I want to congratulate Zeke and his co-authors on their presentation on urban heat island. The main conclusions are:
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).
I assume Zeke (and Steve and Nick who comment often) are all busy at the meeting and will be able to discuss more when each is caught up!
Yes, congrats to all of you.
And Muller saw UHI as upside down!
The poster itself is available here for folks who are interested: http://eposters.agu.org/files/2011/12/Hausfather-et-al-Urban-Trends.pdf
Zeke, et al, unless I am misreading the summary, you are assigning ~.3C/century of warming to UHI? AGW is confirmed? So, the solution is …?
CoRev,
Around 0.03 C per decade urban-rural difference from 1979-present (full possible range of 0 to 0.1C across all sensitivities examined, though the higher end suffers from significant spatial coverage biases as the number of rural stations becomes vanishingly small). We make no claims about global century-scale UHI.
CoRev,
Just to clarify, I’ll also point out that the 0.03 C per decade appears to be for the urban-rural difference. So, since there are rural stations within the temperature record, the actual bias in the combined rural+urban record would be less than that. Furthermore, that is only for the land record, which only represents ~ 30% of the surface temperatures. So the bias they found would not be 0.03 C per decade in the overall surface temperatures.
CoRev (comment 86821)
No this certainly does not confirm AGW theory. It only supports that UHI error by itself is a small source of positive bias in the surface temperature instrumental signal. With the number of error sources involved (UHI, station moves, instrument changes, human error, instrument error, station micro-climate, time of day adjustments, poor coverage, missing measurement data, etc… etc…) it’s contribution should be considered significant (though not a significant portion of the trend all by itself) and added to the others (the sum of which might be significant in proportion to the trend itself). The result of this particular UHI error study is a slight chip off the instrumental signal and a slight increase in known error. For skeptics who had wanted to quantify the approximate impact of UHI error, they now have an answer. In my opinion, overzealous skepics now cannot question whether the overall signal is made up in large part by UHI error. However, that is VERY different than confirming AGW theory… it doesn’t impact attribution studies (though this result may be much more significant if the assessment of natural cycles’ contribution to the signal is increased). It doesn’t impact preinstrumental (proxy) temperature reconstructions (which themselves impact model long-term hindcasting and attribution studies). It barely impacts climate models/errors. MOST learned skeptics don’t dispute that temperatures have been rising for decades. But by how much (impacted by this study), by what cause (not impacted by this study), how unprecedented the change is (not impacted by this study), by how much will it in the future (not impacted by this study), how certain are we (slightly impacted by this study; this is just one small error source out of many many) there is still plenty left to cover.
The climate is the continuation of the oceans by other means, and the oceans are the continuation of the sun by other means.
=======================
I remember reading something from Watts a while back suggesting that some stations counted as rural were adjacent to black asphalt parking areas and would likely show some localized heating. Did your paper go into that at all?
I will also point out that if we back up the start date to 1950
the bias goes down, so much for the extrapolation argument.
Let me just give some stats.
1. 50% of our urban stations are in areas where there is less
than 7 sq km of built area. in the 380 sq km surrounding the site You could call these peri urban.
2. 75% of our urban stations are in areas where there is
less than 30 sq km of urban area. you could call that
small town.
3. 25% of the sample ranges from 30 sqkm up to 380sq km.
The amount of UHI is a linear function of the area built,
up to a certain threshold. (perhaps a log function of area )
I believe you will find that
threshold somewhere around A few hundred sq km
( I have the threshold calculated for the US, just
need to write it up ). Some number of the Urban
stations would appear to be thresholded. That is
their UHI bias is “baked in” at the start of the period
and since its thresholded it cant really go higher.
One could I suppose use the UHI versus area regressions
to back out an all rural network, that would be a
fair amount of work and dependent upon regressions
which are statistically significant but have relatively
low R^2.
I
“…there are many individual cases of urban cooling.”
Or rural warming. Your methodology does not allow to decide, isn’t it?
Heh, phi, nest paw?
========
I also assume that the 0.03 degree C per decade bias is estimated before any considerations of homogeneity adjustments.
As one who thinks that the uncertainty of temperature trends from station measurements is probably more related to individual station micro climate differences than differences from general categorical sources like UHI, I suspect that the paper will have no surprises for me. The model I saw in the paper for trend differentiation appears to me to be based on what Nick Stokes has shown at his blog.
Looking at my own breakpoint analysis with the USHCN station data, appears to me to show some very localized breaks in the climate. It needs more work – as it turns out that the breaks one calculates is dependent on the segment length one chooses for the calculation.
Troyca –
Do you have any quantification for this? [or do you Steve, Zeke et al?] What proportion of the stations are rural?
Perhaps another way of putting it is, if the current UHI adjustment isn’t appropriate, what is? What should it be?
