How to eliminate Pb spikes

chenhongjun

Dear Bence and Joe，
😃 When I do U-Pb dating of the garnet mineral, I will find that there are many abnormally high spikes in the signals I collected (especially Pb), which may be the influence of small internal inclusions, often the age value of these points Or the error value will be very large, how can I filter out these spikes?

Bence

Hi @chenhongjun,

If there are just a few spikes, they will be likely removed from the selection mean/median by iolite's outlier rejection. If you haven't changed this setting in iolite's Preferences -> Stats, it will be set to reject anything that is outside 3 SE for the selection.

If you suspect that the spikes might not be removed by outlier rejection, you can try selecting the intervals around the spikes (either manually, or if too hard to do by hand, you can use automatic selections from channels, using one of your final ratios in your criteria), and then linking the selections. The resulting linked selection will cover all the data not affected by spikes.

Bence

chenhongjun

Bence, 😀 I would like to ask, the specific difference between 2SE and 3SE? Is the range of 3SE 1.5 times that of 2SE?I haven't changed this setting in iolite's Preferences -> Stats, so it will be set to reject anything that is outside 3 SE for the selection.By the way, does the iolite software have the function of automatically filtering spikes? There are many spikes in some points, but the signal range is still very narrow，as the picture shows：
https://imgur.com/a/GWcLH2R，But the data is obviously beyond the range of 2SD, I don’t know if it is caused by these spikes.，But the signal at a certain point looks very stable, but the point is still very discrete, which is very strange, as shown in the figure ：https://imgur.com/a/znDQvsi.How to deal with these spikes really makes me very distressed. For example, when dealing with age, the 207Pb/206Pb signal at certain points fluctuates. I don’t know whether such data is available (because these points have a high U content, which is close to the lower intersection. It is very important for me to date), how to eliminate the influence of these up and down spikes, because the error values of these points seem to be large, as shown in the figure ：
https://imgur.com/a/CGEEWPi https://imgur.com/a/1FBSB90 https://imgur.com/a/iwTpqsV: 😀

Bence

Hi @chenhongjun ,

The difference between 2SE and 3SE is that 2SE is two times the standard error of the sample you're looking at, and 3SE is three times that standard error.

In the first image you show, I wouldn't necessarily say that those spikes are beyond 2SD/2SE: I would say that it looks like there are enough of them that you have a bi-modal distribution, and that your sample is a complex mix of the two. Now you could filter these results out, but it would create many short selections between the U spikes that you could then link together to form a "linked selection". You could do this via the Automatic Selections -> 'from channel data' tool to select for a range of U238/Pb206 ratios. However, I think that this would start to approach data manipulation. Others may have a different interpretation. Please feel free to post here if you have anything to add!

Regarding your second image: it does appear that there are different zones within your ablation. Which of these to select/avoid is somewhat subjective unless you define what and why you're rejecting data before adjusting anything. For example, you could select based on concordance, but it looks from your other images that you may not have significant amounts of common Pb present, so concordance may not be appropriate here.

Which brings me to your final images. From what I can see (i.e. the images you've posted) it appears to be a very well defined spread between a common Pb component and a radiogenic component. The intercept of this group is a possible estimate of the 'true' age of your samples. But you cannot (easily) select just for the most radiogenic parts of your samples if they are a complex mixture. It would actually make the nice linear relationship you see when looking at all samples less well define and thus your intercept would be less well determined. I would suggest that you treat all these selections as a whole defining what is happening in your samples, rather than trying to optimise each selection for the most radiogenic components.

Again, others might have more/better advice, so please feel free to add to the discussion here.

Best regards,
Bence