Anthony Watts' #AGU15 poster on US temperature trends

Credit: NOAA

Anthony Watts has a poster at this year's AGU Fall Meeting. He is very proud of the effort he made (his WUWT article is archived here). He is also looking over his shoulder, fearful of a conspiracy to stop him from publishing. Actually, Anthony's main contribution seems to have been organising other people to do the work, which is a talent well worth having.

Anthony has made his poster available for downloading from his own website, though it's not available on the AGU15 website (yet). His press release is on the AGU website. The poster has some gaps that could be important. But first, what does it show.

Watts' US surface station temperature trends (revised)

According to the poster, the researchers took the following steps with US temperature records dating between 1979 and 2008. First they removed all weather stations that had been moved or had had a change in time of observation. They classified the remainder using one (not all) of the criteria set out in Leroy (2010), specifically "proximity to artificial surfaces, buildings, and other such objects with unnatural thermal mass".

The poster doesn't indicate how many stations were in each class. It did indicate that overall, the 410 weather stations (out of 1218) were "well-distributed", but it doesn't give any other information about the distribution.

The main finding that the researchers are reporting is that weather stations that are not close to heat sources or heat sinks show less of a warming trend than those that are. In the words in the poster:

The United States temperature trends estimated from the relatively few stations in the classes with minimal artificial impact are found to be collectively about 2/3 as large as US trends estimated in the classes with greater expected artificial impact. The trend differences are largest for minimum temperatures and are statistically significant even at the regional scale and across different types of instrumentation and degrees of urbanization. The homogeneity adjustments applied by the National Centers for Environmental Information (formerly the National Climatic Data Center) greatly reduce those differences but produce trends that are more consistent with the stations with greater expected artificial impact. Trend differences are not found during the 1999-2008 sub-period of relatively stable temperatures, suggesting that the observed differences are caused by a physical mechanism that is directly or indirectly caused by changing temperatures. 

The poster has the following chart, which I've annotated to show the years of divergence. As always, click to enlarge.

The three data series are described as:

• Class 1\2 (Compliant)
• Class 3\4\5 (Non Compliant)
• Official, Adjusted NOAA

Some things you'll probably notice in the above chart is that:

• All three series were aligned to the same point at the start. They are all expressed as anomalies from the same zero point in 1979. There was no attempt to plot them from an averaged baseline, such as a mean of 1979-2008 for each. Although that won't affect the trend of any of them, it does make a difference to how they appear relative to each other.
• The largest difference between "Class 1\2 (Compliant" and "Official Adjusted NOAA" is in 2002, which I measured to be 0.035 °C.
• There is little difference between the three data groups for the first decade or so, up until around 1992.
• There is an extended period for which the group the "Class 1-2 Compliant" dropped below the other two groups. This lasted from 1992 to 2008, with the difference declining in the last couple of years of the period. 
• The description of the three groups are value-laden, with words such as the redundant use of "Compliant" and "Non Compliant"; and the words "Official, Adjusted" when referring to the NOAA record, this latter having special meaning for WUWT conspiracy theorists.

There is no indication in the poster of how many weather stations are in the "Class 1\2" or "Class 3\4\5" categories, nor of their geographic dispersion. Grouped together there are 410 weather stations in these two categories, which the authors say is a subset of the 1,218 in the "Official, Adjusted NOAA" category. (See below for more on this from Evan Jones.)

The other point is that it appears that the researchers didn't categorise the weather stations strictly according to Leroy (2010). As far as I can tell, they only categorised them on the proximity of "heat sources or reflective surfaces (buildings, concrete surfaces, car parks etc)".  There was no indication that the researchers attempted to group them according to the other criteria described in Leroy (2010), which were the slope of the land, the vegetative cover, proximity to water, and shade.

The poster shows maps with the 1979-2010 linear trends for different areas in the USA but again, doesn't indicate the number of weather stations in each region on which the trends are based. (Anthony promised that when/if his paper is published, he'll provide all the data.)