Troyca – I take your point, but 0.03C/Decade is still a fifth of the total per decade rise. It isn’t insignificant…
What meeting?
Anteros-
I agree it isn’t insignificant, which is why I’d worked on a similar project with Zeke related to UHI in the USHCN (Zeke presented it at a conference, not sure if it’s going to be published in a journal?). From that project, it was usually around 50% in each bin, so a rough estimate might be 0.015 C/Decade actual bias in the unhomogenized record (but obviously Zeke, Mosher, or Nick could answer better with this particular project).
It is likely most significant when it comes to comparing land and regional trends to models, perhaps with respect to aerosol effects as well. However, if you’re talking about the relationship between the entire surface temperature record in recent decades and UHI, you’re looking at an upper bound of ~ 0.1 C * 50% * 30% (land vs. ocean) = .015 C / decade over THIS time period, which would be around 10% of the total warming. Other methods may produce higher estimates of the signal, and we don’t know the effect going back further, but I don’t think these particular results overturn the general understanding of recent warming in the global surface temperature series.
Troyca –
Thanks. Has there been a good/convincing explanation for the upside-down (or insignificant) BEST UHI results? It still seems such a surprise.. added to with the above results.
@Anteros,
It does significantly drop my confidence in BEST’s results…
Troy,
Don’t worry, our US paper is still going to be submitted. I’ve just been somewhat sidetracked with my day job and getting this presentation ready for the AGU. Starting next week Matt and I will be working on it extensively.
On translating the urban-rural difference to an overall bias, its difficult because the spatial density of urban and rural stations differs by region. I’d say 40% rural 60% urban is a good conservative rule of thumb, but we will try and get a more formal analysis done that controls for spatial bias.
SteveF,
The American Geophysical Union meeting in San Francisco. 17,000 earth scientists in one place makes for a pretty impressive sight!
Anteros,
I’ve talked with Rhode about it a bit, and I’m pretty sure that at least part of their results is due to uncorrected issues with spatial coverage bias. That said, one of the projects that Mosh wants to do is run the BEST data through our method and see if the results change at all.
phi (Comment #86831)
December 8th, 2011 at 11:46 am
“…there are many individual cases of urban cooling.â€
Or rural warming. Your methodology does not allow to decide, isn’t it?
###############################
It’s hard to make sense of this comment.
Looking at Long records you will find a very small number of stations that show cooling over the 1979-2010 period. If you
increase the data quality requirements ( more complete records )
The number of “cooling” stations diminishes. These coolings happen in both urban and rural stations. I can run some checks but you are most likely to find them near coasts or bodies of water.
Zeke et al. use the same methodology than BEST and all others who have tried to quantify UHI or more generally the disturbances of stations. This work is interesting but it can only estimate the difference between rural and urban stations.
There are a few tracks to estimate the total perturbations, such as proxies. We could also try another method based on the results of Zeke et al.
What they get is an estimate of the UHI in the strict sense which could be seen as the influence of urbanization within a radius of 1 km to 10 km. Very roughly, this represents about a third of potential disturbances because we have, by simplifying, slices from 10 to 100 m and 100 m to 1000 m. With this calculation, we see that the total disruption could very well be of the order of 1 ° C per century which is the value that suggests proxies.
Steven Mosher,
“Looking at Long records you will find a very small number of stations that show cooling over the 1979-2010 period.”
I speak of course of relative cooling or warming.
Anteros (Comment #86839)
December 8th, 2011 at 12:26 pm
Troyca –
Thanks. Has there been a good/convincing explanation for the upside-down (or insignificant) BEST UH
################
let me say a few things about BEST.
1. they used a .1 degree “radius” around the sites
at the equator this means 11 km. that grid is say 380 sq km
at 45 degrees North or south.. HA, they are searching less
space.. almost 50% as much, so there definition of rural
get MORE LAX as they move away from the equator.
This leads them to put urban stations in the rural Bucket !
Since the difference between rural and urban is so small,
ANY contamination of the rural class kills you.
2. Modis has a great commission error rate ( 98%) that means
if modis says that the pixel is urban—- Then 98% of the time
its urban. Conversely, modis’s OMMISSION error rate is less
good ( say around 75% in one comparision) That means if
Modis says its rural, then you cannot be sure it is rural. Most
of the time it will be correct.
3. To handle he ommission errors, we apply addition filters to the
rural. We apply the following additional filters
A. No airports : that elimates about 230 sites from the
rural category
B. Nightlights at the site location have to be less than 30
There are a number of sites in areas like australia that
have bright lights ( 1200) but no modis built pixels.
That possible because a modis pixel is 500meters and you
need two contiguous pixels. After applying a filter
of 30, 50% of the sites are pitch black and 75% have
lights less than 7. 30 is the cutoff used by Hansen.
