by Zeke Hausfather , Steven Mosher, Matthew Menne , Claude Williams , and Nick Stokes
[Note: this is an AGU poster displayed at the annual meeting, available here as a PDF. I’ve converted it to plain text and images for your reading pleasure. I’m providing it without comment except to say that Steven Mosher has done a great deal of work in creating a very useful database that better defines rural and urban stations better than the metadata we have available now. – Anthony]
Introduction
Large-scale reconstructions of surface temperature rely on measurements from a global network of instruments. With the exception of remote automated sensors, the locations of the instruments tend to be correlated with inhabited areas. This means that urban
ares [sic] are probably oversampled in surface temperature records relative to the total land surface that is actually urbanized.
It has long been known that urbanized areas tend to have higher temperatures than surrounding less developed (or rural) areas due to the concentration of high thermal mass impermeable surfaces (Oke 1982). This has led to some concern that changes in
urban heat island (UHI) effects due to rapid urbanization in many parts of the world over the past three decades may have been responsible for a portion of the rapid rise in measured global land surface temperatures. This concern is reinforced by lower
observed trends in some interpretations of satellite measurements of lower tropospheric temperature over land areas during the same period (Klotzbach et al 2009).
An analysis of the impact of urbanization on temperature trends faces multiple confounding factors. For example, an instrument originally installed in a city frequently will have warmer absolute temperatures than one in a nearby rural area (especially at night), but will not necessarily show a higher trend over time unless the environs change in such a way that the UHI signal is altered in the vicinity of the instrument. Similarly, microsite characteristics that may be unrelated to the larger urban environment can have
notable effects on temperature trends and act counter to or in concert with the ambient UHI signal.
Moreover, the definition of urban areas is subject to some uncertainty, both in terms of how urban form is characterized and at what distance from built surfaces urban-related effects persist. Published station metadata often includes outdated indications of whether a station is urban or rural, and instrument geolocation data can be imprecise, out of date, or otherwise incorrect.
There is also uncertainty over how much explicit correction is needed for urban warming in global temperature reconstructions, and how well homogenization techniques recently introduced in GHCN-Monthly version 3 both detect and correct for inhomogenities
arising from changes in urban form.
To address these issues and obtain a more accurate estimation of the impact of urbanization on land temperature trends, we examine different urbanity proxies at multiple spatial resolutions and urbanity selection criteria through both simple spatial
weighting and station pairing techniques. This study limits itself to unadjusted average temperature data, though we will examine homogenized data in the future to see how much of the UHI signal is removed.
Methods
We examine GHCN-Daily version 2.80 temperature data rather than the more commonly used GHCN-Monthly data as it contains significantly more stations, particularly during the past thirty years, and allows for separate examination of maximum and minimum
temperatures. A relatively high spatial density of stations is useful to allow sampling into various urban and rural station subsets while minimizing biases due to loss of spatial coverage. After excluding stations that have fewer than 36 months at any time in the
period of record or at least one complete year of data during the 1979 to 2010 period, we are left with 14,789 stations.
A complete set of metadata is calculated for each station using the station location information provided in station inventories and publically available GIS datasets. These datasets include: Distance From Coast (0.1 deg), Hyde 3.1 historical population data (5
arc minute), 2000AD Grump Population density (30 arc seconds), Grump Urban Extent, Land use classes from the Harmonized Land Use inventory (5 arc minutes), radiance calibrated Nightlights (30 arc seconds), ISA- Global Impervious Surfaces (30 arc
seconds), Modis Landcover classes (15 arc seconds), and distance from the closest airport (30 arc seconds). In addition, area statistics at progressive radii are calculated around each putative site location.
Stations are then divided into two classes based on various thresholds for urbanity and two analytical methods are used to estimate the bias in trend due to urbanity: a spatial method and a paired station approach. The spatial averaging method relies on
solving a set of linear equations for the stations in each class. For each group of stations, urban and rural, a time series of average temperature offsets was created by fitting the model:
where T represents the observed temperature for each station, month and year, L is a local average temperature for each station for each month (incorporating seasonal variation) and G is the desired global (or regional) average, varying by year. This is fitted
with a weighting that is inversely proportional to a measure of station density. With a G calculated for both urban and rural, the trends can be compared.
