'Wild' method sampling for calibration

[Christophe Duplais]

Hi!

We are writing a proposal to determine corn earworm native origin (migrant versus local) and we are still hesitating between the 'lab' method versus the 'wild' method for calibration.

Our limitation is that Deuterium oxide-18O costs $1,180 for 1 g! (we want both isotope in our analysis).

Would you have advice on collecting samples (number of site, number of replicate per site) for the 'wild' method. I imagine that 4 points judiciously chosen to have a variation can be enough? Or should we estimate a larger sample as if we wanted to build isoscapes?

Thank in advance for any help you can provide!

Kind regards,

Christophe

[Alexandre Courtiol]

Hi Christophe,

A correct answer would depend on many parameters that are difficult to foresee a priori.

If you are talking about the number of calibration samples to get from the wild, my wild ass guess is that 4 is way too little since the wild method would not be able to use 4 points to compute the required number of variances, covariances and estimates reliably (it the method would run at all).

You could play with the examples in the package by reducing the sample size of the calibration dataset to see the effect.

That would not correspond to your situation because the data would differ but that could give you a good idea.

I hope others will reply too.

Best,

Alex

[francois.rousset.new]

Hi Christophe,
I agree with Alex' reply.
The point of using a correct calibration fit is to estimate the source of uncertainties in assignment, which include the residual variance of the calibration fit. It is sometimes said that one requires 10 points to estimate a mean and 30 to estimate a variance... of course this is no more than a gross rule, but no more can be said without information specific to a given application.

Best,

F.

[Christophe Duplais]

Thank you so much for your very helpful answers! It makes more sense.

In analytical chemistry we need calibration curve to set up a calibration method, which includes several replicates at 4-5 concentrations of the standard. In this great paper they use IsoriX

and only 3 isotope 'concentrations' for the 'lab' method. That's why I was wondering is 4 collecting sites with great variation in d²H, d¹⁸O could have been enough for the 'wild' method.

I really think that having more field data points makes the calibration more accurate across latitudinal and longitudinal gradients.