Tuesday, January 19, 2016

Mind Bending

Eli and Tamino have posted about the obvious long time deviation of the lower troposphere satellite temperature records often called TLT.  Eli and Steve Mosher, with friends and cynics have been doing a dosy do at ATTP and Rank Exploits.

It looks like either a) there is some aging effect in the AMSU receivers or b) the atmospheric/surface composition has shifted in the last 15 years or so (e.g. less ice/ more water vapor) in a way that biases the returns.

It started with Nick Stokes looking at the trends of various temperature anomaly records

clearly showing that UAH 6.0 trends vary strongly for RSS and UAH 6.0 and the various surface records and UAH 5.6

In the midst of the usual ill tempered fro and to about trends at Rank Exploints, it struck Eli that there were really two questions, the long term trends about which much had been said, and the actual measurements which take place over a day or less and about much less has been said, at least in blogs and Congressional hearing, or even on the radio. 

To get at this Eli compared the monthly variation in CRUTEM4 and RSS, showing that they were a pretty good overlay.  Tamino showed both that on the short term (months) there was a perfect match between the UAH and RATPAC balloon sonde records but that they deviated starting in about 2000

Both comparisons show that while the climate system has a fair bit of variability on a monthly or a yearly basis (Hi Judy), the instrumental noise, e.g. the noise inherent in the measurement process is much smaller by comparison.  

Eli's original POV was that the drift is most likely in the AMSU satellites or the processing of the AMSU data. Unanticipated aging of the receiver or the internal hot calibration target seems to Eli most likely, although there might be something involving orbital decay (less likely now because this caused a lot of trouble early on) or even changing land/sea/ice patterns which affect the AMSU response.

However, upon reflection it appears equally likely that there has been some change in the atmosphere (humidity was suggested) or surface emissivity (Mosher's idea) that has befuddled the atmospheric model used by RSS and UAH. That the same effect is seen in RSS and UAH 6.0 indicates that the atmospheric models are idempotent or close.

That the break between the RSS and UAH records and the balloon sonde/surface temperature anomaly records come at the same time as the change over from the MSU to the AMSU, A standing for advanced, makes it hard to choose.

It is well known that the (A)MSU sensors have trouble with ice and snow as well as measuring over high land, so Eli, in his naivety, thinks that point by point comparison with temperatures measured by the (A)MSUs at specific locations might be useful and Mosher has a project moving in this direction
also the weighting function relies on an assumption of a constant emissivity for earth. gimme a few days and I may be able to tell you if gridded delta’s between RSS and BE are correlated with changes in emissivity.
But, rather than go much further into this, Eli would like to point out how this discussion has been mind bending.

A principal attack on RSS and UAH has been that the measurements are five km or so up there and we live on the surface.  The short term global comparisons that Eli and Tamino have done shows that the precision of the (A)MSU measurements as estimates of immediate global temperature are pretty good.

The question now is whether the long term drift is engineering or science

Extending the comparisons between surface, balloon sonde and satellite measurements to specific regions and a daily basis will really nail the precision and maybe the source of drift.  Opportunity exists for using aircraft platforms (research and commercial) to even improve on this and, of course there is the Taiwanese/US COSMIC GPS occultation program.

Closing the loop on temperature anomaly measurements is within our reach.  At that point each of the methods will support confidence in the other and the strengths of each will allow a deeper understanding of the climate system.

ADDED:  In the comments Eric Swanson posted a table which Blogger could not handle really well, of trends.  Here it is prettified.


Tom Gray said...

Wow. Can't say I understand it all, though I did get a few of the simpler sentences, but seems very Impressive. Thanks for puttinh this together, Eli.


I realize that millimeter wave channel allotments try to keep this bandwidth slot clear for geophysical use, but is there any info on flux changes from harmonics-- wifi sferics , for instance?

Everett F Sargent said...


I guess I'm missing out on the ATTP and Blackboard discussions. :( Because I'm way too busy at the moment looking at the global TLT datasets.

RE: RSS TLT bias offsets or RSS v3.3 (most recent data) versus RSS v3.2 (archived at RSS) versus RSS v3.1 (archived at RSS)

The biggest change or bias offset, IMHO, in those three versions is RSS v3.3 minus RSS v3.2, there you will see a sudden divergence starting 1999-01 (January 1999) and a relatively large oscillatory negative bias offset beginning 2002-08 and ending at 2010-12 (the end of the RSS v3.2 dataset).