The pairwise method proceeds with the same classification of stations and the following steps are taken. An urban base pair is selected based on the length of record. To qualify as a base urban pair a station must have 30 complete years of data in the 1979-2010 window.
Ten out of 12 months of data are required to count as a complete year. For every urban base station rural pairs are selected based on distance and data overlap. For every urban base station the rural stations are exhaustively searched and all those rural pairs within 500km are assigned to the base station. Since rural stations may have short records the entire rural ensemble is evaluated for data overlap with the urban base pair. 300 months of overlap are required. If the collection of rural stations has less than 300 months of overlap with its urban pair, it is dropped from the analysis. A weighting function is deÞned in the neighborhood of each urban station, which diminishes with distance and is zero outside a certain radius. An average trend is computed for the rural stations within that radius by fitting the model
where t is time in years, and B is the gradient. This trend is then compared with the OLS trend for the central urban station. The differences in the shapes of the distributions of the trends is a function of the number of stations that form the trend estimation.
Urban trends are trends for individual stations, while rural trends are the result of computing a trend for all the rural pairs taken as a complete ensemble.
Discussion
While urban warming is a real phenomenon, it is overweighted in land temperature reconstructions due to the oversampling of urban areas relative to their global land coverage. Rapid urbanization over the past three decades has likely contributed
to a modest warm bias in unhomogenized global land temperature reconstructions, with urban stations warming about ten percent faster than rural stations in the period from 1979 to 2010. Urban stations are warming faster than rural stations on average across all urbanity proxies, cutoffs, and spatial resolutions examined, though the underlying data is noisy and there are many individual cases of urban cooling. Our estimate for the bias due to UHI in the land record is on the order of 0.03C per decade for urban stations.
This result is consistent with both the expected sign of the effect and regional estimates covering the same time period (Zhou et al 2004) and differs from some recent work suggesting zero or negative UHI bias (Wickham et al, submitted).
Stricter urbanity proxies that result in a smaller set of rural stations show larger urban-rural differences in trend. The upper limit on UHI bias between rural and urban stations is on the order of 0.06 to 0.1C per decade. However, these cases are clearly problematic from the spatial coverage aspect, as the number of rural stations becomes vanishingly small when the most stringent filters are applied. Adopting cutoffs that define rural less strictly leads to more reasonable spatial coverage and an estimate of UHI bias in the record that converges on 0.02C to 0.04C per decade across the proxies. The station pair approach avoids this issue by limiting the analysis to areas with both rural and urban stations available, but has limited global coverage and excludes large areas in India and coastal China where rapid urbanization has been occurring in recent decades.
It is likely that homogenization will further reduce the observed UHI-related bias, as many urbanity biases are detectable through break-point analysis via comparison to surrounding rural stations. We are currently in the process of using the Pairwise Homogenization Algorithm (Menne and Williams 2009) on GHCN-Daily data to examine the effects in more detail. However, it remains to be seen to what degree UHI bias can be removed via homogenization in areas like coastal China and India where there are few rural stations and where station densities are not particularly high in the current version of GHCN-Daily. In any case, the acquisition of additional station data outside of urban areas in these parts of the world would likely be benefitial.
Acquiring more accurate station location data will allow us to use higher-resolution remote sensing tools to identify urban characteristics below the 5 km threshold, and better test effects of site-specifc vs. meso-scale characteristics on urban warming biases. In addition, validated site locations allows for more refinement in the definition of rural stations as a function of distance from urban cores of various sizes.
References
Center for International Earth Science Information Network (CIESIN), Columbia University; International Food Policy Research Institute (IFPRI); The World Bank; and Centro Internacional de Agricultura Tropical (CIAT). 2004. Global Rural-Urban
Mapping Project, Version 1 (GRUMPv1): Population Density Grid. Palisades, NY: Socioeconomic Data and Applications Center (SEDAC), Columbia University. Available at http://sedac.ciesin.columbia.edu/gpw.[Aug 14, 2011].