C. We apply a ISA filter of 10% at the site location. After
applying this filter 75% of all sites have less than 1%
impervious surface.
These additional filters and the relatively small number of stations they apply to have a noticeable effect on the Bias
estimate. You can see that by comparing the modis sensitivity
( which does not have these additional filters )
So its a combination of several things. First, their filter changes as a function of latitude. UHI also changes with latitude so this is NOT a good thing they did. Second, they did not control for Modis Ommission error rates. Third, the scalpal.. Guess what, you apply a scapal on change points and you will get rid of some UHI.
Congrats to all. Any chance to get an explanation on the methods for trend comparisons? The description on the poster is a bit “terse”. (E.g. am I wrong in assuming that at some point you use the word “pair” to mean “one member of a rural-urban pair” ?)
Zeke,
Glad to hear it, I know you’ve been busy! Congrats on the poster, you’re quickly becoming the expert on this UHI topic 🙂
It does not appear that anyone from RealClimate attended the presentation. They have not mentioned it in their posts from, and about AGU.
Zeke and All – For the sampling bias for urban sites (and other landscapes), please see our paper
Montandon, L.M., S. Fall, R.A. Pielke Sr., and D. Niyogi, 2011: Distribution of landscape types in the Global Historical Climatology Network. Earth Interactions, 15:6, doi: 10.1175/2010EI371. http://pielkeclimatesci.files.wordpress.com/2011/02/r-344.pdf
Our abstract reads
The Global Historical Climate Network version 2 (GHCNv.2) surface temperature dataset is widely used for reconstructions such as the global average surface temperature (GAST) anomaly. Because land use and land cover (LULC) affect temperatures, it is important to examine the spatial distribution and the LULC representation of GHCNv.2 stations. Here, nightlight imagery, two LULC datasets, and a population and cropland historical reconstruction are used to estimate the present and historical worldwide occurrence of LULCtypes and the number of GHCNv.2 stations within each. Results show that the GHCNv.2 station locations are biased toward urban and cropland (.50% stations
versus 18.4% of the world’s land) and past century reclaimed cropland areas (35% stations versus 3.4% land). However, widely occurring LULC such as open shrubland, bare, snow/ice, and evergreen broadleaf forests are underrepresented (14% stations versus 48.1% land), as well as nonurban areas that have remained uncultivated in the past century (14.2% stations versus 43.2% land). Results from the temperature trends over the different landscapes confirm that the temperature trends are different for different LULC and that the
GHCNv.2 stations network might be missing on long-term larger positive trends. This opens the possibility that the temperature increases of Earth’s land surface in the last century would be higher than what the GHCNv.2-based GAST analyses report.
Oh, forgot to say congrats on the poster and the hard work! Really great stuff you guys have done, and it’s nice to finally have a serious, stringent quantification of this issue.
Zeke and Steve, I would be really interested in the results of pairing of stations idea you guys were discussing at the conference.
One of the things I really worry about with these studies is there is a higher density of urban stations near coast lines, and that can have a substantial influence on the measured temperature trend. In fact, in many cases, when coastal stations were moved, they were moved inland to avoid coastal development.
At more northern sites, coastal sites can have as much of a 40% (lower) bias on trend, so this is a real issue, and it could be causing you to underestimate the amount of UHI bias present in the data.
I think Zeke’s idea for geographically proximate pairing of long-period staitons that however excludes coastal sites would be an interesting alternative for testing for UHI versus rural effect would be a great way to approach this.
The mythbusters use an ex-military housing complex, abandoned, for some of their driving myths.
I wounder what would happen to the average temp if you painted the tarmac roads with a thin layer of cement and changed the color of the roofs?
Well done boys (in the belief you are all male)
“I think Zeke’s idea for geographically proximate pairing of long-period staitons that however excludes coastal sites would be an interesting alternative for testing for UHI versus rural effect would be a great way to approach this.”
Station temperatures near large bodies of water do not correlate well with those inland. I think it was you that intimated in that if a model like that used by BEST took altitude into account it could also account for some of the coastal variation. Altitude is a variant that should to be included in a model and perhaps separately from that used for proximity to large bodies of water.
I have found that correlation between neighboring stations can be reasonable while at the same time that trends can differ significantly. Trends can vary within local areas to the extent that the noise makes it difficult to estimate statistically significant difference. You need large sample sizes and/or large differences in the trends of interest. I found this situation to be true with the analysis of CRN ratings and trends for USHCN stations.
As a sidelight I have found what Menne found in determining breakpoints for homogeneity adjustments in temperature series and that is that the worse two stations correlate the more difficult it is to determine breakpoints in their difference series. It is like the stations have to be more the same to more readily find differences between them.
