McKitrick & Michaels Were Right: More Evidence of Spurious Warming in the IPCC Surface Temperature Dataset
Guest post by Roy W. Spencer, Ph. D.
The supposed gold standard in surface temperature data is that produced by Univ. of East Anglia, the so-called CRUTem3 dataset. There has always been a lingering suspicion among skeptics that some portion of this IPCC official temperature record contains some level of residual spurious warming due to the urban heat island effect. Several published papers over the years have supported that suspicion.
The Urban Heat Island (UHI) effect is familiar to most people: towns and cities are typically warmer than surrounding rural areas due to the replacement of natural vegetation with manmade structures. If that effect increases over time at thermometer sites, there will be a spurious warming component to regional or global temperature trends computed from the data.
Here I will show based upon unadjusted International Surface Hourly (ISH) data archived at NCDC that the warming trend over the Northern Hemisphere, where virtually all of the thermometer data exist, is a function of population density at the thermometer site.
Depending upon how low in population density one extends the results, the level of spurious warming in the CRUTem3 dataset ranges from 14% to 30% when 3 population density classes are considered, and even 60% with 5 population classes.
DATA & METHOD
Analysis of the raw station data is not for the faint of heart. For the period 1973 through 2011, there are hundreds of thousands of data files in the NCDC ISH archive, each file representing one station of data from one year. The data volume is many gigabytes.
From these files I computed daily average temperatures at each station which had records extending back at least to 1973, the year of a large increase in the number of global stations included in the ISH database. The daily average temperature was computed from the 4 standard synoptic times (00, 06, 12, 18 UTC) which are the most commonly reported times from stations around the world.
At least 20 days of complete data were required for a monthly average temperature to be computed, and the 1973-2011 period of record had to be at least 80% complete for a station to be included in the analysis.
I then stratified the stations based upon the 2000 census population density at each station; the population dataset I used has a spatial resolution of 1 km.
I then accepted all 5×5 deg lat/lon grid boxes (the same ones that Phil Jones uses in constructing the CRUTem3 dataset) which had all of the following present: a CRUTem3 temperature, and at least 1 station from each of 3 population classes, with class boundaries at 0, 15, 500, and 30,000 persons per sq. km.
By requiring all three population classes to be present for grids to be used in the analysis, we get the best ‘apples-to-apples’ comparison between stations of different population densities. The downside is that there is less geographic coverage than that provided in the Jones dataset, since relatively few grids meet such a requirement.
But the intent here is not to get a best estimate of temperature trends for the 1973-2011 period; it is instead to get an estimate of the level of spurious warming in the CRUTem3 dataset. The resulting number of 5×5 deg grids with stations from all three population classes averaged around 100 per month during 1973 through 2011.
RESULTS
The results are shown in the following figure, which indicates that the lower the population density surrounding a temperature station, the lower the average linear warming trend for the 1973-2011 period. Note that the CRUTem3 trend is a little higher than simply averaging all of the accepted ISH stations together, but not as high as when only the highest population stations were used.
The CRUTem3 and lowest population density temperature anomaly time series which go into computing these trends are shown in the next plot, along with polynomial fits to the data:
Again, the above plot is not meant to necessarily be estimates for the entire Northern Hemispheric land area, but only those 5×5 deg grids where there are temperature reporting stations representing all three population classes.
The difference between these two temperature traces is shown next:
From this last plot, we see in recent years there appears to be a growing bias in the CRUTem3 temperatures versus the temperatures from the lowest population class.
The CRUTem3 temperature linear trend is about 15% warmer than the lowest population class temperature trend. But if we extrapolate the results in the first plot above to near-zero population density (0.1 persons per sq. km), we get a 30% overestimate of temperature trends from CRUTem3.
If I increase the number of population classes from 3 to 5, the CRUTem3 trend is overestimated by 60% at 0.1 persons per sq. km, but the number of grids which have stations representing all 5 population classes averages only 10 to 15 per month, instead of 100 per month. So, I suspect those results are less reliable.
I find the above results to be quite compelling evidence for what Anthony Watts, Pat Michaels, Ross McKitrick, et al., have been emphasizing for years: that poor thermometer siting has likely led to spurious warming trends, which has then inflated the official IPCC estimates of warming. These results are roughly consistent with the McKitrick and Michaels (2007) study which suggested as much as 50% of the reported surface warming since 1980 could be spurious.
I would love to write this work up and submit it for publication, but I am growing weary of the IPCC gatekeepers killing my papers; the more damaging any conclusions are to the IPCC narrative, the less likely they are to be published. That’s the world we live in.
UPDATE: I’ve appended the results for the U.S. only, which shows evidence that CRUTem3 has overstated U.S. warming trends during 1973-2011 by at least 50%.
