Comment on Berkeley Earth: raw versus adjusted temperature data by Brandon Shollenberger
Comment on Berkeley Earth: raw versus adjusted temperature data by mike restin
Steven, I read the posts linked at the top and never saw a reference to BEST or Berkeley Earth.
Did I miss it or is this a strawman?
Comment on Berkeley Earth: raw versus adjusted temperature data by Steven Mosher
monthly temperature error.
google nugget
read my comments above.
Jones says the “real error” is .06.
Comment on Berkeley Earth: raw versus adjusted temperature data by Brandon Shollenberger
phi, I can’t say that argument is wrong, but I also can’t say it is right. Verifying it would require looking at a lot of data and seeing how prevalent such patterns are in it. That might be something worth trying, but I have a lot of other things I’d rather do.
Huh. I think I get what you’re saying DocMartyn. I just keep getting distracted by chuckles when I read things like “monsters” and “invisible beasts.”
Comment on Berkeley Earth: raw versus adjusted temperature data by Steven Mosher
then use raw data.
the world will be warmer.
Comment on Berkeley Earth: raw versus adjusted temperature data by Steven Mosher
a better test is to do it blindly on synthetic data.
did that.
then again, only use raw data.
the world is warmer.
Comment on Berkeley Earth: raw versus adjusted temperature data by Carrick
Zeke:
Tests using synthetic data suggest that the Berkeley approach is not too far off, though I agree that its certainly possible that we over-smooth at a regional level. See this memo: http://static.berkeleyearth.org/memos/robert-rohde-memo.pdf
Thank you for the links.
The issue with synthetic data is that ideally it should be created by a third party, and should reflect variation that is found in the environment, without regard to the assumptions of the analysis software.
This prevents assumptions that were written into the code from getting enforced in the synthetic data that is meant to validate the code, or by selection of synthetic data that is unlike to violate the assumptions of the software.
I’ve discussed before some of the assumptions of your analysis code that I think need a closer look at (assumption of uniform, time-invariant correlational field for one), but briefly the Monte Carlo field you are testing against should reflect the realistic noise sources present in the measurement system as well as realistic natural variability of the signal you are trying to measure.
I wrote down a partial list of what I think should be present on Chad’s old blog.
Finally, there’s the question of the metrics you use for testing. For example, you have to start with a definition of what is “optimal” if you are trying to see how close to “optimal” you are able to achieve.
It’s your project so you guys get to define what that means, but it needs to be stated clearly as well as tested for.
Thanks again for comments and your groups hard work in trying to shed more light than heat into this very interesting topic. ;-)
Comment on Berkeley Earth: raw versus adjusted temperature data by Steven Mosher
see my answers above
Comment on Berkeley Earth: raw versus adjusted temperature data by Don Monfort
No, Danny. I am the former Civil War cavalry officer who rode with Gen. Custer at Gettysburg, when the 1st Michigan charged into Jeb Stuart’s vaunted Virginia cavalry and prevented them from crashing through the rear of the Union defenses, during Picket’s charge. Saved the Union. After the war, I became a famous gunfighter. I got out of that business, when I married my Jewish-Jamaican Princess and settled down on a little spread north of town. Send Chester to get me, if you need help.
Your boss is Territorial Circuit Hanging Judge McIntyre. The recalcitrant varmints you don’t shoot, you keep locked up till he comes to town. The Judge will give them a quick trial and a quicker hanging. He’ll give you a tin star and your first month’s pay, when he gits around.
You are going to have to read the script, Danny. And watch this:
The Judge will reimburse you for the $2.99.
Comment on Berkeley Earth: raw versus adjusted temperature data by Carrick
Comment on Berkeley Earth: raw versus adjusted temperature data by Don Monfort
This is why they call him Kid Mosher. Just funnin ya, Steven. But I am serious about the collaboration. Rud can buy you a supercomputer.
Comment on Taxonomy of climate/energy policy perspectives by Pooh, Dixie
This is a fine post. The Discussion section introduces things to consider more broadly. “Do both get fair hearings from policy makers and the public? What are the likely risks and consequences from our current policy approaches?”
For example, What If:
ACTION(1) is adopted but its assumptions are wrong? What happens?