Zeke, I found the material in your link well wriitten and of much interest to an observer like myself. I was wondering, however, after a quick glance if that is the entire paper or there is more – and if more where could I find it?
OT
UAH seems late this month, but RSS came out a few days ago.
November anomaly: 0.033.
http://www.remss.com/data/msu/monthly_time_series/RSS_Monthly_MSU_AMSU_Channel_TLT_Anomalies_Land_and_Ocean_v03_3.txt
JohnM–
Yes. I hope Roy is ok! I’ll post RSS plots tomorrow. Of course, that’s not for the quatloos!
He’s probably at AGU and didn’t get around to posting the result before he left.
Mosher
Here at conference. Best new results. Matching ours.
Roger we don’t use ghcn. I looked at land use and land use changes
Nothing significant except cultivation.which isn’t that significant
Thanks, Lucia,
I didn’t make it to the meeting, but I’m following what I hear with interest.
A little early, but I wish you a happy St Lucia’s day. Do you celebrate?
And John, we do have a TempLS figure. Down 0.12°C from October.
Toto.
two methods: the global method takes all urban and all rural.
A least squares solution is done for each individually which gives
you annual anomalies. Then the difference is taken and the slope of that is calculated.
Pairs method: an urban base is selected. Then a rural composite is constructed. The trend for both are calculated and differenced
Carrick
Urban stations: 10% are within 2km of the coast
Rural: 15%
As you know there are two elephants that can confound any finding
Latitude differences and proximity to coast. i’m noodling on the coast thingy as the effect falls off with distance from coast..
need a good transform for the distance from coast data.. or I might just hunt for the threshold. Any thoughts on the underlying physical process? should the effect fall off log like? etc. Note,
even a log transform gives me bi modal dist in the regressor, so
its quite ugly just regressing trend verus distance from coast.
Significant but r^2 sucks.. eg 100km is no diferent than 1000km.
I could make coastal a factor, but then I have the threshold question. Islands dont make matters simple either
Steven, I had suggested just using elevation as a proxy for coast line. That should be easy to code in right?
And I would do a contiguous US only reconstruction (Zeke’s method would work, I’d update his anomaly method to subtract off the linear trend like I do…maybe a cheaper way of getting to what Nick is doing).
US only eliminates islands and allows you to be really picky about which data you keep.
Zeke,
for the dummies like me, what has been done to assign this difference in the trends between stations to UHI as opposed to alien visitations? What has been done to show this difference is the TOTAL UHI for the stations as opposed to it being the NET difference in UHI between the stations?
I see these studies as only showing the net differrence in TREND between stations that could be based on multiple and differing reasons. How do you attribute causation?
Roger,
You used the wrong nightlights data. the dataset you used was the f15 stable lights. You want the f16 radiance calibrated lights.
Also the population numbers you use from Imhoff are based on US data and cant be generalized to the world. You should look at
The recent papers by Ziskin and baugh given at the asian pacific conference. There you will see how poorly nightlights does in areas like india. Or yet use Landscan population and then calibrate it against Hyde.
Also, your land cover data was outdated
“Do you have any quantification for this? [or do you Steve, Zeke et al?] What proportion of the stations are rural?”
1/3
A holiday involving cookies? Of course. 🙂
kuhnkat,
If aliens (or leprechauns) only visit rural stations, they may well be the cause. We aren’t examining the modalities per se, just the correlations. That said, we are looking at all different proxies for urbanity (modis, nightlights, isa, historical population growth, airport locations, vegetative cover, historical changes in vegetative cover, etc.) for different radii around the station location and with different cutoffs for urbanity.
As for how we determine the total rather than net UHI trend, the best way we can do this is look at the most rural stations. The trend in urban rural differences does increase as you become more stringent in your selection criteria, which inandof itself is an interesting result, but with very very stringent criteria you start running into issues with spatial coverage of rural stations. That said, when using the pair method that controls for spatial coverage we still don’t see that much more than a 0.3-0.5 C difference at the extremes.
Me being almost totally naive on this issue, I think I can help you 🙂 The problem is there is virtually no coverage, where the people ain’t. Let’s pretend we have had those missing stations in place throughout the period of your study. So, we got to create them. Get really stringent on the ruralness criteria, until you are left with one station. Use the anomaly data from that model rural station to create enough stations to cover those vast parts of the land surface that you know to be places where the wild critters reign. Then do your thing. My guess is you will find that the UHI bias is bigger than you thought it was, before I helped you.
Dr. Pielke,
Thanks for the link to the paper. We’ve looked at a few different vegetative cover datasets (both contemporary and the Hyde historical data), but we haven’t done too much analysis with them to date. Its definitely high on our to-do list, however!
Oops, I meant 0.03 to 0.05 C per decade in the earlier post. Silly orders of magnitude.
I believe this is the reference Steven was referring to.