The Watts et al hypothesis

The poster sets out a hypothesis, which is that heat sinks cause an artificially warmer trend over a long period of warming, and an artificially cooler trend over a period of cooling. The mechanism for how that could occur isn't explained. From the poster:

It is well established that heat sinks have an effect on temperature offset. But, we contend the effect manifests itself into temperature trends, as well, in contradiction to the conclusions of Menne et al. (2009, 2010). The overall warming effect of a heat sink on a nearby sensor is greater at the end of a warming phase than at the start of it. Therefore, the trend will be spuriously exaggerated by warming over the 30-year study period (Figure 4).

Conversely, the effect of a heat sink is less at the end than at the beginning of an overall cooling phase. Therefore, the cooling phase is exaggerated in the reverse manner as during a warming phase. This explains why warming is exaggerated from 1979 - 2008 and from 1979 - 1998, when overall warming was reported. Likewise, It also explains why the cooling from 1999 - 2008 is exaggerated. We contend that heat sinks will amplify the trends.

Furthermore, if there is no trend in either direction (as during the last decade-plus), there will be no divergence. This explains why there has been no overall divergence between USHCN and CRN since CRN was not activated until 2005, and the CONUS Tmean trend has been flat.

The Watts et al hypothesis doesn't seem to be supported

If you look at the chart (which apart from some regional trends on maps, is all the data there is at present), it doesn't appear to support the hypothesis. The authors didn't describe the physical mechanism that would explain why an established heat sink would store heat at a rate greater than the temperature increase of the surrounding area over years and decades.

It's expected there would be differences in absolute temperature as a result of artificial structures, and in diurnal variation (maxima and minima), depending on the proximity of buildings and asphalt etc.

A difference in the annual trend would be expected if the nearby buildings are changed over time, as happens when towns are rapidly growing. When there is a structure close to a weather station it will affect the temperature. The structure will absorb heat during the day and release it at night and that will moderate the air temperature, for example, it can raise the minimum. Structures can also affect air flow and wind. So the temperature (and humidity) recorded will be different to what it would have been had a weather station been sited further from buildings.

However if the surrounding structures are not changed, then it's not clear to me how the average annual temperature trend would rise more quickly than the surrounds for weather stations near building structures, or drop more quickly over a period of years when the surrounds are in a cooling trend. (I might not be thinking this through properly, so if you think this is what is expected, please describe how it works in the comments - for example, by affecting the minimum, hence the average of Tmin and Tmax. If you're interested, here's a brochure that describes how concrete slabs work, which is probably fairly similar to the effect of concrete buildings close to weather stations. Different materials, orientations etc will have different thermal responses. The response at any site probably also depends on the diurnal amplitude in temperature.)

Now the WUWT article indicates there's a paper that's been submitted for publication. Anthony doesn't say that it's been accepted. So we'll have to wait for data that will allow an analysis, and wait to see if there is more analysis behind their "key findings", which might support their hypothesis. For now, I can't see it, and nor have they provided a physical explanation other than the assertion that "The overall warming effect of a heat sink on a nearby sensor is greater at the end of a warming phase than at the start of it." In the poster is the claim:

...warming is exaggerated from 1979 - 2008 and from 1979 - 1998, when overall warming was reported. Likewise, It also explains why the cooling from 1999 - 2008 is exaggerated. We contend that heat sinks will amplify the trends.

Problem is that the period from 1999 to 2008 doesn't appear to show any consistent difference between the categories from beginning to end. The difference between 1979 and 1998 began in 2002. It wasn't a gradual dispersion. It was more like a marked shift. If the hypotheses held, then the temperatures would have diverged when the temperature was rising. Instead, the temperatures began to diverge in 1993 when temperatures dipped, and stayed diverged over the warmer period from 1992 to around 2006, when they began to converge again.