Elvidge, C.D., B.T. Tuttle, P.C. Sutton, K.E. Baugh, A.T. Howard, C. Milesi, B. Bhaduri, and R. Nemani, 2007, “Global distribution and density of constructed impervious surfaces”, Sensors, 7, 1962-1979
Fischer, G., F. Nachtergaele, S. Prieler, H.T. van Velthuizen, L. Verelst, D. Wiberg, 2008. Global Agro-ecological Zones Assessment for Agriculture (GAEZ 2008). IIASA, Laxenburg, Austria and FAO, Rome, Italy.
Klein Goldewijk, K. , A. Beusen, and P. Janssen (2010). Long term dynamic modeling of global population and built-up area in a spatially explicit way, HYDE 3 .1. The Holocene20(4):565-573.
Klotzbach, P., R. Pielke Sr., R. Pielke Jr., J. Christy, and R. T. McNider, 2009. An alternative explanation for differential temperature trends at the surface and in the lower troposphere. J. Geophys. Res.
Menne, M.J., I. Durre, R.S. Vose, B.E. Gleason, and T.G. Houston, 2011: An overview of the Global Historical Climatology Network Daily Database. Journal of Atmospheric and Oceanic Technology, submitted.
Menne, M.J., and C.N. Williams, Jr., 2009. Homogenization of temperature series via pairwise comparisons. J. Climate, 22, 1700-1717.
Schneider, A., M. A. Friedl and D. Potere (2009) A new map of global urban extent from MODIS data. Environmental Research Letters, volume 4, article 044003.
Schneider, A., M. A. Friedl and D. Potere (2010) Monitoring urban areas globally using MODIS 500m data: New methods and datasets based on urban ecoregions. Remote Sensing of Environment, vol. 114, p. 1733-1746.
T. R. Oke (1982). “The energetic basis of the urban heat island”. Quarterly Journal of the Royal Meteorological Society 108: 1–24.
Wickham, C., J. Curry, D Groom, R. Jacobsen, R. Muller, S. Perlmutter, R. Rohde, A. Rosenfeld, and J. Wurtele, 2011. Inßuence of Urban Heating on the Global Temperature Land Average Using Rural Sites IdentiÞed from MODIS ClassiÞcations.
Submitted.
Zhou, L., R. Dickinson, Y. Tian, J. Fang, Q. Li, R. Kaufmann, C. Tucker, and R. Myneni, 2004. Evidence for a signiÞcant urbanization effect on climate in China. Proceedings of the National Academy of Sciences.
Ziskin, D., K. Baugh, F. Chi Hsu, T. Ghosh, and C. Elvidge, 2010, “Methods Used For the 2006 Radiance Lights”, Proceedings of the 30th Asia-PaciÞc Advanced Network Meeting, 131-142
Discover more from Watts Up With That?
Subscribe to get the latest posts sent to your email.
Another microsite issue is “snuggling up”, for lack of a better term. As data collection went electronic, direct cabling substituted for observers’ feet, and recording stations ended up close in to recording buildings, virtually guaranteeing that their measurements lost almost all benefits of natural surroundings and isolation. This applies even to many rural stations.
As I read it, this paper finds average UHI in USA to be 0.06 to 0.1C per decade with the number of genuine rural stations being “vanishingly small”. As the spatial reality is that the rural percentage is dominant, the sample is clearly unrepresentative.
However, as it is not “reasonable” to have such biased spatial coverage, the definition of “urban” has been eased off so that the category covers only those stations which converge on 0.2 to 0.4C.
Is this a case of changing the data to fit the hypothesis?
The upper limit on UHI bias between rural and urban stations is on the order of 0.06 to 0.1C per decade. However, these cases are clearly problematic from the spatial coverage aspect, as the number of rural stations becomes vanishingly small when the most stringent filters are applied.
This is not a justification for using less stringent rurality criteria. That the right answer is difficult to achieve does not make the wrong answer more correct.
While it is beyond the scope of your research, I would also be interested in the impact of multidecadal changes in atmospheric weather patterns on UHI over long periods of time. For example, if we were to see a fairly reliable seasonal storm track make a persistent move in location, we might see a significant change in regional temperature trends. This would be particularly so if the track moved from impacting a more urban location to a more rural or vice versa. If a city, for example, were to get fairly regular seasonal rainfall for a period of, say 30 years, and then the dominate storm track were to move, that urban area might experience a lot more solar heating. A rural area that has gone from mostly sunny conditions to more stormy conditions might not see as much of a change in temperature so the impact of the movement of the storm track might act to cause an amplified response.
