near the absorption edge.
Carlo
On Wed, 22 Jan 2020, Daria Boglaienko wrote:
Hello,
In what case smoothing of the data is OK? My data have a lot of noise and smoothing really helps (visually). I compared fit done in Artemis for smoothed and non-smoothed data sets and the result is very similar, however when I searched about it online, it looks like it is not recommended. What is a good way to justify it?
Thank you!
-Daria
-- Carlo U. Segre -- Duchossois Leadership Professor of Physics Directory, Center for Synchrotron Radiation Research and Instrumentation Illinois Institute of Technology Voice: 312.567.3498 Fax: 312.567.3494 segre@iit.edu http://phys.iit.edu/~segre segre@debian.org _______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit Unsubscribe: http://millenia.cars.aps.anl.gov/mailman/options/ifeffit
I entirely agree with Mike and Scott.
I never smooth data because it is a slippery slope to start on! How much should you smooth? Should you only smooth data if it looks "bad"? What does "bad data" look like anyway?
If the only point of smoothing is to improve the visual look of data, then you are doing it for the wrong reason - it is better to present the true data and let the reader decide if your interpretation is robust. In my view, smoothing data is somewhat scientifically dishonest unless you have a strong justification that extends beyond trying to make the data look "better"...
As someone who works with environmental samples at low concentrations, I'm very used to seeing noisy data. You just have to get used to long count times and lots of replicate scans!
Cheers,
Will
Dr William W. Bennett
Senior Lecturer
Environmental Futures Research Institute
School of Environment and Science
Griffith University
Gold Coast, Queensland, Australia
M +61 401 186 488
w.bennett@griffith.edu.au
Assistant Professor
Nordcee, Department of Biology
University of Southern Denmark
Odense, Denmark
M +45 8193 8111
wbennett@biology.sdu.dk
________________________________
From: Ifeffit
On Jan 22, 2020, at 5:44 PM, Carlo Segre
wrote: Hi Daria:
I smooth (or boxcar average) the data which it is oversampled, such as in a continuous scan where the data in the exafs region is spaced closer than the usual delta k of 0.05.
I generally don't like to smooth XANES data since smoothing does tend to attenuate sharp features such as the ones which exist near the absorption edge.
Carlo
On Wed, 22 Jan 2020, Daria Boglaienko wrote:
Hello,
In what case smoothing of the data is OK? My data have a lot of noise and smoothing really helps (visually). I compared fit done in Artemis for smoothed and non-smoothed data sets and the result is very similar, however when I searched about it online, it looks like it is not recommended. What is a good way to justify it?
Thank you!
-Daria
-- Carlo U. Segre -- Duchossois Leadership Professor of Physics Directory, Center for Synchrotron Radiation Research and Instrumentation Illinois Institute of Technology Voice: 312.567.3498 Fax: 312.567.3494 segre@iit.edu http://phys.iit.edu/~segrehttp://phys.iit.edu/~segre segre@debian.org _______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffithttp://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit Unsubscribe: http://millenia.cars.aps.anl.gov/mailman/options/ifeffithttp://millenia.cars.aps.anl.gov/mailman/options/ifeffit
_______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffithttp://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit Unsubscribe: http://millenia.cars.aps.anl.gov/mailman/options/ifeffithttp://millenia.cars.aps.anl.gov/mailman/options/ifeffit
I agree with all of the comments. If I take data using step scans, then I never smooth. However, in continuous scan mode where the data is oversampled, a boxcar average in the EXAFS region is OK. The alternative is that the fitting software interpolates using an algorithm which effectively averages over the delta k = 0.05 bin. By using a boxcar average the statistics of the measurement are all included. This is not really smoothing though. Carlo On Thu, 23 Jan 2020, Will Bennett wrote:
I entirely agree with Mike and Scott.
I never smooth data because it is a slippery slope to start on! How much should you smooth? Should you only smooth data if it looks "bad"? What does "bad data" look like anyway?
If the only point of smoothing is to improve the visual look of data, then you are doing it for the wrong reason - it is better to present the true data and let the reader decide if your interpretation is robust. In my view, smoothing data is somewhat scientifically dishonest unless you have a strong justification that extends beyond trying to make the data look "better"...
As someone who works with environmental samples at low concentrations, I'm very used to seeing noisy data. You just have to get used to long count times and lots of replicate scans!
