Does anyone can share information about the minimum data/samples for using PARAFAC (parallel factor analysis) to characterize organic matter in river or lake samples?
For example, I only have 10 samples fluorescence spectroscopy data. is 10 samples suitable to analyze the component of organic fractions by using PARAFAC?
Dear Euis Nurul,
In factor analysis, the problem is always to have more samples than factors needed to express all variations. I'm not a PARAFAC expert but I think you can try. I'm sure you can find this kind of information from Rasmus Bro (http://www.models.kvl.dk/users/rasmus/)
Regards.
Ludovic.
Hi,
There is not a minimum number of samples to run PARAFAC. However, according to Stedmon and Bro, 2008 (Characterizing dissolved organic matter fluorescence with parallel factor analysis: a tutorial, Limnol. Oceanogr. Methods), a number between 20 and 100 would be preferable to resolve mixtures such as DOM. From my own experience, over 100 samples would be preferable for consistent results. It should also be considered that if split-half validation is going to be conducted, a small number of samples in each sub-data set could be problematic. I have usually run PARAFAC analysis using large data sets (> 1000 samples) and results have been very robust.
Hope this helps,
Nancy
I agree with the above.
There is no benefit at all in having few samples, but lots of benefits of having many. You will likely benefit by increasing sample sizes even up to the thousands.
Related: you can actually extract more components than samples in PARAFAC. Even with as little as two samples, you can extract many more components if the data support it. So PARAFAC is a funny kind of animal!
Best regards
Rsmus
I vaguely remember having at least "20 samples" for PARAFAC modeling while having a friendly informal discussion with Dr. Colin Stedmon (who is a pioneer in PARAFAC application for DOM) at University of Birmingham during a PARAFAC workshop. It would be more beneficial to have samples from diverse ecosystems than lots of samples from just one area.
Truly, there is no standard minimum set of data to conduct parafac and a larger data set probably gives stability to the model and more accurate analytic informations. Nevertheless, avoiding the combination of data will spare you from the confusion of having unecessary coumpound formation. Thanks
For ensample, in aim of getting more data for parafac, it won't make sense to combine water samples from lakes and rivers. The formed coumpounds could not be real and related to specific area. So unecessary compounds will be formed as result of mixing.
Can we generate multiple EEM matrices from the single sample? It means taking multiple samples from the single sample. Let's say we have 10 samples. If we can obtain 5 different EEM matrices from each sample, we can get 50 EEM matrices. It is possible?
- Badges
- Science topic
- Similar topics
- Correction