If the move of the weather pattern moved from one rural area to another or one urban to another, it might not have as great an impact. What prompted this thought was a paper I read a while back where evidence was found of past movement of the ITCZ in Africa where it has apparently changed position by some 300 miles for a considerable period of time. Basically, it moves from time to time. I was then thinking of how the PDO might impact storm tracks in the US, particularly along the West Coast. Storms making landfall around the Portland/Beaverton/Vancouver area reliably over a long period of time might have considerable impact on the “normal” temperatures of those cities. Should the storm track move a considerable distance for a fairly significant amount of time leaving those cities with more sunshine, the temperature trends for the Pacific Northwest might be improperly biased when what we are actually seeing in the trend is solar heating of the concrete and not really much of a real change in climate. Whereas a movement in the opposite direction, from landfall in a relatively rural area to a track over a major population center would show a rather pronounced negative trend in the data that is exaggerated by the attenuation of UHI due to increased cloud cover.
Considering how sensitive policy makers are these days to tiny changes measured in tenths of a degree and how these tiny changes over short periods are often extrapolated over long periods for planning purposes, such changes could cause over-reactions and adoption of needless and expensive policy measures to “combat” something that is simply a reflection of a natural cyclical storm track change and not a general change in global climate.
I’ll be interested to see what impact your homogenization techniques have on the difference in urban-rural trends. Frankly, for purposes of climate monitoring, I don’t think we should be using urban stations at all. I believe we should be using rural or parkland for locations for such monitoring. Most states have many acres of state/national forest/park/preserve land available for such monitoring locations.
Well, the first thing I want to say is that what a biased map you show up at the very beginning! How many rural sites are partially (or even fully) obscured by urban sites—particularly in the US? This is not really better than the way Mann et al. enjoy putting a big fat line on top of their reconstructions to hide any recent insensitivity. I’m very sensitive to people who try to lie (or mislead) with graphs.
I wondered about that, too, Alexander, because I know there is some pretty rural area in Kansas and Nebraska and the Dakotas that show up as “urban” in that map. As in most of those states are rural except for a few smallish cities. Heck, some places out there you can go 40 miles between tiny little towns.
Thanks for posting this poster. It deserves a careful and thoughtful read. I have used some of these same techniques in my work on ore-bodies. The pair wise comparison can be very useful but as pointed out can sometimes be problematic and needs to be applied with care. In some locations it may not work. It seems to me, instead of always needing to make adjustments for urban growth and other changes, we should quickly and diligently establish new rural locations that will remain unchanged or unaffected for a long period of time. If enough true rural stations are available we would not need to use “adjusted data” which will always present difficulties.
Considering that virtually NONE of the existing global historical temperature data is of proper sample size nor of a proper random sampling regime to obtain a reasonable confidence level or interval, it barely matters that UHI sampling bias is present. A sample size of 1 from a effectively infinite population size for each of 30 days does not entitle the scientist to 30 DF for minimum nor maximum temperature. The corruption of the science of sampling is truly awesome all the way down to the purported precision down to hundredth of a degree..
Neither have I ever seen an honest accounting of the actual extent / range of all the various errors in the data set.
GIGO – the data is trash no matter how you stack it.
One thing to consider is that for a lot of these sites, there will also be wind data. You could also extract wind data from the weather maps at the time on a large scale.
So now you could test a site that is to the east of an urban area. You also need a site to the west of an urban area. Depending on the wind direction you either get no difference, or a bias one way or the other.
Interesting work.
You say;
“Ten out of 12 months of data are required to count as a complete year.”
Do those ‘missing months’ vary-i.e one record misses out January and February, whilst another misses out June and July?
How is the missing data accounted for?
tonyb
Thanks for posting this, v interesting work to read. Congratulations.
The pdf link shows the whole poster.