Kuhkat
“kuhnkat (Comment #86882)
December 8th, 2011 at 7:01 pm
Zeke,
for the dummies like me, what has been done to assign this difference in the trends between stations to UHI as opposed to alien visitations? What has been done to show this difference is the TOTAL UHI for the stations as opposed to it being the NET difference in UHI between the stations?
I see these studies as only showing the net differrence in TREND between stations that could be based on multiple and differing reasons. How do you attribute causation?”
###########################
very good question. First off we accept climate science 101. UHI is real. We do not accept this on faith but we accept it based on many studies that show that urban areas are warmer than their rural surrounding. The are empirical studies. For example, see Anthony Watts who ran a transect across Reno.
Now, collecting temperatures from Urban locations and comparing it to rural does involve a supposition that invisible gremlins did not fuck with the data. Like most experiments. Also, to back up these correlations you have a physical theory that explains why urban areas should be warmer. If you like I can explain that.
But We didnt stop there. ALthough the poster doesnt show the work I will explain a little bit of what else I did.
I took rural stations and I looked at the trends. Then I looked at all the features that could explain the trends. For the trends we
see in rural stations the only variables that mattered were latitude and distance from Coast. I checked for the presence of aliens.
When aliens were present the Trend was the same as the trend you see when aliens were not present. Alien visitation data is available online. Just ask John O’Sullivan, he’s got the links.
I then looked at urban stations: here the story was different.
The Trends were explained by three variables; latitude, distance from coast and the amount of built area near the site. Again, with and without alien visitation made no difference. I even tested for the presence of snarks. They too were not significant.
I hope that helps.
Sorry carrick I confused Kims paper with Doll’s
There were three papers from that conference and I should have
posted the links ( in a rush )
here is the Doll paper, see table 2 for an explanation of why using
nighlights to judge population on a global scale is problematic.
http://www.ngdc.noaa.gov/dmsp/pubs/APAN_30_Doll.pdf
Also, this paper is a must read
http://www.ngdc.noaa.gov/dmsp/pubs/APAN_30_Ziskin.pdf
If you look at the DN numbers that roger refers to it appears he is using stable lights as opposed to radiance calibrated lights.
The radiance calibrated lights gives you a much wider dynamic range.
For our work we dont use population at all to classify stations. If I had to make an argument about that it would go something like this.
1. Population in an of itself does not cause UHI. The infra structure
required by higher populations does. Put another way,
50000 people living in grass huts does not cause as much UHI
as 10000 people living in a concrete jungle.
2. Population data in my experience has some really weird
outliers. Some areas with huge over estimates and other
areas with underestimates.
That said we do use population data to sanity check. but in the cases where the population data didnt match the other data
( like a urban place with zero population ) it is always the pop
data that is flakey
My favorite
Moomba
http://en.wikipedia.org/wiki/Moomba,_South_Australia
No official population
and hella nightlights for the 1200 guys who fly in and out
http://maps.google.com/maps?ll=-28.11,140.21&spn=0.3,0.3&t=m&q=-28.11,140.21&lci=org.wikipedia.en
The other examples were places in Russia with huge population densities… and nothing there when you actually look at the map
So pop data in my mind is something you want to check, but relying on it is another matter.
Landscan pop data is another matter. Its closed data so I have not evaluated it. My sense is that it is the best of the lot.
Don
“Use the anomaly data from that model rural station to create enough stations to cover those vast parts of the land surface that you know to be places where the wild critters reign. ”
well what most people dont realize is that the most rural most remote stations are in a couple of places. south pole and north pole.
Take the 32 stations that are in the tundra for example. Do you really want to use one of them as your standard?
If you did that, then the global warming Trend for Rural stations would be TWICE as large as the urban stations. That is, the Trend for stations in the Tundra is about .54C over the 1979 -2010 period
urban stations are about 1/2 of that!
Why? well tundra is in the far north. and LATITUDE drives the trend. Thats called polar amplification. If I picked a low latitude
rural station you see a much lower trend for rurals.
That is why we want representative spatial coverage when we do these comparisons. Think of it this way
Rural Trend = f(latitude, coast)
Urban Trend = f(latitude, coast, built area)
To tease out the contribution of built area to the urban Trend
we have to minimize the sampling differences in latitude and
coastal proximity. basically, you cant get to the right answer by your method. But hey, I’ve thought of the same proceedure myself, and after scratching my head and reading carrick I decided that I was mistaken.
Carrick.
I have a raster that provides distance from coast for every 1km
grid, with 0 for cells on the coast. I can also get distance from inland water bodies.
The issue is this. You’ve got a lump of stations that are
within a few km of the coast, ( say 10-15%) then
that bump thins out the futher away from the coast you get
Then you get a huge lump that are very far away.
