All models I fit have at least one component with "anti-Stokes" shift (excitation wavelengths higher than emission wavelengths). I am using a normalised dataset, with non-negative constraints.
At what EX/EM wavelength pair is your component? It might be an artefact of 1st order rayleigh scatter smoothing. Maybe you could try to increase the margins of smoothing or remove the according parts of the scatter.
could you show us a picture of your components? The answer to this question depends on the issue you're having. As Astrid said, sometimes scatter can be an issue producing components that are not viable. Sometimes, rectangles of only missing numbers in each EEM cause completely random results since the model cannot fit against data. Lastly, if your components look nice & smooth, but emission occurs below excitation: Did emission and excitation get swapped during the data import by accident?
Thanks, Urban. We took a closer look and increased a bit the excised section of the first-order Rayleigh-Tyndall scatter, and also increased the excised area under that scatter band, and it is not giving us anti-stokes shifts anymore.
I am looking forward to working with version 0.5 of the drEEM toolbox, but in the meantime, is there an update version of the tutorial for version 0.4?
Great to hear that you were able to troubleshoot the issue. The new functions scanview, eemreview, and spectralvariance will hopefully provide an easy way to spot this in the future.
We are really close to releasing it, just ironing out some platform- and version-dependent issues. Expect it this week!