The two maps on the bottom left of the poster, when viewed at 150% show that
1. Map of Urban & Rural Stations – One station in the Antarctica ice & one to the NE in the sea. None in the Arctic
2. Map of Urban % Rural Pairs – None in the Antarctic and none in the Arctic
Studies that use twins would provide a useful baseline methodology!
Why do I, in these studies, not see more of formulations like
– CHANGE through time in the DEGREE of urbanization/ruralness (probably most important to some rural areas),
– CHANGE through time in the way we “furnish” our urban areas (more or less parks and gardens and more or less of offices and residential usage),
– CHANGE through time in the way we heat our urban areas (during the last 40 years all major urban areas in Sweden has changed to area wide heating systems – more efficient and thus less leakage to the general environment )?
Gösta Oscarsson
Stockholm
–
BioBob, I think for purposes of “climate change” it really isn’t all that important how accurate it data are so much as you capture the proper trending of it. And when you are doing things like creating a global aggregate average and simply paying attention to the trend and not so much the actual value, you can probably get by with a small sample. The notion isn’t really to accurately measure the “average” temperature of the planet as it is to get an accurate “signal” or the variation of the data.
Even a single station per US state could do it provided you had a location with no other influences to its trend except changes in the climate. That is what make land use changes so tricky. If you have 100 stations in a given area and over half the record suddenly a dam was built and massive irrigation started, you get a change in trend. I believe that FEWER stations would actually make it easier to spot trends but at the same time you have to be very careful to select stations with minimal surrounding changes over the years. A station in a county with about the same population it had 100 years ago with a surrounding terrain that has remained basically unchanged for 100 years would probably give a good indication of the local climate trends.
My personal belief is that for the most part, the station records have too much locally induced noise to be of much value. That is the price that you pay for using stations near people. A station in the middle of Redden State Forest, DE or Tumbling Run Game Preserve, PA or Dixie National Forest, UT probably won’t show much human induced variation.
Places away from airports, away from towns, away from heavy ag use are probably going to give you are more accurate climate signal. More stations just means more chances for noise and makes it harder to see any signal. Heck, a collection of stations on Islands in the various oceans might even be good enough but in general, I would want a network of isolated stations about 250 miles from each other.
do you mean to say urban areas are warmer than rural areas?. i guess every one knows that.
BioBob @ur momisugly 12.22
Touché
And with that level of knowledge perhaps you would consider assisting Donna L and plot the distribution of lead and secondary authors, the sample they were likely to have been obtained from and then the final sample offered of authors in the IPCC summaries. Three maps?
And then construct a sample of media outlets for disseminating that work. These days collection & dissemination is the field of market research and economics. And visuals always have great impact.
As McKintrick stated so well in his recent report
What is Wrong with the IPCC: Proposals for Radical Reform
3.1 Appearance and Reality
“Simple card tricks often work by adding in steps that seem to make the outcome impossible, but which in reality have no effect. For example, if the card that needs to be revealed is known to be at the top of the deck, it is a simple matter to shuffle the deck repeatedly without changing the position of the top card. While it looks like the entire deck has been randomised, in reality the shuffle was neutralised with respect to the one card that matters.
The analogy applies to the IPCC process”
In some parts of the westernised world there has been a pervasive shift in vegetative cover. Trees have been making a comeback in many rural areas that were previously completely cleared. This affects the local and regional albedo and local wind patterns. Relatively recent (last 30 years) changes in the nature of the land surface and the extent of irrigation need to be controlled for. Has this been carried out in the work that the post above summarises?
Even if the months with extremes of heat and cold were ditched the data would still only give a false positive result. Aristotle would be dumbfounded at the continual error of climatology in arguing from the particular to the universal. A a good enough guess in science ain’t science it’s pseudoscience.
Reconstructions of temperature rely on a global network of stations?
The problem is they are NOT global and probably not fall within the definition of ‘network’.
The alarmist reconstructions rely on a global network of scientists who are not bothered about altering, deleting or otherwise changing their data to fall in with the AGW ideas.
I really need to do some proper science about UHI as i live in the perfect city for it.
I remember as a kid doing some simple geography field trip work in a village and we could dectect a 1-2 differnce in temperature between the village and the sourrounding countryside.