CHALLENGED(2)is adopted but its assumptions are wrong? What happens?
NURTURE(3)is adopted but its assumptions are wrong? What happens?
DELAY(4)is adopted but its assumptions are wrong? What happens?
Your turn: You are in city X. Your job interview is in city Y. There is a long road and a short road between them. You are a bit low on gas. The short road has curves and precipices, but the long road has gas stations. You can just make the interview if you drive fast on the short road. (You have no cell phone.)
Comment on Berkeley Earth: raw versus adjusted temperature data by Steven Mosher
the vast majority of data comes from GHCN Daily unadjusted.
daily data doesnt get adjusted.
the rest of the data comes from hourly GCOS. unadjusted.
a small fraction of the early record comes from monthly.
So if you look at TMAX and TMIN files you see that they go back to around
1820.. 1810 around there.
before that time you dont have many daily files and hence no tmin/tmax.
before I joined BEST I used GHCN D exclusively. with no adjustments.
for anything.
using an algorithm developed by skeptics JeffId and RomanM, I got the same answers.
everyone forgets that skeptics actually did their own algorithm
Why?
because their answer was warmer than CRU
oh ya.. they didnt think it was much warmer..and didnt matter.
Comment on Berkeley Earth: raw versus adjusted temperature data by EdG
After reading this entire comment thread I must say that, thanks to the Josh Earnest type answers from Mr. Mosher, I now take the so called ‘BEST’ data even less seriously than I did before.
It would appear that the whole conveniently named ‘BEST’ project is just an effort to create a slightly tweaked version of the other data sets that somebody can claim is better. Sort of like creating a Good Cop to go along with the Bad Cop.
Comment on Berkeley Earth: raw versus adjusted temperature data by Curious George
Excellent. Did your synthetic data show a warming trend and a cooling trend after a time reversal? I imagined that synthetic metadata might be a problem.
Comment on Berkeley Earth: raw versus adjusted temperature data by Bad Andrew
“the world is warmer”
You mean it’s a Warmer’s World. Play Scientist and tell us how much warmer it is.
Andrew
Comment on Berkeley Earth: raw versus adjusted temperature data by Brandon Shollenberger
Carrick:
Now you just need to repeat in on synthetic data that fully tests your analysis script.
And publish the results so people can actually verify them :P
Comment on Berkeley Earth: raw versus adjusted temperature data by Steven Mosher
Don,
I know you are serious about the collaboration. And yes Rud has the money to get the computer. I’m on board. you wont find anyone else.
dont forget I did my own series on my own dime using skeptics algorithms.
people forget that skeptics already DID their own series.
warmer than CRU.
they all just moved on.
of course nobody harped about the minor issues.. cause welll…look at the HS over there..
Comment on Berkeley Earth: raw versus adjusted temperature data by Carrick
Svend:
If nothing really changes with the adjustments, why do it?
That’s part of the optimization criteria that seem to be getting ignored (the missing ones that need to be there to prevent over-smoothing).
When the change is minor, but it’s destroying regional scale structure in the process, that alternation needs to be eschewed.
Comment on Berkeley Earth: raw versus adjusted temperature data by Steven Mosher
I’ll repeat this phi.
last time willis played the saw tooth card.. ( if unicorns, farted in harmony, it might set up a of constructive resonance that.. well could lead to a station being wrong)
last time he did that I went through the pain excercise of looking for unicorns.
1. The algorithm is not perfect.
2. the algorithm doesnt need to be perfect.
3. you will always find something to argue about whether an algorithm
adjusts or a person does it.
So what do you do. We decided to test the algorithm on synthetic data with errors injected.
What did we want to see?
ON average did the algorithm move you closer to the truth.
it did.
could there be unicorns that it misses. ya. without a doubt.
could it over correct some, and undercorrect others
oh ya.
Will you be able to explain how each individual case is adjusted?
only if you STEP THROUGH THE CODE.
why use an algorithm?
1. to remove the human bias.
2. to create one process that is the same for all cases.
the alternative is to do what CRU do. They trust the experts
in different countries.
3. to create something that you can do SENSITIVITY testing on.
cant do that with humans.
4. Because its open and repeatable, unlike human adjusters.
5. because it scales to 40K stations.