Cheers,
Will
Dr William W. Bennett
Senior Lecturer Environmental Futures Research Institute School of Environment and Science Griffith University Gold Coast, Queensland, Australia M +61 401 186 488 w.bennett@griffith.edu.au
Assistant Professor Nordcee, Department of Biology University of Southern Denmark Odense, Denmark M +45 8193 8111 wbennett@biology.sdu.dk
________________________________ From: Ifeffit
on behalf of Mike Massey Sent: Thursday, 23 January 2020 2:50 PM To: Carlo Segre ; XAFS Analysis using Ifeffit Subject: Re: [Ifeffit] smoothing XAS data At this juncture, I'd like to bring up a "rule" I've made for myself regarding data quality and smoothing: I'd rather walk away from a beam run with one good spectrum than a hundred poor ones.
Meaning, for me anyway, if the data aren't smooth enough on their own, I'd rather spend more precious time counting than move on. I'd count for three days on a sample if I had to (I've never had to).
While it has been proven that, contrary to the popular saying, one _can_ polish a turd, I'd personally rather not try.
I realize this approach was more difficult when I was a grad student and everyone in the research group was clamoring for data, but even in that situation, good data > not-good data.
Apologies for the tangent, but I hope someone out there might find it useful.
Cheers,
Mike
On Jan 22, 2020, at 5:44 PM, Carlo Segre
wrote: Hi Daria:
I smooth (or boxcar average) the data which it is oversampled, such as in a continuous scan where the data in the exafs region is spaced closer than the usual delta k of 0.05.
I generally don't like to smooth XANES data since smoothing does tend to attenuate sharp features such as the ones which exist near the absorption edge.
Carlo
On Wed, 22 Jan 2020, Daria Boglaienko wrote:
Hello,
In what case smoothing of the data is OK? My data have a lot of noise and smoothing really helps (visually). I compared fit done in Artemis for smoothed and non-smoothed data sets and the result is very similar, however when I searched about it online, it looks like it is not recommended. What is a good way to justify it?
Thank you!
-Daria
-- Carlo U. Segre -- Duchossois Leadership Professor of Physics Directory, Center for Synchrotron Radiation Research and Instrumentation Illinois Institute of Technology Voice: 312.567.3498 Fax: 312.567.3494 segre@iit.edu http://phys.iit.edu/~segrehttp://phys.iit.edu/~segre segre@debian.org _______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffithttp://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit Unsubscribe: http://millenia.cars.aps.anl.gov/mailman/options/ifeffithttp://millenia.cars.aps.anl.gov/mailman/options/ifeffit
_______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffithttp://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit Unsubscribe: http://millenia.cars.aps.anl.gov/mailman/options/ifeffithttp://millenia.cars.aps.anl.gov/mailman/options/ifeffit
-- Carlo U. Segre -- Duchossois Leadership Professor of Physics Directory, Center for Synchrotron Radiation Research and Instrumentation Illinois Institute of Technology Voice: 312.567.3498 Fax: 312.567.3494 segre@iit.edu http://phys.iit.edu/~segre segre@debian.org
Dear Daria, as you stated the fitting result is very similar for the smoothed and unsmoothed data. So I think the answer to some commenters' question "Why do you smooth the data?" is that it just looks better. First of all, if you smooth your data you must, obviously, say so and explain how you did the smoothing when publishing/presenting these results. Secondly, the quality of the raw EXAFS data is an important measure of how good the measurement is. Smoothing it will remove this information. This is a strong argument to NOT use smoothing. When you do Fourier-Transformation from k-space to R-space (and possibly back from R-space to q-space) this automatically smooths the data. I am aware that this is a filtering of frequencies rather than "real" smoothing, but it will give the effect you want: The data will look less noisy. In conclusion, I would suggest never smoothing, instead showing the back-transform from R-space to q-space if you want to have nice looking curves but your data are not that good but ALWAYS supplying raw data in publications. With kind regards, Felix -- Dr. Felix E. Feiten Fritz-Haber-Institut der Max-Planck-Gesellschaft Department of Interface Science Faradayweg 4-6 14195 Berlin phone: +49 30 8413 4142
participants (6)
-
Carlo Segre
-
Daria Boglaienko
-
Felix E. Feiten
-
Mike Massey
-
Scott Calvin
-
Will Bennett