I’ve computed results for just the United States, and these are a little more specific. The ISH stations were once again stratified by local population density. Temperature trends were computed for each station individually, and the upper and lower 5% trend ‘outliers’ in each of the 3 population classes were excluded from the analysis. For each population class, I also computed the ‘official’ CRUTem3 trends, and averaged those just like I averaged the ISH station data.
The results in the following plot show that for the 87 stations in the lowest population class, the average CRUTem3 temperature trend was 57% warmer than the trend computed from the ISH station data.
These are apples-to-apples comparisons…for each station trend included in the averaging for each population class, a corresponding, nearest-neighbor CRUTem3 trend was also included in the averaging for that population class.
How can one explain such results, other than to conclude that there is spurious warming in the CRUTem3 dataset? I already see in the comments, below, that there are a few attempts to divert attention from this central issue. I would like to hear an alternative explanation for such results.
Discover more from Watts Up With That?
Subscribe to get the latest posts sent to your email.
I am a little confused by the analysis. I am with Hector (3/30, 12:46PM) here. If the UHI effects the trend then you have to identify some mechanism or proxy for a mechanism that leads to an impact on the local temperature “trend”. Population growth is a proxy, energy consumption might be another. What the data suggests to me is that the rate of growth in population has been greater in already dense areas than in less dense areas – essentially the continued urbanization of the population – hence a higher trend. However, surely, as Hector suggests, you also have to factor in actual changes in population over the 1973-2010 period. You can see the UHI effect at work in small settlements like Cambridge Bay in the Arctic as the number of dwellings increased. Don’t you have to incorporate a similar mechanism to really gauge the strength of the UHI effect? Isn’t that what McKitrick and Michaels did?
Since Roy’s method does not take into account the amount of rural-station warming that is attributable to UHI it would seem to systematically undererstimate the amount of total warming that is attributable to UHI. One would expect the first increments of urbanization to have the strongest warming effects, so just because a station is still rural doesn’t mean it hasn’t been significantly affected by local heat sources, expecially if the thermometers tend to be placed near local outposts.
I am not sure that this is showing the same thing as McKtrick and Michaels. They had a mechanism as to why the rate of change in temperature was different in different areas, i.e., higher rates of growth in economic activities was associated with higher rates of surface temperature increase. Hector @12:46PM correctly points out that the analysis needs to incorporate actual changes in population growth – otherwise no mechanism for impacting the rate of change in the temperature has been proffered – except the implicit notion that population growth is stronger in existing densely populated areas. This makes sense but is not demonstrated and might not be true where there are few controls on urban sprawl and rural development.
Jones was involved originally in a UHI study of Chinese stations that was subject to a complaint against one of the authors (Wang IIRC then at SUNYA) on the basis that the statement that there “few if any” station moves was untrue. It was based on this flawed study that Jones concluded that UHI was not a factor. SUNYA eventuually rejected the complaint but censored the details. After Climategate I (IIRC), it came out that there had been numerous station moves of the “few if any” group, including some that required altitude corrections (also undisclosed)!!.
BEST came up with its own interesting definitions, where they classify stations as “very rural” and “not very rural:”
One tenth of a degree would be about 6 nautical miles, hardly what a disinterested observer would call very rural. IIRC “urban” was defined as more than 50% “built.” So, these “very rural” sites could be as close as only a 5 or 6 minute drive by car from places that were more than 50% built!
Large cities have large heat “bubbles” above them (UHI in 3D) that probably affect temperatures many miles from the city perimeter, depending on local environmental conditions like wind speed and direction. Had they used a stricter definition of, for example, 60 miles from “urban regions”, this might be more believable. However, even this definition is strained when one considers regions like the Washington to New York corridor, LA to San Diego, Miami to Ft. Lauderdale, SE China or mega-cities like Mexico City, Tokyo, São Paulo, etc. where I would submit there probably aren’t any stations withing hundreds of miles that would be both (a) truly “very rural” AND (b) comparable climatologically. This, I would submit, makes the use of “paired” stations to study UHI questionable.
The UHI study in Barrow, AK took a different approach. They set up a grid of identical temperature data loggers to map the temperature field around the town. This permitted the real picture of UHI for this location to be determined and separated from confounding variables like seasons, wind speed and direction and open water leads. Once they were able to map the temperature field, they were able to objectively determine which station (in the first paper) and which average of five stations (in the second paper) best represented the UHI and which station and average of five stations best represented the controls, where no UHI was evident. This took five years, after which they dismantled the rest of the network and kept up just the representative data loggers, due to the trouble and expense of maintaining the network.
I don’t believe any pairing study (other than Barrow), including BEST, has ever objectively determined that particular stations are or are not representative of UHI areas and areas not affected by UHI. Representativeness is simply assumed unquestionably. Given Anthony’s find during the surface station project that something like 80% of stations have significant siting issues, the representativeness assumption should not be accepted without support.
