The fluorescence of DOM is a function of both the molecules that are in it and interactions between molecules, especially for fluorescence in the "humic-like region" (http://www.ncbi.nlm.nih.gov/pubmed/24509887).  Thus, the fluorescence of DOM is impacted both by its composition and its structure.  Unfortunately, since the number of possible configurations and hence subtly different fluorescence signatures is effectively infinite, a(n accurate) universal PARAFAC model cannot be created.  DOM is the epitome of a stubbornly complex and dynamic system (and that's why it's so much fun to study)! :)

One possible way would be for some person or group to compile a master set of EEM, that they pick the number of components and trim and reduce all the data to the same range/intervals. With this other people could calculate component concentrations based on the PARAFAC of this master data set, and report the values in their own papers. The science will continue to improve, so it would make sense to incorporate new data into the master list and eventually remove old data that isn't sufficiently detailed. I think only doing this every few years would be sufficient to stay up-to-date with current science, while allowing enough time for studies done near the same time to be compared.

thanks, and i think this site has done some work you describe. the site address is: http://www.openfluor.org .Hope this can help!

Anyway, PARAFAC is just one method to decompose the data and we cannot expect it too much!

Interesting. That's a nice looking website.

It is slightly different than my idea but this implementation might be more useful for those actively measuring EEM Fluorescence, which I am not and I recognize how much work Colin Stedmon and the other creators have done into this field.

They allow you to compare your model (of a dataset) against multiple published models (of many datasets) which lets you determine if any of your model components are similar to past published model components. This would allow you to look back at model components in literature to learn what that component has typically been attributed to (even across many different models). Not having to reconstruct or replicate all of this from available literature data and by directly contacting authors, and basically automatic on the website is very useful.

My idea probably addresses a different scenario, and one I'm not sure how frequently it occurs. That would be if you had too few EEM to do PARAFAC and decided to assume an existing model. That may be a time when a universal model would be somewhat helpful, but there is difficulty creating one for some of the reasons you noted.

I agree with you on the universal model, but i think that possibly not realistic. For the dissolved organic matter is complex and the environment is the also very complex. Additionally, the measuring condition may differ from lab to lab. That all cause the versatility and variations of peak positions.

But what is also confusing me is that if there is too few samples, how should we deal with it? Recently, one postdoc asked me about this question. In her paper, there are ten samples taken from wastewater treatment processes. She submitted her paper, but the reviewer asked her to do PARAFAC, I think this is difficult for her just based on so few samples, for PARAFAC based on statistics and large number of samples in need to get it validated. And maybe she can put her data with someone's databases, but i think her data would be just the minority, and on that condition, maybe the components identified would not be suitable.

But sometimes the reviewers need this to be executed, so this maybe a question.

Recently, the feasibility and limitation of global PARAFAC model have been discussed by Yu et al. in this paper "Impact of dataset diversity on accuracy and sensitivity of parallel factor analysis model of dissolved organic matter fluorescence excitation-emission matrix" published on Scientific Reports. http://www.nature.com/srep/2015/150511/srep10207/full/srep10207.html

And the core conclusion is that to create one universal model for all environments is not suitable, but to create one relatively universal model for a narrow/similar environments is feasible and realistic.

The fluorescence of DOM is a function of both the molecules that are in it and interactions between molecules, especially for fluorescence in the "humic-like region" (http://www.ncbi.nlm.nih.gov/pubmed/24509887).  Thus, the fluorescence of DOM is impacted both by its composition and its structure.  Unfortunately, since the number of possible configurations and hence subtly different fluorescence signatures is effectively infinite, a(n accurate) universal PARAFAC model cannot be created.  DOM is the epitome of a stubbornly complex and dynamic system (and that's why it's so much fun to study)! :)

Chad's point is valid, and it is probably reflected in the quote from Yu et al.'s abstract in Nature (http://www.nature.com/srep/2015/150511/srep10207/full/srep10207.html): "building a universal PARAFAC model that can be widely available for fitting new EEMs would be quite difficult, but fitting EEMs to existing PARAFAC model that belong to a similar environment would be more realistic".  An important point in the "criticism" of PARAFAC in our paper (http://www.ncbi.nlm.nih.gov/pubmed/24509887) is the danger of assuming that components in PARAFAC of DOM represent discreet types of fluorophores, because they could easily be due to electronic interactions that could change with solution environment.  So, each end-member source of DOM (mainly algal versus terrestrial) could have unique fluorophores AND unique interactions between those fluorophores, so a universal component set would not work.  To the extent that the interactions between fluorophores are not highly variable WITHIN a certain source/class of DOM, then a model set of components for "similar environment(s)" (quoting Yu et al.) would probably be useful for assessing things like source fluctuations and maybe metal complexation and redox state.  But it's still a dicey proposition to assume PARAFAC components represent unique structural features.

Agree with Chad and Charles. When I firstly used PARAFAC to model fluorescence data six years ago, I was confused by the comparison of models with others'. That's why I raised this question. During the past several years, I have measured thousands of EEMs and conducted PARAFAC modelling many many times. The more I deal with the samples and EEMs, the more I feel and recognize the complexity of DOM and fluorescence. For my study, the same fluorophore identified by PARAFAC, even in one model, responds to microbial- and photo degradation in different samples in the same condition. So like other statistical tools, PARAFAC helps you deal with the data, but can never gives you the truth. Also the same, the EEM characteristics of natural DOM could just tell you where the fluorophore most probably derive from and how the reactivity is probably for the fluorophore, but also not the truth.