Didn’t realize Don Quixote worked at CDC now…

You mean to tell me that a model trained on average quality code is producing code that is of average quality? I thought we could spin straw into gold!

Basically, the argument “we don’t need as much P inputs because we seldom see a yield response”, is meaningfully different than “we don’t need as much P inputs because we defined a response range and most soils are below that” in terms of the way it gets put into practice.

Yeah, I saw similar things in some of my PhD work: no yield response even after 12 years of no P. My concern is mostly about being right for the wrong reasons, since that intervening step is where decision making occurs

acsess.onlinelibrary.wiley.com/doi/abs/10.2...

I haven’t had time to fully digest here, but isn’t another interpretation here that Olsen P is just a bad proxy for crop P response? The huge drop in R2 between initial model to LOOCV (from 0.61 to 0.42) doesn’t inspire confidence in the subsequent analysis of fertilizer P needs in Europe.

I didn’t realize you were taking it over - very cool! I use the package all the time, so I appreciate your work on maintaining it 😊

Oh and get a better calorie conversion from field to human stomach, ie food waste and livestock production.

I will further underscore my seriousness and humility by wearing a cape while posing dramatically for movie posters.