I'm not disputing the authors when they claimed statistical significance for the overall difference between their "Classes 1\2" and "Classes 3\4\5". But the paper goes beyond that and talks about this cooling and heating effect, which as far as I can tell isn't supported by any analysis.

The discussion with Evan Jones at HotWhopper

The poster looks to be along the lines of the discussion that Evan Jones had with Victor Venema, here at HotWhopper. Anthony gives a backhanded thanks to Dr Venema for his contribution, writing on his blog:

This work is a continuation of the surface stations project started in 2007, our first publication, Fall et al. in 2010, and our early draft paper in 2012. Putting out that draft paper in 2012 provided us with valuable feedback from critics, and we’ve incorporated that into the effort. Even input from openly hostile professional people, such as Victor Venema, have been highly useful, and I thank him for it.

Anthony couldn't bring himself to mention HotWhopper, but these are the discussions to which I believe he's referring:

• HotWhopper Competition: Best Name for a Denier Lobby Group (in 25 words or less) - the discussion began in the comments
• Heat sinking, temperatures rising in the US of A - this article was specifically to allow continued discussion by Evan Jones of the US surface station analysis.

Evan Jones made the following points in those discussions:

• Only partial Leroy criteria: They were only categorising the weather stations on "what Leroy (2010) considers to be a heat source". They did not factor in the other criteria in Leroy (2010), such as shade, vegetation, or ground angle. Evan explained "We do not even know how the stations would be rated in these other categories". (link)
• Very few stations would meet all Leroy criteria: If they had applied strictly the Leroy criteria for Class 1 and 2, there might have been only a dozen or so weather stations in their Classes 1 and 2 (link)
• Still only around 80 stations in Watts' "Class 1\2": Only about 82 weather stations, or one in five of the 410 stations in the WUWT research are included as "Class 1\2" (link) Update: In a late comment at WUWT, Evan Jones has said there were 92 "Class 1\2" and 318 "Class 3\4\5" - a bit more than his April 2014 estimate (link) - Sou 9:56 pm AEDT 18 Dec 15).
• No indication of geographic spread: I couldn't find any comment from Evan that indicated how well those 80 or so weather stations were dispersed across the contiguous USA.
• Evan saw what he wanted to see in regard to the difference in warming, writing "It was the '90s that saw a strong warming trend. That's when your difference pop out. If you know what to look for, it sticks out of the data a mile."

On that last point, here again is the chart on which I've marked the 1990's for which Evan maintains the trends are different. Also marked is the period from 1999 to 2008, when the poster maintains the "the cooling from 1999 - 2008 is exaggerated". Neither is the case. 

The difference in annual temperature anomaly began around 1992 (with cooler temperatures) and continued to the end of the 1990s, being much the same difference from the mid 1990s to the late 1990s. As for the "cooling trend" being exaggerated, in fact the difference in the anomalies in 1998, 2006 and 2007 (and 2008) was less than the difference in the some of anomalies in the middle of that period [amended - Sou].


What conclusions would you draw? I can't draw any conclusions without more data. I would have some questions, however.

• What would the data have shown if they had grouped the classifications in the manner that Leroy (2010) grouped them, with Classes 1-3 together and Classes 4 and 5 separately?
• What difference would it have made if they'd been able to use all the criteria in Leroy to classify the weather stations?
• How many weather stations are included in Watts' Class 1\2?
• How representative are the Watts Class 1\2 weather stations? Are they spread across the USA in the same proportion as Watts' "Class 3\4\5" weather stations?
• What is the basis for the claim that "the cooling from 1999 - 2008 is exaggerated", which isn't what the chart shows?
• In relation to the statement "warming is exaggerated from 1979 - 2008 and from 1979 - 1998", were any other possible reasons explored to explain the step drop in Watts' "Class 1\2" observed in the early 1990s?
• By what physical mechanism would "warming effect of a heat sink on a nearby sensor (be) greater at the end of a warming phase than at the start of it" and the effect of a heat sink be "less at the end than at the beginning of an overall cooling phase". What data is there that would support those hypotheses?
• What happened to the data for 1988? The data point appears to be missing from the chart.

