Friday, May 22, 2015

How bad is naive temperature averaging?

In my last post, I described as "naive averaging" the idea of calculating an average temperature anomaly G by simply subtracting from each of a number of local records the (varying) lifetime average, and averaging the resulting differences. There, and in an earlier post I gave simple examples of why it didn't work. And in that last post, I showed how the naive average could be made right by iteration.

The underlying principle is that in making an anomaly you should subtract your best estimate of what the value would be. That leaves the question of how good does "best" have to be; it has to be good enough to resolve the thing you are trying to deduce. If that is the change of global temperature, your estimate has to be accurate to the effect of that change.

If you add a global G to a station mean, then the mean of the result isn't right unless mean G is zero. There is freedom to set average G over an interval, but only one - not over all station intervals. So as in the standard method, you can set G to have mean zero over a period like 1961-90, and use station means over that period as the offsets. Providing there are observations there, which is the rub. But much work has been done on methods for this, which itself is evidence that the faults of naive averaging are well known.

Anyway, here I want to check just how much difference the real variation of intervals in temperature datasets makes, and whether the iterations (and TempLS) do correct it properly. I take my usual collection of GHCN V3 and ERSST data, at 9875 locations total, and the associated area-based weights, which are zero for all missing values. But then I change the data to a uniformly rising value - in fact, equal to the time in century units since the start in 1899. That implies a uniform trend of 1 C/century. Do the various methods recover that?

Monday, May 18, 2015

How to average temperature over space and time

This post is about the basics of TempLS. But I want to present it in terms of first principles of averaging, with reference to other methods that people try that just don't work. I'll mention the least squares approach described here.

I once described a very simple version of TempLS. In a way, what I am talking about here is even simpler. It incorporates an iterative approach which not only gives context to the more naive approaches, which are in fact the first steps, but is a highly efficient solution process.

Sunday, May 17, 2015

Emissions hiatus?

John Baez has a post on the latest preliminary data from EIA for 2014. Global CO2 emissions are the same as for 2013 at 32.3 Gtons. EIA says that it is the first non-increase for 40 years that was not tied to an economic downturn.

They attribute the pause to greater use of renewables, mentioning China. Greenpeace expands on this, saying that China's use of coal dropped by 8%, with a consequent 5% drop in CO2 emission. They give the calc with sources here. This source says April coal mined in China was down 7.4% on last year, which they do partly attribute to economic slowdown there.

It's just one year, and may be influenced by China's economy. We'll see.

Friday, May 15, 2015

JAXA Hiatus

If you're following JAXA Arctic sea ice, they have announced an interruption for maintenance, until May 20. Pity, it is getting interesting, with 2015 now at record low levels for the day. Maybe it was going off the rails - we'll see.

NSIDC NH isn't looking so reliable either - they showed a massive re-freeze yesterday.

Update - JAXA has unhelpfully taken down all its data, and replaced it with the warning (in Japanese), which overwrote my local data. So no JAXA table data here till that is restored. Fortunately the plot is OK.
Update - Jaxa is back.

Thursday, May 14, 2015

GISS down by 0.1°C in April

GISS has reported an average anomaly of 0.75°C (h/t JCH). They raised their March estimate to 0.85°C (from 0.84), so that makes a difference of 0.1, which is what I also reported for TempLS. In fact, TempLS has since crept up with extra data, so the difference is 0.09°C. Metzomagic comments on recent GISS changes here.

Update: GISS has published an update
May 15, 2015: Due to an oversight several Antarctic stations were excluded from the analysis on May 13, 2015. The analysis was repeated today after including those stations.
This actually made a big difference. The April temperature anomaly comes back from 0.75°C to 0.71°C, and March from 0.85°C to 0.84, making a drop of 0.13°C. GISS uses an extra Antarctic data set from GHCN, which I presume is the one involved. So this won't affect the TempLs calc.

Below the fold, as usual, I'll show the GISS plot, and the TempLS spherical harmonics smoothed map.

Tuesday, May 12, 2015

BoM declares El Niño status

Australia's Bureau of Meteorology has upgraded its ENSO Tracker to El Niño status, saying
"El Niño–Southern Oscillation (ENSO) indicators have shown a steady trend towards El Niño levels since the start of the year. Sea surface temperatures in the tropical Pacific Ocean have exceeded El Niño thresholds for the past month, supported by warmer-than-average waters below the surface. Trade winds have remained consistently weaker than average since the start of the year, cloudiness at the Date Line has increased and the Southern Oscillation Index (SOI) has remained negative for several months. These indicators suggest the tropical Pacific Ocean and atmosphere have started to couple and reinforce each other, indicating El Niño is likely to persist in the coming months.

International climate models surveyed by the Bureau indicate that tropical Pacific Ocean temperatures are likely to remain above El Niño thresholds through the coming southern winter and at least into spring."
You can see an animation of the last 50 days of ENSO-responsive SST here.

Sou has more on the story here.

ps In terms of Arctic sea ice, 2015 has been lately well ahead in melting of most recent years. It has just passed 2006 (which is fading after an early spurt) to have the least ice for this day (12 May).

Monday, May 11, 2015

Station temperature trends - new page

I've promoted another earlier post that I find I quite frequently want to refer to. It is upgraded to use WebGL and current data. The page is here.

 This page shows a shaded WebGL plot of station trends of monthly average temperature over various periods to present (you can choose). The coloring is intended to represent the station values exactly, with continuous shading in between. It uses primarily unadjusted GHCN V3 with ERSST, though adjusted GHCN is an option. It follows posts here and later here. Operating advice below.