Ever since NOAA released their latest update to sea surface temperature, version ER-SSTv4 (Extended Reconstruction Sea Surface Temperature version 4), it — and they — have been the target of vicious attack. It has come not just from climate denier bloggers, but from politicians like Lamar Smith (R-TX, chairman of the House committee on Science, Space, and Technology). The accusations haven’t been limited to error, rather they have focused on claims of outright fraud by NOAA scientists, saying, without any justification whatever, that the new version was an attempt to deceive, simply because it shows faster recent warming than other versions.
But new research has not only vindicated them, it establishes that their latest update shows every sign of being the best sea surface temperature data set yet. As in, the best.
Composite sea surface temperature data sets come from numerous instruments, including shipboard measurements from buckets (both wooden and canvas), ship data from engine intakes, buoys at sea, ARGO floats, and satellite radiometry. One of the trickiest aspects of making a global historical data set is piecing together the records from all these different instruments. In the new study, Hausfather et al. compared sea surface temperature data sets to data which are instrumentally homogeneous, coming from only a single type of instrument. These were compared, for recent data (when they’re all available), to the main historical reconstructions, to see which best match the instrumentally homogeneous sea surface temperatures (IHSST). The data sets so tested include HadSST3 from the Hadley Centre/Climate Research Unit in the U.K., the Japanese Meteorological Agency’s Centennial Observation-Based Estimates of SSTs (COBE-SST), the older NOAA data set ER-SSTv3b, and the updated ER-SSTv4.
And the winner is — hands down — NOAA’s update, ER-SSTv4.
What impresses me about this research is how thorough they were. They compared data to IHSST both using only regions where data was present, and infilling to make more complete geographical coverage. They used multiple versions of IHSST for ARGO floats and for satellite radiometry. The main work was to compute the difference between the main data sets and the IHSST data sets, and estimate the trend of the difference series (by linear regression) for recent times, so that a difference trend of zero indicates a good match (no trend bias). Uncertainties in trend estimates were corrected for autocorrelation using an ARMA(1,1) model (which pleases me personally). They even applied a correction to estimates of the autocorrelation parameters (which doesn’t make a huge difference, but is still noticeable when using short time spans).
As an example, using regions with data in common (no infilling), the estimated trend rates of the difference series look like this (the second panel uses a shorter time span so comparison can be made to ARGO floats):
In every case the ER-SSTv4 data compare more favorably than its rivals. In fact, the others show a significant cool bias, underestimate the sea surface temperature trend and, because SST is a big part of global temperature estimates, contributing to the mistaken notion of a “pause” or “hiatus” in global warming. In my opinion, much of the motive for the false accusations against NOAA’s latest update is the fact that it undermines climate deniers’ favorite talking point, the “hiatus” that never was.
Of course the new research has already been criticized, but what’s striking about it is how feeble the criticism is. In part this is due to the thoroughness of the research, in part due to the desperation of deniers.
My overall opinion: the new research makes blatantly obvious that the accusers owe a huge apology to NOAA scientists, who were “rewarded” for making the best SST data yet by being slandered. The biggest apology should come from Lamar Smith, who should be removed from the House committee on Science, Space, and Technology for his gross incompetence and unethical behavior. Just my opinion.
As for any apologies, or even retraction of false accusations against NOAA scientists, my advice is: don’t hold your breath.
This blog is made possible by readers like you; join others by donating at Peaseblossom’s Closet.
Open Mind 
At another blog somebody claimed it’s improper to use lower resolution and higher uncertainty data to adjust higher resolution and lower uncertainty data.
First, is that always true? Second, did Karl do that?
[Response: No, it’s not true. One should, however, take the uncertainties into account, weighting things appropriately to account for their variable uncertainties.
That’s one of the innovations of ER-SSTv4, to give greater weight to more precise data.]
JCH, I think it is interesting to examine this question from the point of view of information theory. A lower-resolution dataset may contain within it information not contained in the dataset with higher resolution. As such, the combined information may yield more understanding than either dataset singly. Ideally, if, as Tamino says, the uncertainties are properly accounted for, the combined dataset should at a minimum be no worse than the best dataset you started with.
“But new research has not only vindicated them, it establishes that their latest update shows every sign of being the best sea surface temperature data set yet. As in, the best.”
For the recent trends in the global mean. For data around WWII, for example, I would advice to use HadSST.
http://variable-variability.blogspot.com/2016/08/naive-empiricism-and-what-theory.html
[Response: I agree there are definite signs of problems in the WW2-era data.]