I can turn that into a binary factor but I have to pick a distance.
arbitrary, dont you think? So I could say 20km or 30km or 2km
We know the coastal effect falls off with distance, So
800km from the coast is no different than 300km. you can see how that effs up the regression.
maybe i misunderstand what you want to use elevation for
Carrick I did a US only a while back to
play around with my regression ideas.
kinda proof of concept
OT:
concerning UAH, i think spencer gave a talk at the AGU yesterday
What explanations could there be for the differences found amongst different studies that look into UHI effects on temp trends? E.g. Hansen (2010) and Wickham (2011) found results that are slightly different. Is that a reflection of unresolved uncertainties (ie the effect may currently be indistinguishable from zero, even if its real effect is non-zero) or does it come down to spatial coverage bias or something else (e.g. choice of data)?
Zeke, Mosth, et al, thanks for your earlier response. I neglected to thank and compliment y’all on this superior effort.
My comment re: .3C/century UHI signal is admittedly a gross simplification, but not more so than what we see too frequently from the CAGW supporters.
Your efforts are giving us a reasonable explanation of the UHI signal, which in turn allows us to extract a more true CO2 sensitivity. Clearly doubling CO2 will have some affect on temperature, but your and others recent research is showing us how poor/overstated were the ACO2 impact estimates.
Keep up the good work folks! Your independent efforts are adding credibility and clarity to the overall GW understanding.
Mosher, excuse the typo, please.
Top Job, well done all concerned
.
Lucia, as you’ve closed the Nov UAH fred
.
Given Roy Spencer’s message regarding the delayed November UAH update due to undersea cable issues!, will we have to bet on December before November’s data is released? I don’t think I can stand the suspense!
.
Frankly, I do think his daughter might have shown better timing with her hatching too but what the hell, you don’t become a Granddad every day, my best to all
My comment re: .3C/century UHI signal is admittedly a gross simplification,
IIUC, it’s not a gross simplification, it’s just wrong. As Zeke has pointed out repeatedly, what they show is that the UHI effect on the temperature records must be lower than roughly half of ~.3 per century, in the un-homogenized data (the major products are based on homogenized data, which should reduce the effect further).
IOW, their work doesn’t refute the consensus, it confirms it.
Maybe I’m just confused.
We only looked at 1979-2010, so extending our analysis for a full century is not appropriate. Also, 0.03 C per decade is about one tenth of the land warming during the 1979-2010 period. Add in oceans (not subject to UHI), and the resulting bias on the global temperature record is pretty negligible. Thats not to say that UHI isn’t real and doesn’t affect the trends, just the effect is small relative to the common warming trend affecting urban and rural stations alike.
One interesting note: the Berkeley folks revised their analysis to find a positive UHI signal of 0.02 C from 1950-present, which is pretty close to what we get for that period.
Kuhnkat,
Since Mosher is the most likely here to have been abducted and anal probed by an alien at a costume party, I’d say he is most likely to be the most qualified person here to answer your alien vs trend questions. I suggest giving him your full confidence on this one.
Looks like RSS came in at 0.03C for November 2011. Please check if correct
Steven M,
Thanks for the explanation. I didn’t mean to be taken literally, on the one station suggestion. Use your imagination on the infilling. The point I was attempting to make is this: “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.” and “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.” Leading to my naive theory that if the unpopulated expanses of the land surface had been properly sampled over the period of your study, the UHI bias might be estimated to be in the .06 to 0.1C per decade range, and would likely be more than .03.
To my untrained eye (connected to my untrained brain) the analysis that you guys have done with the available data seems to be better than anything else I have seen. Better than the BEST, and the rest. And it didn’t cost us anything. I nominate your little group, and Steve Mc, to be the recipients of a $BILLION$ (taken out of the consensus mob’s obscene budget) to re-examine the crap that has passed for climate science, up to now. But keep a close eye on Nick.
Zeke, in your link the following comment excerpted below is of much interest to me. I have been analyzing the USHCN TOB and Adjusted temperature series for breakpoints using the R function breakpoints (strucchange) and I cannot account for all the changes that I see between the individual station data for TOB and Adjusted series – even when I use the R function at its most sensitive. The differences appear in a difference series between TOB and Adjusted (for the same station) as plateaus of varying length. If I assume all these plateaus resulted from breakpoints of differences series of a station with its nearest neighbors I come up short in my breakpoints calculations.
The methods applied by Menne, and I assume that you are using in your initial efforts noted in the comment above, are different than I use in R and are known in the climate science community as the Standard Normal Homogeneity Test or (SNHT). I have searched for R coding for that test without success. Do you know of any such code or a stand alone program that I could conveniently run on my desktop computer? Breakpoint analysis and particularly those required of the Menne algorithm take long, long computer time. I also have noted that Menne used a something with more computing power than a desktop computer for his calculations.
