Posted on January 3, 2020 7 Comments
apologies for my generation’s inability to treat your generation in a just and compassionate manner. My generation is unwilling to choose collectively to inconvenience itself so that your generation might grow up in a minimally stable environment. I can’t explain that, I can only apologize for it.
“Earth Angel” is unfortunate if you remember the lyrics.
Here’s wishing Greta a happy birthday, and many years of joyfully being a pain in the ass to hypocrites who aspire to power.
If possible, could Dr. Foster please answer a question I have based on his recent post on warming trends (a post that’s now closed for comments)?:
I wanted to know if the analysis he performed there for ERA-5 would also work for JRA-55, which I’ve posted below:
I ask, in part, because contrarians/denialists like Ryan Maue and Patrick Michaels laud JRA-55, and try to use it justify low warming projections (even though it actually justifies no such thing). For instance:
So if JRA-55 showed accelerated warming, or even warming roughly on par with analyses such as Berkeley Earth or NASA’s GISTEMP, then that would make Maue’s, Michaels’, etc. position look quite silly.
I’m not speaking for @Tamino, but it seems to me that since Maue’s, Michael’s position in the minority in the published literature, they should be the ones who apply their techniques to Berkeley Earth, GISTEMP, etc, and report.
Moreover, an assessment of JRA-55 by some of its principals describes its predecessor, JRA-25, as having “a cold bias in the lower stratosphere” and that in JRA-55 this “has been diminished”. Admittedly that’s from 2015.
As a Bayesian I find this kind of selection and preference for certain datasets to ascertain a global condition especially problematic, since the Bayesian way is to include all data, even outliers, but downweight them. A lot of this happens with appropriately chosen priors.
My advice to people who want to use JRA-55 is to fuse them together with all the datasets, like Berlekey Earth and GISTEMP, and draw conclusions from the resulting time-varying posterior density. Techniques for doing so are quite standard, e.g.,
Reich, Cotter, Probabilistic Forecasting and Bayesian Data Assimilation, Cambridge University Press, 2015
and software which has been incorporated in publicly available R code:
Gelfand, Banerjee, ““,
Annu Rev Stat Appl., 2017 Mar; 4: 245-266.
Packages spBayes, and the neat new spNNGP.
There are also courses one can take.
I’m sure there are others, e.g., in Python, but Python isn’t my world.
A premature Bayesian named Arnaud Almaric once almost said , “Kill them all and let God upweight the outliers.”
Sorry, the Gelfand and Banerjee paper title should be “Bayesian Modeling and Analysis of Geostatistical Data“.