Tuesday, January 12, 2016

Sea Surface Temperatures Blow Away Existing Record

Today the Hadley Centre posted its anomaly for December -- it's here -- and 2015's annual average completely destroyed the previous record -- 2014's -- by 0.11°C.

Coming in third is the El Nino year of 1998, at 0.17°C below 2015's value.

This is a huge jump, almost a decade's worth of warming in just one year. (The 30-year trend for HadSST3 is +0.135°C/decade.)


5 comments:

JoeT said...

Thanks for this post, David. Do you know if all the SST data still comes from buoys and ship intake rather than satellites?

Did you see the new Berkley Earth announcement?

"2015 Unambiguously the Hottest Year on Record
According to new Berkeley Earth analysis, 2015 was unambiguously the hottest year on record. For the first time in recorded history, the Earth’s temperature is clearly more than 1.0 C (1.8 F) above the 1850-1900 average...2015 set the record with 99.996% confidence."

It's the confidence level than interested me in particular. Also from the report,

"This year exceeded the previous warmest year, 2014, by 0.14 C. As this amount greatly exceeds the margin of error by over 5 standard deviations, giving 2015 an
unambiguous claim as the warmest year..... The estimated margin of uncertainty on the combined global average is 0.05 C with 95% confidence."

From the 95% confidence level, I have to conclude that 0.05 is 2-sigma not 1-sigma. That also allows 0.14 C to exceed the margin of error by 5 standard deviations. And finally, if we just calculate:

0.5*(1+erf(0.14/(2*0.025)) one gets exactly the 99.996% probability that 2015 is warmer than 2014 as stated above.

However, I'm confused by the 1-sigma = 0.025. Since it's not always clear to me when climatologists refer to an uncertainty as 1-sigma or 2-sigma, last year I used this chart by Gavin Schmidt to conclude that 1-sigma = 0.05. What he's plotting is the pdf for the temperature given the mean and 1-sigma so that the integral under the curve equals one. Here is my comparison of the pdf using 0.05 and 0.025 as 1-sigma. The 0.05 value looks like what Gavin drew.

Do you have any other independent knowledge of what the 1-sigma uncertainty is for NASA versus Berkley Earth? Is it even possible that Berkley Earth has a smaller 1-sigma value than NASA does?

David Appell said...

Joe, thanks.

I think these SST data are taken from ships & buoys, per this page:

http://www.metoffice.gov.uk/hadobs/hadsst3/

"HadSST3 is produced (a slightly more detailed description) by taking in-situ measurements of SST from ships and buoys, rejecting measurements that fail quality checks, converting the measurements to anomalies by subtracting climatological values from the measurements, and calculating a robust average of the resulting anomalies on a 5° by 5° degree monthly grid."

David Appell said...

Joe: Last year I thought that GISS's error bars of ±0.05°C were 2-sigma:

http://davidappell.blogspot.com/2015/01/probability-2014-is-warmest-year-66.html

But I'm not 100% sure. I'll try to clarify that this year when GISS announces a new record.

But I certainly agree this can all be very confusing. Fortunately I think this year's record is so much higher than the previous record that these questions don't matter. But it may matter in the future. But then again, warmest years are mostly PR and don't matter in the long run.

JoeT said...

Thanks David for the comment. I'm interested in the this issue first because probability is interesting in itself. But the secondary reason is that last year we saw so much of the denier class fixate on the low probabilities. This year we know the probabilities will be high, so the blow back will be on the satellite measurements. That's one of the reasons why I appreciate all the discussion of the satellite data on so many blogs.

Just to wrap this part up, I can't go back and reconstruct the temps from 2014 because NASA revised their data since last year. However if I just look at the chart you showed on your blog, 2014 had a probability of 38% for 2014 and 23% for 2010.

So if I take 1 sigma = 0.025, from 0.5*(1+erf(0.02/(2*0.025)= 0.714 then the ratio of 2014 to 2010 is 2.5. If take 1 sigma = 0.05 then the ratio is 1.6. The ratio of the probabilities is 38%/23% = 1.7 That tells me 1 sigma is 0.05

Here's what I get for 2015. I assumed that 2015 will be 0.13 C higher than 2014 because I stole that number from Nick Stokes' blog. Then I took the 12 warmest years and did 5000 samples of a random normal distribution and compared them. Here's what I get:

1 sigma = 0.05 1 sigma = 0.025
2015 95% 99.99%
2014 3% 0.01%
2010 1% -

Here's my takeaway. If Gavin says something like just under 100%, then 1 sigma = 0.025. If he goes with something around 95%, then 1-sigma = 0.05

JoeT said...

I realize that this blog is on "hiatus" (or should I call it a "pause"?).

But in the press conference this morning one reporter brought up the low probabilities last year and asked what they were this year. Gavin said they were 94% for NASA, 99% for NOAA. (BTW, Nick Stokes nailed it by forecasting a 0.13C increase for NASA which turned out to be on the money.)

From the 94% probability for NASA, 0.05 has to be 1-sigma, not 2-sigma.