search

rosalieb / serac

an R package for ShortlivEd RAdionuclide Chronology of recent sediment cores (Bruel & Sabatier, 2020, Journal of Environmental Radioactivity)
MIT License
10 stars 2 forks source link

Negative Pb-210_xs #2

Open Danapit opened 1 year ago

Danapit commented 1 year ago

Hello Rosalie, First of all, thank you very much for this great tool! I wonder what you'd recommend to do when there are negative values of excess 210Pb in the profile. Because excess 210Pb is a calculated value from two measurements, which both have some uncertainty connected to them, it's only natural when the values close to the base of the core, where no or little excess is present, tend to be distributed on both sides of 0. I think ignoring these values introduces a bias. Could a model with a fixed offset be used to avoid negatives, when the linearized fit (using log values) is used? Or a non-linear fit on original values, which could directly deal with negative? Cheers! Dana

rosalieb commented 1 year ago

Hi Dana, My first thought is that negative Pbex in the profile should be ignored, but maybe I have not given this enough thoughts. Could you please share an illustration of the issue and an illustration of the offset solution? I think I understand what you mean, but I prefer to be sure. Hand drawn diagram is fine if you cannot share the original data!

Danapit commented 1 year ago

Thank you, I'll come up with an example, maybe the best will be a synthetic profile to illustrate the issue. It might take me a couple of days, sorry!

rosalieb commented 1 year ago

Thanks a lot, and no rush!

Danapit commented 1 year ago

Hey! I took a while and still kind of draft status... I was thinking about this problem and I don't know if this really is a problem, and whether negative values could actually contribute to a bias (towards "flatter" profiles -> high calculated sedimentation rates, I would expect). I don't have such a solid background in maths, so I tried to sum it up in this repo and played around with an artificially generated profiles: https://github.com/Danapit/210Pb_dating_CFCS_SERAC_test I didn't get to the last part which should be some kind of statistical testing, which I have to think about first - long time ago I did this training. But from the results on the first glance it appears to me it doesn't make too much difference if there are negative values present in the profile. Maybe you can have a look and tell me what you think. Thanks!