Also interesting that I find the literature on methods for breakpoint analyses comes in two flavors: apparently with one from climate science and one from econometrics and seldom of never have I seen a cross referencing between these flavors in the literature. What’s that all about?
My interest in breakpoint analysis for temperature series has evolved from a better understanding of what Menne et al did in their paper on homogenization of USHCN temperature series to what information and insights the breakpoints might hold for local climate changes and trend variability in local temperature stations.
“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 beneficial.”
Kenneth,
I won’t claim to be an expert in the PHA process, and Menne and Williams are handling that part of the analysis for us (as an aside, it does apparently take quite a bit of processing time, especially for all of the 25k GHCN-Daily stations).
I’m not aware of any existing R packages that include the same functionality. Nick and Mosher work much more with R than I do, however, and might have some thoughts.
We did find in our US analysis that there were considerably smaller urban-rural differences post-homogenization, even when only using rural co-op stations to homogenize both urban and rural USHCN stations.
Steven Mosher:
Yes that is a good idea regardless. With the US you can cut down the coverage to what it is in other parts of the world, and see how well your method handles that. It’s at least a verification that your method works where you know the answer from the higher-resolution data already.
Zeke/Mosher:
Surprised that Steve noted that altitude was not a factor in your model or did I misinterpret?
Did I miss the CIs for your estimate of the trend difference between rural and urban?
Also I see in the 4 graphs shown under the title, Trends in Urban Warming via Other Urbanity Proxies, have what appears to be large differences in the plot of trend difference versus the proxy value for the methods labeled Spatial and Pairs methods. Care to expound on these differences?
Kenneth,
In the extreme cases (e.g. 0 nightlights within 11 km), you are mostly left with U.S. and Australian rural stations being compared with all the world’s urban stations, which introduces some pretty big spatial coverage bias into the results. The pair method avoids this bias by restricting the analysis to areas that have both types of stations in the same (400 km or so) area.
In general, when the line becomes dotted for the spatial method it means that there is < 50% rural coverage for global land, and you should be wary of relying on those numbers too much.
We are going to rethink the pairs approach a bit to try an increase the spatial coverage we can get over the next few weeks. Limiting the analysis to areas with urban stations with continuous 30-year records ends up excluding some pretty large parts of the world, unfortunately.
Thanks Don.
Zeke and Steven,
Is there any way to use the satellite data to check your work, or to add data where it is lacking in the surface station record? I know I am probably talking about apples and oranges, but can we make some fruit salad and come up with something palatable?
Also Zeke: “One interesting note: the Berkeley folks revised their analysis to find a positive UHI signal of 0.02 C from 1950-present, which is pretty close to what we get for that period.” Was that .02C per decade from 1950? If not, it doesn’t look that close.
Steven,
You are welcome and I am working on that $BILLION$. We need some integrity and diligence in the climate science.
Zeke/Mosher:
Surprised that Steve noted that altitude was not a factor in your model or did I misinterpret?
######################
which model. The global spatial model works on anomalies.
The Pairs approach, That is one area where Altitude might
be important, but again we are looking at trend
Regressions to explain trend? ehh, I think I looked at that
long ago. I’ll look at it again at some point, but dont expect altitude differences to explain differences in trends. So, like
stations at 100 meters will warm faster than stations at 130
meters? nope, dont think so
Don Monfort,
Yep, BEST now shows 0.02 C per decade urban rural difference from 1950-present. We get pretty similar results for that period.
toto,
“…the major products are based on homogenized data, which should reduce the effect further…”
See : http://pubs.giss.nasa.gov/docs/2001/2001_Hansen_etal.pdf
Zeke,
They must have seen your poster. At least they have the sign right now.
Don Monfort,
Nah, I think they independently realized an error in their analysis, as they made the change before our poster was available (and our methodologies are somewhat different). Its just nice to see that we are finding close to the same results now.
Don,
Satellite data to check our work.
Well there are two ways
1. Looking at trends in TLT we can say that we explain
Some of the small difference between UHA&RSS and
the land record. Recall, that if you look at the trends
in TLT we saw at most .1C per decade difference
in trends. That .1C difference amounts to what I call
the bias budget. We have a trend at 2 meters
call it .28 per decade over the 1979-2010 period
The estimates for the trend at TLT vary between
.18C and .22C per decade. That leaves a Bias Budget
of around .04C to .1C per decade. We explain
about .03-.04C of that budget. Lets put that another
way. Our All rural trend for the globe is around .26 C
per decade. At TLT the estimates are .18C to .22C depending
on the sat. group you believe in. Looking at regional studies for China ( we cant get their raw data ) we see them estimating .05C
for their region. basically, we are within striking distance of
reconciling the sat records over land with the land record.
There is no huge bias that makes the warming since 1979
vanish.
