More calibration data sets
This addresses the following email question:
I have a question concerning the CRONUS-Earth online tool. Updated Be-10 production rates were recently published by
and I would like to ask, if it is possible to calculate exposure ages with the CRONUS-Earth online tool using these Be-10 production rates. I could not find a feature to change production rates.
A couple of answers to this. First, there is in fact no feature to change production rates. This is an explicit behavioral-engineering scheme to force people to think about production rate calibration the correct way. As always, the point here is that we don’t actually measure the reference production rate at sea level and high latitude — we measure time-averaged production rates at sites in various locations that are not at sea level and high latitude, and use a scaling method that describes how production rates vary in space and time to convert the actual measurements to an estimate of the reference production rate. Obviously, one must then use the same scaling method to scale the reference production rate estimate to the production rate at a new site of unknown age. So the online exposure age calculators don’t allow you to simply enter a reference production rate value, because then you might potentially commit sloppy and incorrect thinking by using different scaling methods to estimate the reference production rate from the calibration data and then the production rate at the unknown-age site from the reference production rate. If all the information you have is a reference production rate estimate, there is no way to avoid committing this error, because you don’t necessarily know what scaling method was used to generate it from the calibration data that were actually measured.
Thus, not allowing users to input reference production rates of their choice is purely a behavioral-engineering strategy to force users to focus on the data that were actually measured — nuclide concentrations at calibration sites of known age — and not the reference production rate estimates that don’t represent actual measurements. This simply implements the principle that to a man with a hammer, everything looks like a nail. So if you don’t want nails driven into things, don’t give him a hammer.
Instead, we provide a method to enter actual, directly measured, calibration data. That takes place on this page. Submitting those data then generates estimates of the reference production rate that can be used to calculate exposure ages at unknown sites. This workflow ensures that scaling is done exactly the same way in both directions.
Because of how HTML forms are written, this workflow has a useful feature that no one knows about. If you enter a calibration data set and press “submit,” the system will return a new web page that can be used to submit data from unknown age sites and calculate their exposure ages using reference production rates that best fit the calibration data you entered. Because these reference production rates are hard-coded into the web page (they are hidden variables in the HTML code that defines the input form), you can save this page to your local computer. Then, you can open it later whenever you want (of course, you still need to have a network connection to the actual calculation server on hess.ess.washington.edu) and use it to compute exposure ages according to those reference production rate estimates, without having to redo the calibration procedure. This is a useful timesaver because the optimization procedure to find the best-fitting reference production rate estimates can be time-consuming with a lot of data.
So this gets closer to the answer to the other part of the question above. You can use the Heyman (2014) calibration not by entering the reference production rate estimates from that paper (in his Table 1), but by entering the calibration data that are used to derive those estimates. A link to the calibration data from that paper (that I’ve excerpted from the supplementary data in that paper by removing the outliers as described in the paper) is here. You can enter that into the calibration input page to derive reference production rates.
Of course, this generates two problems. One, there are a lot of data, so the calculation is quite slow. It takes about 5 minutes. Be patient. Second, this calculation yields reference production rate estimates that are slightly different from those in Table 1 of the Heyman paper. The reason for this is that the method of averaging is different. The online calibration code weights each sample according to its uncertainty; more precisely measured data have more weight in the averaging, and all samples are treated equally otherwise. In the Heyman paper, samples from each individual calibration site are used to generate a production rate value for that site; all site values are then averaged to obtain a summary value. The idea here is to treat all sites equally, no matter how many samples were collected at each site. Frankly, I don’t see any strong rationale for deciding which one of these strategies is better. However, the difference between the two estimates is smaller than their uncertainties, so the issue doesn’t appear to be particularly important.
This brings up the last subject. You don’t actually have to wait 5 minutes for the fitting calculation to finish because I’ve already done it, and posted the results on the alternate calibration data sets page. At the bottom of this page, there are links to input pages that use reference production rate estimates from the Heyman data set calculated with both averaging schemes.
In addition, I’ve also added input pages that use production rate estimates derived from calibration data in recent papers by Meredith Kelly and PH Blard. These are discussed at length in previous blog entries here and here, and I should have done this months ago.
Finally, a couple of comments about the Heyman paper. Basically, what Jakob has done is to simply observe that since the compilation of available calibration data that went into the initial calibration used in the online exposure age calculators in 2008, i) there have been a lot of new calibration data collected that showed that the 2008 data set overestimated actual production rates by something in the area of 10%; and ii) all these new calibration data, at least when normalized to each other with scaling schemes based on the work of Devendra Lal (the St and Lm schemes in the online calculators), all pretty much agree with each other. Thus, Jakob has simply piled up all these new data sets and obtained a best-fitting reference production rate to all the data together. This is perfectly sensible and, in my view, is also a pretty good way to estimate the global scaling uncertainty (he concludes it is 6%, which I agree with). I would have excluded some of the data sets (for example, the Be-10 measurement standardization of Gosse, 1995, is not actually known because no one can find the relevant lab books from the UPenn accelerator; and the Larsen, 1996 data have some problems including both analytical reproducibility and the fact that really the radiocarbon ages are only younger limits on the age of the sites). However, his compilation shows fairly clearly that most of the recent calibration data, when properly standardized, agree with each other pretty well.