The 2002-08 through 2010-12 negative bias offset has a annual anomaly cycle (something That I would not have expected except for the fact that it occurs outside the RSS baseline anomaly period of 1979-1998).

What caused this negative offset bias?

See this RSS "Version 3.2 to Version 3.3 Differences" (dated "Carl Mears, February 7, 2010" which is incorrect it should be dated "Carl Mears, February 7, 2011" (the original creation date of at least one version of that PDF file and the Mears 2011 paper appeared right at that time)).
Publication History
Issue published online:
20 April 2011
Article first published online:
20 April 2011
Manuscript Accepted:
9 February 2011
Manuscript Revised:
7 February 2011
Manuscript Received:
25 August 2010

Note: Mears, et., al. used RSS v3.2 for that particular paper.

----End of Part One----

Everett F Sargent said...

---Start of Part Two---

Anyways, from the "Version 3.2 to Version 3.3 Differences" change log file Figures 1 and 2 are most useful (particularly Figure 1 though).

I can reproduce all of Figure 2 including the +1C plotting offset for the "V3.3 miinus V3.2" differenced time series (but it's best to look at the differenced time series in isolation without the offset).

From Figure 1, you will see a "switch" or inclusion of the "AQUA" satellite at almost exactly 2002-08.

The 1999-01 adjustment is not nearly as "egregious" to the overall long term trend change but it occurs right about the time NOAA-13 was retired and NOAA-15 came online (I seem to remember that there were some issues with NOAA-14, so whatever).

Back to the 2002-08 through 2010-12 negative bias offset.

If one only corrects for that eight year five month offset of approximately -0.046C (my anomaly calculation for the eight complete years, there is no discernible trend for that eight year five month period, however the period from 1999-01 through 2002-07 has a very linear negative trend (but alas w-a-a-a-a-a-y too short to do anything with)) AND extrapolates that same offset through 2015-12 (or five additional years (I can't support a trend change given the differenced dataset, only a constant offset would appear or seem to be justified), then I call this a "rollback" in this case, a rollback from v3.3 to v3.2.

The 0.046C is then added back to all years starting 2002-08 through 2015-12 (current month) to v3.3 (that is the only change I've done), a period of 13 years and five months then the current RSS v3.3 trendline changes from ...

+1.23 C/century to ...
+1.40 C/century

The spreadsheet is available to anyone who might wish to see what kind of "trouble" I can get into. I have not looked at the RSS TMT time series yet. Also from the nine regions contained in the global file all have this very quirky eight year five month issue between 2002-08 and 2010-12 (the end of the archived RSS v4.2 dataset).

My biggest problem is adjudicating if this was an issue with RSS v3.2 (error in that version) or RSS v3.3.

As most people appear to suggest 2002-2003 as the point in time when the TLT time series "drifted" from the earlier record, this may be due to switching satellites and going from MSU to AMSU instrumentation or somesuch (I don't know the exact history, satellites and instrumentation types used throughout the satellite era by the various groups).

Very long post. Hope I"ve helped in some rather very minor way. If the above is too technical or too gibberish sounding I'll try to answer those types of questions to communicate the above better.

That is all.

Everett F Sargent said...

Minor typographical correction to 2nd part of my two posts ...

"archived RSS v4.2 dataset"

... should read ...

"archived RSS v3.2 dataset"

Sorry about that.

Everett F Sargent said...

Found another one (error) ...

"it occurs right about the time NOAA-13 was retired and NOAA-15 came online"

... should read ...

"it occurs right about the time NOAA-12 was retired and NOAA-15 came online"

Per the change log file Figure 1.

Sorry about that one too. :(

E. Swanson said...

I've been playing with the latest UAH v6 data. I used the TMT and TLS time series to build a "TTT" series with the thought of comparing that with the RSS TTT, which is calculated as (1.10*TMT -0.10*TLS). UAH and RSS use different latitude ranges in their data sets, for example, UAH Extra Tropical latitude is 20-90 and RSS Mid Latitude is 25-82.5.