Karsten Haustein showed elsewhere a glimpse of the upcoming new HADISST2:
It follows ERSST4 well in the satellite and buoy era, but it doesn’t exaggerate the peaks and valleys like ERSST4 further back in time. It looks good and will probably be another nail in the coffin for the Lamaristic Inquisition and its followers
JCH,
A common complaint was that buoy temperatures were shifted up 0.12 C, instead of shifting ship temperatures down by 0.12 C. But ERSST reports _anomalies_, and when you substract the baseline off it removes that effect from reported values or trends. Doing it the other way around has zero effect. Literally. Zero.
For the critics who’ve spent time on this, the arrogance of having a pop at scientists when you can’t get your head round pretty basic maths concepts is shocking.
Tamino: thanks for covering this and I think your work implicitly contributed to the choice of ARMA(1,1). Could you add a clarification though? You say that “Composite sea surface temperature data sets come from numerous instruments…”, which is true, but one of the things I like about this paper is that ERSST4 doesn’t use Argo or radiometers. So the buoys are verification and the satellites/Argo are validation.
(posting outside work hours, everything here is personal opinion and doesn’t represent NASA, Caltech or JPL)
“For the critics who’ve spent time on this, the arrogance of having a pop at scientists when you can’t get your head round pretty basic maths concepts is shocking.”
It shouldn’t be all that shocking since this is a well worn tactic, and not an error on their part. There are no rules to follow if you think, as deniers do, that this is a war to be won, rather than a truth to be found.
As I understand this, the NOAA series, ER-SSTv4, is used in GISS L&O.
Shouldn’t these more accurate post 1998 satellites be casting some doubt on the UAH and RSS surface series? If ZH can affirm NOAA ocean with post 1998 satellites, what would post 1998 satellites indicate about the accuracy of UAH ocean and RSS ocean? If not, why not?
Hi JCH,
At least for longer time series (1979-2016) MSU data agrees pretty well with surface data over oceans. Its over land that you see big differences, though there are a number of reasons why its more challenging to correct for diurnal drift over land in satellite data.
Zeke Hausfather,
Thank you. That is an exceedingly useful piece of learning. Certainly the Ocean component of UAH TLT v6.0 demonstrably matches the NOAA Ocean component very closely. (RSS do not provide an Ocean value for public consumption.) This leaves the trolls with only the land temperatures to argue about and that is a whole different ball game.
ZH – thanks for taking the time to communicate science to average me.
Al Rodger – it looks to me like the divergence originates in 2004; it sort gets back in line after that; and then it widens from 2011 onward. It would be interesting to see if you agree.
Hi Al,
RSS has troposphere-over-ocean data on their FTP site here: ftp://ftp.remss.com/msu/monthly_time_series/
In general, you should find this figure of interest comparing surface and satellite trends over land and oceans (only NOAA is shown for surface data, but they all are pretty similar).
JCH,
I’ll tell you what I’m looking at:- I grabbed the NOAA global ocean and the UAH TLTv6 global ocean, calculated the average of each series 12/79 to 11/16, subtracted it to align them and plotted their 12-month rolling averages.
And what do I see:- the two wobbly traces sit quite sweetly except when the wobbliess becomes more meander and less sine wave, during he periods 92-96, 99-04 and 2011-to-date. During the first two of these periods, UAH seems to hang high & since 2011 it seems to hang low. (I note that your two mentioned years 2004 & 2011 both feature in this description, but not helpfully.) As far as my three periods are concerned, they all three periods when (with a bit of licence time-wise) MEI was not showing strong El Nino/La Nina cycles.
Sorry if this isn’t helping describe what you are looking at.
Myself, I am impressed by the match. The two glolbal ocean series yield trends that are not much different given they have such differently sized wobbles – UAH +0.107ºC/decade(+/-0.014), NOAA +0.118ºC/decade(+/-0.007). So, given the difference in the land+ocean trends, the Land trends must be very different (something like 0.16ºC/dec UAH, NOAA 0.25ºC/dec?).
Hi Zeke,
Could you explain a bit why it is more challenging to correct for diurnal drift over land. Does this have anything to do with why RSS TLTv3.3 hasn’t been updated yet?
JoeT:
Could you explain a bit why it is more challenging to correct for diurnal drift over land.
Yes, I was intrigued by that comment too. It did occur to me that land temperatures have a larger spread over 24hrs than SSTs. That would mean that failing to correct for diurnal drift is more important for land temps; but that isn’t quite what Zeke said – so I too would welcome some expert insight.