Leroy (2010) land weather station siting classifications

For the sake of completeness, I Googled and found the document Anthony referenced as Leroy (2010) (he neglected to provide a link). It is a paper submitted to the JMA/WMO Workshop on Quality Management in Surface, Climate and Upper-air Observations in RA II (Asia) held in Tokyo, Japan on 27-30 July 2010. The paper was submitted by Michael Leroy of Meteo-France and is called "Siting Classification for Surface Observing Stations on Land". 

I do not know if the classifications discussed in the Leroy workshop paper were adopted by the WMO.  The current WMO Guide to Meteorological Instruments and Methods of Observation is the 2008 edition, updated in 2010, with a corrigenda table dated Sept. 2012. Some of the sections of the guide were updated in 2010, but it's not clear to me if the classification system proposed by Michael Leroy has been adopted. (Victor Venema - can you help?)

The paper makes clear that different classifications can apply depending on what is being monitored:

A site as a whole has no single classification number. Each parameter being measured at a site has its own class, and is sometimes different from others. If a global classification of a site is required, the maximum value of the parameters’ classes can be used.

The classifications described in the main paper relate to precipitation. An annex to the paper describes site classifications for air temperature and humidity, surface wind, global and diffuse solar radiation, direct radiation and sunshine duration. As I understand it, Evan Jones went through each of the 410 weather stations and sorted them according to the classification scheme proposed for air temperature and humidity - but only on the "proximity to buildings etc" criteria. Not the other criteria. Leroy observed that in regard to temperature and humidity, the "main discrepancies are caused by unnatural surfaces and shading". (Shading wasn't included in the Watts project as far as I can tell, only the unnatural surfaces.)

The five classifications proposed in the annex to Leroy (2010) for temperature and humidity observations are summarised as follows, showing the proposed measurement uncertainty:

• Class 1 are the best, having a slope less than 1/3 (19°), natural vegetation less than 10 cm high, unshaded and at least 100 m from buildings or bodies of water.
• Class 2 are next best, having a slope less than 1/3 (19°), natural vegetation less than 10 cm high, unshaded and at least 30 m from buildings or bodies of water.
• Class 3 (additional estimated uncertainty added by siting up to 1 °C) - natural vegetation less than 25 cm high, unshaded and at least 10 m from buildings or bodies of water. 
• Class 4 (additional estimated uncertainty added by siting up to 2 °C) - Close artificial heat sources and reflective surfaces (buildings, concrete surfaces, car parks etc.) or expanse of water (unless significant of the region, occupying: Less than 50% of the surface within a circular area of 10 m around the screen; o Less than 30% of the surface within a circular area of 3 m around the screen.
• Class 5 (additional estimated uncertainty added by siting up to 5 °C) are the sites not in Class 1 to 4 above.

In addition to the above five classifications, Leroy (2010) discussed an additional classification scheme, which "was not considered as enough mature to be proposed as a WMO standard". It was described as an organisational classification:

In order to document the performance characteristics of the various surface observing networks used, Météo-France defined another classification, called "maintained performance classification", including the uncertainty of the instrument and the periodicity and the procedures of preventive maintenance and calibration.

I only mention it because of the diagram below, in which Leroy shows the classification of the Radome network, which is the the main surface-observing network of Météo-France. According to Leroy (2010), 85% of the weather stations in that network are rated as Class 1 to 3, 5% as Class 4 and 1% as Class 5 - using the temperature and humidity classification.

As you can see from the above diagram, Leroy (2010) grouped his Classes 1 to 5 differently to the Watts poster. Leroy grouped Classes 1, 2 and 3 together, and Classes 4 and 5 separately. The authors of the poster grouped the classifications differently. Instead of reporting Classes 1 to 3 together, Anthony's team grouped the 410 stations with Classes 1 and 2 together, and Classes 3 to 5 together.