Here's what I found:
Comparison of “satellite Temperature” trends, degrees/decade

Region UAH 5.6 UAH 6.0 ES TTT RSS TTT
Globe 0.14 0.11 0.11 0.11
NH 0.19 0.14 0.13 0.16
SH 0.09 0.09 0.09 0.07
N Polar 0.43 0.22 0.19 0.25
N Ex Tropics 0.25 0.16 0.15 -----
N Mid Lat ----- ----- ----- 0.16
Tropics 0.08 0.10 0.10 0.12
S Mid Lat ----- ----- ----- 0.05
S Ex Tropics 0.09 0.08 0.08 -----
S Polar -0.02 -0.01 -0.01 -0.06
(sorry for the lack of proper formatting, but the "pre" tag doesn't work)

Eric Swanson
19 Jan 2016

Spencer's description of their algorithm for each grid point is:
TLTv6 = (1.538xMSU2 - 0.548xMSU3 + 0.01xMSU4)
He didn't indicate what they do with the AMSU data, which exhibits slightly different theoretical emission weighting profiles...

EliRabett said...

The bunny feels quite Tom Sawyerish about this. Thanks Eric and Everett

Unknown said...

Longer term it seems the MSU and AMSU units will be phased out and replaced by Advanced Technology Microwave Sounder (ATMS) units.

Accord to the RSS site, "The first ATMS was launched on October 28, 2011. Measurements made by the ATMS are not yet used in our data set. We are working to cross-calibrate ATMS with AMSU so that ATMS measurements can be included in the future."

The first mission was Suomi NPP and the next will be the JPSS-1 mission in 2017.

Hopefully these satellites will give greater precision. Does any one know more about the new ATMS sounders?

Everett F Sargent said...

David Sanger,

Truth be told, we could already have perfect satellite data.

So even if we did, which we don't, the various channels have filters applied (the old ones might have been mechanical moved in place 'analog' filters, the newest ones might be pure digital filters for all I know).

There is no such thing as a perfect filter, mechanical, analog or digital.

Those sensors spinning up there in space can't be calibrated in situ through a known medium with specific optical properties. So the sensors (channels) see the entire atmosphere through those rather imperfect filters (mechanical or analog or digital).

So we get numbers of something called the TLT or TMT or whatever. But what you won't find are numbers for exactly what/where the TLT or TMT is, that's for sure, which part(s) of the air column do they represent. Where is it (not a specific isobar or elevation/altitude) or how thick is it (err, a few kilometers or more (not likely to be less), come on, deal with it). We already know that it can't be a surface measurement simply because IT"S NOT A SET OF MEASUREMENTS TAKEN AT THE SURFACE. Those satellites can't distinguish the surface from 1km above the surface or 2km above the surface or 3km above the surface, etceteras, ad nauseam, ad infinitum.

So what's a good climate scientist going to do with measurements somewhere's in the troposphere and somewhere's in the stratosphere?

Three things in lieu of those measurements:

(1) Observations from elsewhere's (radiosonde data and there is a heck of a lot of that type of work being done today, updating radiosonde databases, that is, along with homogenization to derive global products).

(2) Reanalysis data (e. g. model data based partly on the underpinnings of the radiosonde and other 'short term' observational data (many of those from space)).

(3) Earth system models (e. g. AOGCM's).

The advantage of all three of the above is that they can tell us what is going on at different (constant) pressure levels (isobars, which in turn can be used to infer much more accurately elevation/altitude), specifically they can help us to independently determine the temporal and spatial temperature trends as we rise up throughout the air column.

I've tripped over quite a few recent papers that incorporate one to all three of the above methods.

So maybe the TLT and TMT satellite measurements are all correct, although through ever changing metrics due to instrument changes etceteras. You get to choose whatever one suits you like Monkers does.

I believe that the TLT TMT measurements are the only set of measurements I've ever come across that don't have a set of well defined spatial and temporal benchmarks (they're not grounded, but are somewheres up in the atmosphere, no weigh stations to tell if you are still on the right track and are not drifting, we'll just assume stuff, a lot of stuff)

Also, I don't know what all I am talking about. So how was my special pleasing and/or appeals to emotion? YMMV

Everett F Sargent said...

"special pleasing" should be "special pleading"

Hank Roberts said...

> Russell ... is there any info on flux changes from harmonics
> -- wifi sferics , for instance?

Damned good question.
Briefly 'oogled
Electronic Noise Is Drowning Out the Internet of Things
Our increasingly connected world needs better protection against RF noise pollution
18 Aug 2015

I have a LED flashlight (with a driver circuit board) that has set off some car security alarms when I happened to fiddle with brightness levels while close by.