There are a couple reasons why MSU temperatures are harder to get right over land. One big one is that the satellites measure the same location at different times of day on each pass-by, and require a model to account for the effect of observation time on temperatures. Over the oceans this is easy, since the difference between daytime and nighttime temperatures is smaller and relatively well-constrained. Over land it varies widely over time even for the same location, and requires a climate/reanalysis model to account for. There are also differences in emissivity due to the heterogeneity of surface types (and elevations) over land that are largely absent over water (apart from areas with sea ice).
Zeke, Thanks for taking the time to reply.
Hi Zeke,
Correct me if I am wrong, but aren’t the birds on which these units fly in sun-synchronous orbits? Shouldn’t that limit TOD issues?
[Response: Zeke probably knows better than I do, but if I’m not mistaken one of the motives for the latest RSS update (to v4) was that the orbits drift, and that drift wasn’t adequately addressed in RSSv3 (or in UAH for that matter).]
Hi Tamino,
Certainly orbital drift is an issue, but that is distinct from TOD issues. Drift is an issue that takes place over time due to drag, inhomogeneities in the gravitational field, etc. I would think that drift would mainly be important on timescales of years.
Hi Snark,
You are correct, MSU sensors are nominally in sunsynchronous orbit. But the effect of orbital drift on observation time is non-trivial over the life of the satellite, and the change in measured temperature due to changes in observation time is large relative to the magnitude of the observed trend. I should have been a bit clearer about that.
I imagine that one reason that this is harder over land is that the land a pretty variable background seen by the MSUs, while the ocean is pretty homogenous. The diurnal variation of temperature of a parking lot, residential neighborhood, downtown city, desert, farmland, forest or swamp are all rather different to each other.
I’m not sure of the beam size of the MSUs, but there must be substantial variation of the diurnal cycle in different beams.
Thanks, Zeke, and sorry to be a pedant. That makes sense.
Pete,
The instruments are passive–there is no “beam”. They merely receive microwave emissions over their field of view.
While the Ocean data for the TLT satellite & the surface records gives a very impressive result, I was surprised (enought to wonder if I’ve made a slip somewhere, but I cannot see it) to see trend-wise the GISS Land record was a halfway-house between the NOAA (& BEST & CRUTem4) Land data and UAH TLT v6.0, GISS is also a lot less wobbly than NOAA/BEST/CRUTEM as well as being less slopey. (The UAH wobbles are not much greater than NOAA/BEST?CRUTEM.) The trend numbers work out as follows (the 2013 end dates being to allow a comparison with BEST).
NOAA – land
12/78 to 09/13 +0.260C(+/-0.031)
BEST – land
12/78 to 09/13 +0.256C(+/-0.032)
CRUTem4
12/78 to 09/13 +0.261C(+/-0.024)
GISS – land
12/78 to 09/13 +0.192C(+/-0.017)
UAH v6.0 – land
12/78 to 09/13 +0.166C(+/-0.024)
[Response: If you’re using the GISS met stations record, be advised it’s not really a land-area estimate. Instead it’s a *global* estimate based on land-station data only. You can get a land-area GISS record by averaging the gridded GISTEMP data over land areas (Climate Explorer will do that for you).]
Thank you. That has indeed sorted it.
NOAA – land
12/78 to 09/13 +0.260C(+/-0.031)
BEST – land
12/78 to 09/13 +0.256C(+/-0.032)
CRUTem4
12/78 to 09/13 +0.261C(+/-0.024)
GISS – land
12/78 to 09/13 +0.265C(+/-0.030)
UAH v6.0 – land
12/78 to 09/13 +0.166C(+/-0.024)
snarkrates,
the drift in the orbital parameters chances also the orbital plane of the initial sun synchronous orbit. The change can be several hours of local time over the years, see the figure 2 in this (open access) paper
http://journals.ametsoc.org/doi/abs/10.1175/JCLI-D-15-0744.1
Zeke Hausfather,
how about sea ice in the ocean. Does it have the same behaviour as land or as ocean for the diurnal drift in satellite data?
How does the difference behave for different latitude bands? Or different seasons?
In the case of inversions near the ground in winter, I suspect a large difference between surface and satellite data sets. Am I right?
snarkrates: All receiving instruments have a beam pattern (usually referred to as just a “beam” in radioastronomy and microwave remote observing). It’s the angular sensitivity dependence of each pixel in the instrument.
For a single pixel instrument like the MSUs, the beam pattern is same as the field of view.
Point of order. Can we stop calling these people climate change deniers. It’s long since gone past that point. These people are full on science deniers. They are anti-science, not just one aspect of it. They can’t just pretend to be pro-science in one area and anti-science in another because the application of the scientific method with all its pros and cons is the same across all areas of science. To be claiming it’s false in one area is to be claiming it’s false in all of them.
So science deniers, not (just) climate change deniers.