Conspiracy ideation from Anthony Watts

Anthony couldn't resist some conspiracy ideation in his WUWT article. There were hints of it in his poster as well, as mentioned above. He described Victor Venema as an "openly hostile" professional, though there's no sign of open hostility in his discussion with Evan Jones that Anthony referred to. In another spot he wrote:

We are submitting this to publication in a well respected journal. No, I won’t say which one because we don’t need any attempts at journal gate-keeping like we saw in the Climategate emails....

...When the journal article publishes, we’ll make all of the data, code, and methods available so that the study is entirely replicable. We feel this is very important, even if it allows unscrupulous types to launch “creative”  attacks via journal publications, blog posts, and comments. When the data and paper is available, we’ll welcome real and well-founded criticism.

Anthony also said he doesn't want any comments of the following kind:

This is a technical thread, so those who simply want to scream vitriol about deniers, Koch Brothers, and Exxon aren’t welcome here. Same for people that just want to hurl accusations without backing them up (especially those using fake names/emails, we have a few). Moderators should use pro-active discretion to weed out such detritus. Genuine comments and/or questions are welcome.

From the WUWT comments

Here are some comments from the WUWT detritus. The kind that is welcomed at WUWT.

ntesdorf  might be surprised to learn that the data Anthony's dogsbodies analysed is a subset of the same data analysed by the NOAA.

December 17, 2015 at 1:39 pm
This is a colossal effort and achievement by Anthony Watts and deserves the widest study and acknowledgement. I hope that there are no mis-guided efforts to block its publication. The benefits of this study are self-evident. Reliable data is the basis of all science, and reliable data has been missing from the Climate debate for a long while.

Aphan isn't quite correct, but it's nice to know there are HotWhopper fans at WUWT. (The discussions with Evan Jones, here and here, took place in April 2014:

December 17, 2015 at 3:28 pm
Sou (as in Soooooie!) over at Snot Plopper been trying to debunk this paper since she saw Anthony’s tweet in October! I’m so excited to read it Anthony! 

Jeff Id congratulates Anthony for his effort in getting Evan Jones, and John Nielsen-Gammon and his students, to pull together the analysis and the poster.

December 17, 2015 at 2:05 pm
You should be very proud of the time and effort put into this.
My biggest congratulations. Very impressive sir! Also to the coauthors and those who put so much time into supporting this effort. 

Peter Sable wants to extrapolate the trend from Anthony's 80 or so weather stations in the USA to the entire land and ocean surface:

December 17, 2015 at 2:48 pm (excerpt)
Nice work, Anthony. You set a standard to which others should attempt to emulate.

If I read the numbers correctly, it looks as if we should take any warming trend we see from adjusted temperature record trends and multiply it by ~2/3. e.g. for HadCrut, GISS, BEST, etc. 

Ardy was the only person I saw who noticed that even the Watts' "Class 1\2" weather stations showed a strong warming trend since 1979:

December 17, 2015 at 3:05 pm
Am I missing something. IF the gold standard sites are warming at 0.204 per decade. If that continues we get to 2c per century. Now I understand the variability but I am not sure this report lets the heaters off the hook. They may well say ‘so what it is still 2c per century.’ 

The remainder were much the same. Heartfelt gratitude and congratulations all around. I didn't see anyone asking any of the questions I've posed. There weren't too many commenters who showed much sign of reading the poster.

I don't begrudge them their celebrations. It's nice to see that they've an example where a climate conspiracy promoter is able to show a poster at AGU. (He wasn't the only one, by the way. There was also Peter Ward, who thinks global warming is caused by the hole in the ozone layer, or something.)

Related articles

After posting, I found that Victor Venema has already written an article about the work, which you can read on his blog.

The two main discussions with Evan Jones in April 2014 at HotWhopper are here and here.