Given all of the above, I would submit that it is Dr. Spencer’s methodology (building on MM2007) or something similar that should be considered as the best estimate of UHI and not the other way around.
Hugh Pepper says:
March 30, 2012 at 12:46 pm
“Were they too part of the great conspiracy?”
That makes YOU officially a conspiracy theorist, Hugh.
Even in the boonies many stations are near buildings, roads, dissipating electrical equipment, etc. I think most surface data are contaminated by anthropogenic waste heat and albedo mods. We should stop calling it UHI, the problem is far more pervasive and affects many “rural” locations albeit to a lesser degree than in more developed ones.
Roy
Very nice post. The problems with siting, inconsistencies with readings, instrumental problems etc etc was known of over 100 years ago. I wrote about them here;
http://wattsupwiththat.com/2011/05/23/little-ice-age-thermometers-%E2%80%93-history-and-reliability-2/
The Met office adjust the British record by .2C to account for UHI-we suspect it should be a lot more as the UK has been described as one large Urban heat island (it is relatively small with a considerable pulation density)
All in all there are so many potential flaws and blind spots in the temperature record that we shouldnt rely on it to make such far reaching ppolicy. There is an additional complications with the number of regions that are cooling, these are overwhelmned by the warming signal-likely UHI-but which disguises the complexity of the climate.
tonyb
The Other Tex said:
“Correct me if I am wrong Dr. Spencer, but my impression was that you were not comparing your dataset to the overall HadCruT trend, but to the trend in their data on a grid square basis. So if you were only comparing the HadCruT trend for that grid to the trend at the stations in that grid, it doesn’t matter if you have SH grids involved or not, right? What you have established is that UHI does in fact affect the calculated trends. A much more detailed analysis would be necessary to discern the degree to which the UHI contamination affects the entire global dataset; but what this analysis has done is significant, because it has shown the UHI contamination to be real and significant.”
Yes, thank you for noticing, Tex. You apparently read my entire post. 🙂
-Roy
Following up on a few comments about the strength of the warming bias increasing (not decreasing) with average population density, I agree this is opposite of what I expected. I don’t have an explanation for it, but I haven’t taken the time to think about it, either.
@Hugh Pepper
This question was dealt with by the BEST study. They concluded that, since only 0.5% of the world is urbanized, even a 2 degree rise in urban temperature would contribute negligibly to the global average. Were they too part of the great conspiracy?
====================
So in your opinion assumptions made by BEST are to be preferred, and careful analysis rejected. And on top of that, throw in suggestions of conspiracy theory…
Perhaps this sort of reasoning is actually a product of working backwards from preferred conclusions?
Steven Mosher said @ur momisugly March 30, 2012 at 1:40 pm
What is wrong with treating the Southern and Northern Hemispheres as separate entities? Not only is UHI different in the Southern hemisphere, the temperature trend over the last 100 years is also quite different. The only reason I can see for averaging the temperatures of the Southern and Northern hemispheres is to disguise the fact that there is no global trend. Southern hemisphere temperatures have been almost static for a century!
Peter Miller said very nicely,
“A classic goofy, alarmist half truth: Obviously only a tiny part of the world is urbanised, but equally obviously most of the world’s temperature monitoring stations are located in that tiny part. These stations then bias the results from the other circa 99.5% of the world.”
But, unfortunately, people really don’t know that. I lived in big cities most of my life and held that false impression myself. When I was twenty-five I took a job where a flew quite a bit and I was truly shocked at the amount of rural land. Now I live in the country and know first hand. CIty scientists see the earth as an endless city and need to get out of town. This urban heat island stuff is huge. Y’all need to get out in the country, and freeze your butt a little.
Another, obvious thing, something I haven’t seen mentioned here, though someone must have brought it up: That cities contains tens of thousands of windblocks called “buildings”. Wouldn’t blocking this lower wind slow down the thermostatic cooling effect that Willis E. describes so well, besides obviously giving the air more time to heat up?
Can someone explain to me the difference between gridding data and inventing data? I’m not being flippant either. I’m genuinely interested in the answer.
RockyRoad said @ur momisugly March 30, 2012 at 12:49 pm
Let’s meet in Bondi and discuss the relative merits of moving closer to the equator versus living in a city densely populated by pretty young girls in bikinis. I’ll tell the missus I want to spend some time with my older son who lives there. The coffee is first class; the food at Bombay to Bondi cheap and delicious 🙂
Will Nitschke says to Hugh Pepper:
“Perhaps this sort of reasoning is actually a product of working backwards from preferred conclusions?”
Of course it is.
Thank you for all your hard work Dr. Spencer. It is nice to see a rigorous treatment of the subject of UHI in the CRUTem3 data set. To bad this very important work will never make it into publication.