JCH,
Remember that the satellites are not measuring temperature, but rather proxies for temperature. They require models to constrain temperature with the proxy data. One of the issues identified in this work is the fact that whether the data are taken at the surface of the ocean or below it accounts for a lot of the relative disagreement. RSS and UAH invariably suffer from noise introduced by the layers of atmosphere above the lower troposphere. You can see that in the radiosonde data. The latest RSS generally does a better job than UAH, but this is a difficult problem.
snarkrates – people are thinking there is a double standard.
ZH used a satellite record to affirm the ocean component of GISS L&O, and yet argue the RSS and UAH L&O series do not call the GISS L&O series into question (I one of them who argues RSS and UAH cast little if any doubt on the GISS series.)
It just seems impossible that RSS ocean and UAH ocean (they have to have some version of oceans only) can agree with the satellite data ZH used, as it apparently agrees well with the buoy data and the ARGO data… and GISS Oceans (ER-SSTv4).
JCH,
UAH and RSS is troposphere above the oceans, and I believe that ZH used satellite readings of ocean surface skin temperature, or near surface temperatures, more or less the satellite part of the OISST 0.25 degrees dataset.
I think that RSS used its own Atmospheric water vapor product (oceans 60N-60S) to validate RSS v4.
UAH 6 would probably not pass such a validation. I did this exercise one year ago:
Olof R – wood for trees has land only series for UAH 5.6 and 6.0 and RSS. It just seems like you guys should be able to splice ZH’s post 1998 satellite ocean record onto those land only series and end up with a product that is “way much better”. But maybe not as the land only could be the really bad part and the UAH/RSS ocean somewhat reasonable.
The HadISST2 looks good.
Christy and Spencer UAH has a lower rate of increase(which some find “interesting), and Dr Christy in congressional testimony touted the difference between his measurements and model temperatures(discussed at http://www.realclimate.org/index.php/archives/2016/05/comparing-models-to-the-satellite-datasets/) . He kinda glossed over the issue that his measurement is TMT, Temperature of the Middle Troposphere, not surface temperatures, and the impacts of this. Indeed, comparing UAH v6 versus GISS shows less warming in the middle troposhere than at the surface
(http://www.woodfortrees.org/plot/uah6/mean:12/plot/uah6/trend/plot/uah6/from:1998/trend/plot/uah6/last:100/trend/plot/gistemp/from:1950/mean:12/offset:-0.3/plot/esrl-co2/normalise). UAHv6 does show warming since 1998, contrary to the claim that “The satellites confirm the global warming pause continues now for over18 years now.”
Whatever; this likely means that inaccurate cloud parameterizations in the models causes models to overestimate TLT temperatures due to changes in the lapse rate. My SWAG is that higher absolute humidity but similar relative humidities cause higher growth rates of cloud droplets above the dew point; this causes fatter droplets that more efficiently move water to lower levels in the atmosphere. A supporting observation is that the temperature at the top of the troposphere is cooler in the tropics, where the surface is warmer and more humid.
http://weather.uwyo.edu/cgi-bin/sounding?region=samer&TYPE=GIF%3ASTUVE10&YEAR=2017&MONTH=01&FROM=0712&TO=0712&STNM=04360
http://weather.uwyo.edu/cgi-bin/sounding?region=samer&TYPE=GIF%3ASTUVE10&YEAR=2017&MONTH=01&FROM=0712&TO=0712&STNM=82281
When you jack up the atmosphere GHG’s and put a moister, warmer layer near the surface, it pushes the bottom of the curve at the surface to the right; moisture/cloud/lapse rate dynamics pushes the tom pf the lapse rate curves to the left, cooling the stratosphere and the top of the troposphere.
Since the lapse rate feedback is a NEGATIVE feedback on sensitivity, what Dr Christy has demonstrated is that climate models have too high (negative) lapse rate feedback, and must therefore underestimate climate sensitivity. I predict that somewhere around UAH v 9 or 10, Dr Christy will have conclusively eliminated the negative lapse rate feedback, driving the final nail in the coffin of climate sensitivity being less than 3 degrees C (to coin a phrase &:>)
“what Dr Christy has demonstrated is that climate models have too high (negative) lapse rate feedback, and must therefore underestimate climate sensitivity. ”
I don’t agree. Soden & Held (2006) and others showed that lapse rate and water vapour feedbacks are anticorrelated in climate models. More warming high up means more vapour high up too.
It’s possible that something else is going on beyond what the models simulate, but it’s possible to have lower lapse rate feedback and the same sensitivity.