A decade ago the only thing I owned capable of triggering car alarms was a 5-watt 2-meter ham radio.

Everything around nowaeays is both noisier and more sensitive to noise.

I can walk around with an AM transistor radio and find rather alarming levels of electronic smog from most of the compact fluorescents and, even more so, the new LED lights I've bought.

China must be a hugely noisy place in the radio spectrum nowadays.

And the satellites are just barely above the atmosphere.

Tangentially, I wonder:

Ya know the "rogue wave" problem: They thought they were just mythical for centuries, until lately. Turns out you combine a variety of different wavelengths and intensities and eventually you get a point or line on the ocean that's the sum of them all, and a huge wave (or, presumably, hole) finds your ship and maybe breaks it.

Well, that happens in 3-D for radio frequencies too, doesn't it? There must be sporadic high intensity spikes happening everywhere, more and more often.

All we need is a Maxwell's Demon operating in that range and we can power everything by collecting random energy from everything, eh?

steven said...


Glad I found this. Well the first stage of processing is all done.

Here is a short synopsis of where I stand and what I have.. if you have questions or ideas lemme know.

I started with Berkeley Earth (BE) 1 degree gridded fields for
Land Ocean, and Land Tmax and Land Tmin. In addition I used the TEMPERATURE fields for RSS ( yes they do have Temperature (K) grids in 2.5 degrees )

The main reason for Using RSS temperatures is I wanted to check for Temperature inversions.
Yes the data show temperature inversions at the poles.Quite a bit in fact.

The next step was getting the data to a common resolution and common masking. So the data was re gridded to 1/2 degree and RSS was used to mask BE. RSS lacks coverage at the poles and over certain high elevation locations-- andes, tibet, etc.

So I end up with a bunch of datasets: Land Ocean, Land Tave, Ocean, Land Tmax and Land Tmin. For RSS I have Land Ocean, Land, and Ocean.

The main reason for making a Tmax Land and Tmin Land is that it would appear that RSS normalize all measurements to a local noon. And with the land surface temps we Tmax ( happens at different times ) and Tmin.
For the Ocean there is just one temp.

The purpose of all this was to see if any of these cuts through the data
made a difference. Lets say the trends in Land Ocean are .2C/decade
for BE and .17 for RSS ( notional numbers ) On the assumption of a drifting sensor... one might expect to see the same kind of trend difference
accross the various cuts.. so if delta trend was .03 in Land Ocean
we might expect the same for just the Land.. and Just the ocean.

Well I can tell you they are all different.. details to follow ( kinda swamped now ). The biggest difference in Trend comes between BE Tmax
and RSS. This is a land only measure. The trend is almost double for BE.

To figure out if this is related to some non receiver based issue I have constructed monthly maps of the Temperature difference ( lat/lon,month)
These are 1/2 degree maps. Some Hovmoller charts gave the impression that the divergence was growing over time and growing at northern latitude.

side note.. from 1979-1999 the difference between RSS and BE Tmax is essentially trendless. from 1999 on is where the divergence starts.
nearly coincident with the addition of noaa 15.

So for every 1/2 degree cell in the map I calculated the trend in the difference. Biggest trends are all in the northern latitudes..
and also in south america and tibet at high elevation locations.

Need to dig into the inversion layer issues and may do some seasonal cuts.

Emmissivity might take longer cause I want to use Aster.. all the other emissivity data is merely based on land class ( at its heart ) Aster is
90m data.. so a long data crunch.

Tom Dayton said...

Eli, on the issue of balloon-versus-satellite temperature index drift versus matching ups and downs, especially post 2000: Do you think Christy and Spencer's R^2 approach is sufficient to speak to RSS 4 versus UAH 6? http://www.drroyspencer.com/2016/03/comments-on-new-rss-v4-pause-busting-global-temperature-dataset/

EliRabett said...

Short answer is don't know. Long one is to match up the records vs. Sonde on a very short time basis w. geographic location, That at least would say something about the variation. The problem with satellite/detector/calibration source would be much harder to get at, although if some MSUs flew a long time into the AMSU record or the newer systems show variation that might be a strong hint,

Tom Dayton said...

barry at Open Mind gave a heads up to a reply by Carl Mears to criticisms of the RSS 4.0 approach: http://www.remss.com/blog/RSS-TMT-updated