It is a shame that science has sunk so low that journals will no longer publish anything but pal reviewed “propaganda” It seems to me we are seeing more and more real science “published” and “peer-reviewed” at sites like WUWT and Climateaudit. It maybe the saving of science.
News papers are losing market share to the blogosphere for a similar reason. People want to see something besides carefully crafted propaganda. Now the politicians want to start internet censorship. as well.
If I did not believe in censorship of journals and the news media before, the move towards censorship would have convinced me.
“…understand the UHI bias in the complete record…”
Steven: Why? If that’s constructive criticism, I at a loss to see the constructive part. Are you implying that it is not possible to identify a specific urban area and then identify the UHI bias of that area? That one must someohow include Australia, Malaysia, Chile, etc in order to identify or quantify the UHI specific to an area?
To me, one should first prove there is UHI bias in the record and if possible, quantify it explicitly by region. Go build the world afterwards as I think that is how the CAGW alarmists have been masking their machinations through the baffle with BS routine. If you prove UHI for one urban area, the rest are dominoes.
I’d be careful being too loose with Dr. Spencer’s analysis. In CA there is a reason counties have low populations – they tend also to have different climate zones like mountains, deserts, military/government land, uninhabited islands, all offering low quality of life in the urban sense, and unlikely to be well-instrumented for collecting quality data for purposes of climate trending. I would also take note of the activities in CA in the early 1970’s when the BLM ran a lot of people off public lands.
If consideration of this is also part of Dr. Spencer’s analysis then a tip o’ me hat to the good professor for a job well done. If it holds up the results are devastating.
Roy Spencer says:
March 30, 2012 at 4:08 pm
“UHI does in fact affect the calculated trends.”
Just use an algorithm to counteract the UHI.
It’s all the rage in climatology.
Dr. Roy,
IF Mosher reads the whole thing, perhaps he will notice also. Then I am sure he will retract this gratuitous insult:
“So, you need to add in other data sources IF you want to understand the UHI bias in the complete record.”
Just to be clear – AGW is real – people do cause it to get warmer. The big reveal is it only happens where there are lots of people and their stuff piled up in one place.
So this means we can stop talking about CO2 and instead talk about getting off the asphalt during the summer… right?
Well that’s going to save a lot of money. We should have a conference at a luxury hotel in a tropical paradise – or something.
so, dp, now we should prefer urban to rural thermometer data? Those backwards rural types must be under-reporting temperatures just to spite the elitist city folks. 🙂
Actually, I am probably one of the very few here who was certified as an aviation weather observer, I worked at an NWS office taking hourly observations. I know something about the issues involved in temperature measurement, whether liquid-in-glass or electronic.
Tom_R says:
March 30, 2012 at 1:36 pm
Dr. Spencer,
Because UHI warming depends on an increase over time, what your study here has shown is that the UHI warming increases faster in higher population density areas. Doesn’t this disagree with your previous results?
Yeah – I thought the same thing. If I understand your post you seem to be saying that there is no reason that an urban area should be any more affected by an UHI Trend than a rural area. I remember Roy did do a post on WUWT some time ago which graphed the UHI trends v population. The basic message was that an increase in population of say, 1000, in a rural area had a similar effect on trend as an increase in population of say, 100000, in an urban area. This makes sense to me.
I can’t see any reason why urban TRENDS (not temperatures) should be any different to rural TRENDS (not temperatures) unless ALL urban populations were growing while ALL rural populations remained static.
Dr Spencer:
Thankyou for the valuable and interesting analysis in your article.
In the subsequent comments you say (at March 30, 2012 at 4:15 pm):
“Following up on a few comments about the strength of the warming bias increasing (not decreasing) with average population density, I agree this is opposite of what I expected. I don’t have an explanation for it, but I haven’t taken the time to think about it, either.”
I write to suggest one possible hypothesis to explain it.
In general, measurement sites are near the edges of populated regions (e.g. they are often at airfields).
But
(a) expansion of regions with high population density is mostly achieved by ‘urban sprawl’ (i.e. by spread of urbanisation) and is not mostly achieved by ‘infilling’ the populated area.
while
(b) expansion of regions with low population density is mostly achieved by ‘infilling’ the populated area and is not mostly achieved by ‘urban sprawl’ (i.e. by spread of urbanisation).
If this is true to some degree then the amount of urban development near the edges of populated regions (i.e. where most measurement sites are situated) would be greater for the regions with high population density.
Richard
Oooh I love this stuff, questions, questions. So if NOAA show a “cooling” trend across all their geographical zones but differeing rates for each, do Dr Spencers’ “urban” areas correspond to the NOAA zones that cooled the least and can population numbers of those zones roughly correlate to the implied UHI. Or to put it another way would the whole of the US have cooled the same if there was no urbanisation? Anyone good with numbers?