It says this Santer paper was published on January 1, 2017:
Comparing Tropospheric Warming in Climate Models and Satellite Data
… When the impact of lower-stratospheric cooling on TMT is accounted for, and when the most recent versions of satellite datasets are used, the previously claimed ratio of three between simulated and observed near-global TMT trends is reduced to approximately 1.7. Next, the validity of the statement that satellite data show no significant tropospheric warming over the last 18 years is assessed. This claim is not supported by the current analysis: in five out of six corrected satellite TMT records, significant global-scale tropospheric warming has occurred within the last 18 years…
I call them “climate science deniers”–that gets past the Rubios (“climate is always changing”).
As far as I’m concerned (and I know many others may not agree), the science is telling us urgent action is necessary, and anyone who is working to confuse or cloud that message (e.g., say, Roy Spencer or Bjorn Lomborg or Matt Ridley) is a climate science denier. MHO.
I thought Karl 2015 said that it was the buoys that had a cooling bias and the ships had it right.
Now, this paper says the buoys were good all along and so were the satellites that were dropped a few years ago.
So can we go back to the earlier versions of the SST record then if the buoys and satellites reflect the proper record.
Karl et al 2015 did not say “the buoys had a cooling bias” and that “the ships had it right”.
Moreover, ERSST v 4.0 (used in Karl et al) does not include satellite measurements because they do not add information and are problematic prior to 1985. Hausfather et al only used satellite data post-1985, as far as I can see.
You might want to reconsider the sources for your talking points.
PaulW, the new paper shows that the latest version of ERSST4 agrees best with satellites, buoys-only and Argo. The new version is the best, the old version was drifting away from reality (assuming that lots of independent things agreeing = most likely closest to reality).
Your explanation of ERSST4 is very different from how it’s described in its paper, Huang et al. (2015, doi: 10.1175/JCLI-D-14-00006.1).
Equations (2) and (3) and the following paragraph explain how more weight is given to buoys rather than ships. Maybe that’s the opposite of what you read, but lots of the critics say the opposite of what was actually in the paper.
Maybe you’re talking about the 0.12 C adjustment from the bottom of Table 1? Since this is done before the anomalies are calculated it makes zero difference to trends or anomalies. You can do any addition or subtraction you want, so long as the sum of (T from buoy) + (buoy adjustment) – (T from ships) – (ship adjustment) = 0.12 C. That’s because buoys tend to report a temperature about 0.12 C cooler than ships do when they’re measuring the same location.
Some bloggers made a lot of noise about this, even though it makes zero difference to the final data. Sure, they sounded super confident when they were having a go at the scientists, but it seemed to me like they didn’t get their heads round how the maths works.
PaulW,
The thing is that each measurement tool has its own issues (depth of measurement, noise, sources of heat–e.g. engines…). Since we are mainly interested in deltas, these aren’t as important as long as the dataset looks at only one measurement tool. The issue is when the mix of measurements changes over time. As always, the difficulties have to do with splicing across changes.
Has there ever been any further discussion on the choice of 0.12C offset and how that was improperly an unweighted algebraic average of differences in the different ocean basins? Ross McKitrick had pointed this out a while ago when I wrote on it for the Michigan SSA way back when, though according to some rudimentary follow-up I did on that particular complaint if a rough area-weighting was applied it would result in a greater difference than 0.12C (and so more warming). I’m not surprised that McKitrick had stopped short of attempting to figure out the more correct number he thought it should be (especially after misdirecting his readers about the distribution of the differences and the distribution of the *mean* of the differences), but all the same I thought it was an interesting point and don’t know if anything had become of it in the literature.
(If I am not mistaken, the particular averaging isn’t found in the Karl et al. paper but in Huang et al., whose work was what Karl et al. was chiefly based on.)
One criticism was that the result was skewed by an injudicious choice of start and end point in relation to the El nine la nine events… Is there anything to this?
[Response: It’s total bullshit. For one thing, the time span studied ends with 2015, and much of the el Nino impact was in 2016. For another, it starts with the 1997-1998 el Nino. Add to that the fact that they’ve done a fine job accounting for autocorrelation in their uncertainty estimates.
The main thing that makes such criticisms bullshit rather than mere error, is that they come from the same people who for years have pushed the “no warming since” meme without mentioning (let alone accounting for) that 1997-1998 el Nino. When it comes to el Nino, they want to have their cake and eat it too.]
That should read El nino la nina
Or even El Niño and La Niña, but that get to be a lot of work if you use them much.
Tamino: You of all people should know the denier truism that el Nino’s only count if they appear on the right side of the graph and never if they appear on the left side.
Reblogged this on Greenpress and commented:
They were right all along, but nobody “wins” with rapid climate change – awareness is just step one. We need action, urgently.