Which is almost exactly the global warming signal for the past 100 years.
From the city centre where i live Sheffield UK, a city with 500,000 people and the 4-5 biggest in the UK i can be within a national park in 20 minutes drive and a few miles.
Almost directly along a straight road, out into the countryside. The temperature in winter can drop at least 2-3 degrees from work in the city centre to where my folks live in the countryside.
This is actually greater thean the global warming human induced signal.
Intrestingly sheffields weather station is within a few 100m of where i work.
http://www.museums-sheffield.org.uk/museums/weston-park/home
What is e(smy), it has not been described above.
I find it interesting that they would look at Rural stations up to 500km away to compare to an urban station. Everyone knows that “climate” can change in as little as 10km. I know they “grade” them by distance, but it still seems a strange thing to do.
What is very misleading is that there are very few Rural stations compared to Urban ones and yet spatially there is much, much more rural land area than urban. Where in their algorithms is this allowed for. Shouldn’t there be a major biasing for rural stations to match the rural area difference?
What is the difference in Trend between all RAW Rural Vs all RAW Urban without any kind “adjusment”?
The most interesting UHI study that I have read is about Barrow, AK (Hinkel, et al. 2003 and Hinkel, et al. 2007). They used enough sensors to map the temperature field around Barrow. This then allowed them to establish a scientific basis for which urban sites were most representative of the urban UHI and which rural sites were most representative of the “normal” temperature. On this basis, mean Winter UHI was about 2°C. Barrow grew from a population of about 300 in 1900 to about 4,500 in 2000. Hinkel attributed substantially all of the UHI to use of natural gas extracted from local fields. On the basis that these fields were not in production in 1900 (my guess) and that the population then was much smaller, one could estimate that the UHI in 1900 was probably close to 0°C. Thus, the mean Winter UHI of 2°C would have developed over about 100 years or ten decades. I would submit then that the Winter UHI trend bias for Barrow would be 2°C divided by 10 or about 0.2°C per decade. Since fossil fuel use in Barrow is primarily for Winter heating, I would then divide by 4 seasons to obtain an estimate of 0.05°C per decade.
Given that Barrow would likely be classified as a “rural” station in the lower USA based on population alone, and that this study assumes that the mean UHI for “rural” stations is zero, I think one could question whether this study significantly underestimates the “rural” mean UHI trend bias. The argument could be made that using the Barrow study results as an estimate of the “rural” trend bias in the lower 48 overestimates it, because there is a larger difference between inside and outside temperatures in the Winter in Barrow than would be the case on average for the lower 48. However, this study implicitly assumes that the “rural” station UHI trend bias is zero, when it may well be significantly different from zero. ANY non-zero “rural” UHI trend bias would be in addition to the difference bias that this study estimates.
Unlike Hinkel, this study does not seem to develop a scientific basis for representativeness of the urban or rural stations as estimators of the “urban” mean UHI or the “rural” mean UHI. Such representativenes is simply assumed. Also, the study appears to be heavy on statistical examination of the station data, while not checking for the presence of other factors that might cause differences between “urban” and “rural” station pairs. Hinkel refers to Daly 2006:
Re AC Osborn +1 for how climate can change over very short distances. 20 miles from me, farmers grow wheat and oilseeds, I’d struggle to get a decent crop of barley even with its shorter growing season. (Same altitude, 100M above sea level)
Also, in the same way as you get ‘light pollution’ whereby one can see the (usually orange) glow of city streetlights many tens of miles away from the actual city, there must surely be an equivalent for the down-welling infra-red. Can rural thermometers ‘see’ an infra red glow from urban areas that may be dozens of miles away and over the horizon – are all thermometers thus affected?
It might just explain this entire global warming scam…..
Menne ? …hmmm… isnt that one of the guys who did a trick on Anthony regarding station data?
Incredibly there is only ONE rural station in the GISS database in the whole of Africa that has a record back to 1941 and with less than 20% missing data in the last decade. This is Calvinia in S. Africa and this shows NO increase in temps.
Unsurprisingly UAH temps show a much smaller rise than GISS since 1979.
http://notalotofpeopleknowthat.wordpress.com/2011/11/19/giss-and-their-african-temperatures/