The second way to use satellites is to look at LST. but LST
cant be compared to SAT.
YA Zeke, they show .02 rural versus ALL. I think we were around
that number for 1950 to present.
Steven Mosher:
What about ≤ 10 m verus > 10 m?
I suspect you’ll see an effect there (sea level correlates with marine influenced atmospheric boundary layer).
ya carrick, the other place where I’ve seen the effect is in mountain valley locations. Ferreting those out is a walk into DEM territory.
The issue with altitude is the uncertainty in location.
you tell me where the station is exactly and I can download
DEM for it (90 meter SRTM) I suppose I could look at all the terrain within a few km.
But gimme a few minutes and I’ll fire up the data for
the stations
of the 14000 + stations we use
25 are below sea level. 75% of them are more than 200km
from a coast. One is within 3km of the water.
Steven,
Reconciling the sat records to the land records is mostly what I had in mind. Is it possible to get any useful data for areas that are not covered in the surface station record, to fill in the blanks? An analysis that combined sat with surface station data to get full coverage would be interesting, to me. If it wouldn’t make any difference and I am barking up the wrong tree, tell me to stop.
I just have the sneaking suspicion that the surface station record is inferior to the satellites, because of coverage and siting issues. Of course, I might be suffering from my own issues. I am also wondering if cleaning up the air/aerosols over cities (particularly in the developed world), since 1950, had any significant effect on temp trends in urban areas.
Zeke,
They should have independently realized their error, before they launched their PR blitz. My guess is they were tipped off to their goof, by a citizen auditor. Maybe Jeff Id, or Steven Mosher? It will be interesting to see what other errors they admit to.
Mosher
SRTM might be problematic cause it only goes up to 60 N and G top is of too low a resolution for you I imagine?
I think that comparing purely the altitudinal differences is very difficult without incorporating biomes into the issue ie that 10 for tundra would be a more true test rather than testing the whole database… you could afterall be looking at 10 in a completely different one.
Don Monfort (Comment #86968)
December 9th, 2011 at 4:35 pm
“An analysis that combined sat with surface station data to get full coverage would be interesting, to me…
I just have the sneaking suspicion that the surface station record is inferior to the satellites, because of coverage and siting issues.”
the SAT and the TLT measurements are different thinks in so many different ways that eventhough you could probably find analysis to reconcile the two, one would have to wonder about exactly what is the output. Lets remember too that the Satellite data does not have full coverage. In fact it does pretty similar to what GISS does in the high Arctic 80+ (ish) and does not include much of the Antarctic continent.
Finally we also have to consider that the satellite record is the one that has been show to have significant discrepancies in the past so I am wary of using it as the defining source of accuracy. Issues are still being recognized now (Zhou et al. 2010 (or 2011)) which show biases in the way the TLT record is currently being produced…
Now for me an interesting way of taking it would be to look at Land Surface Temperatures across the planet using MODIS/Landsat and to automate monitoring of that. Looking at averaged trends using those datasets and comparing to the station data would not be directly comparable because of the thermal inertia of the land surface but it would tell us things about the UHI. Remember we measure SSTs and the oceans tend to warm slower the air anyways whereas land has a faster response. Compare subsets of completely rural locations using land surface temperatures (like tundra etc) and compare trends to other regions etc…
Here Carrick
http://dl.dropbox.com/u/52801067/Altitude.pdf
and log of the distance to coast ( +1km to keep log sane )
( abs handles those cases where the station is 1km
in the ocean –mislocated )
http://dl.dropbox.com/u/52801067/LogCoast.pdf
Don, on BEST
I think it had something to do with a simple processing error.
I spent a few minutes with the Author. I’m headed over there next week to exchange some ideas and maybe get started on a project.
Robert
“SRTM might be problematic cause it only goes up to 60 N and G top is of too low a resolution for you I imagine?”
curses I forgot that about SRTM. crap one day I started downloading 64000 folders and just gave up, hoping for a better solution.
geonames has a jason interface, but I’m limited to 5K look ups per day.
G top might work ( remind me the resolution– brain fade )
Biomes? did you see my stuff on Tundra? That was pretty cool.
After reading Imhoff ( he is very helpful dude ) I took the
Olson 2001 and reclassified it using Imhoff’s schema. Tundra popped out as the biggest trend. but then you knew that. Still
it was cool to verify that stuff was making sense
ya Robert, we have a totally new turn key interface to Modis LST.
Also, looks like Landsat has a big project to make a bunch of old stuff
available in common formats. I sat through the presentation and I recall that some form of surface temp is on the list.
The biggest issue is retrieval and storage and look up.
I should probably define a workflow where I use the station locations to download the granules that I need and then process from that. issue is also picking the time of day, cloud free etc etc etc.
nasty effin problem.
Terabytes.
Also, as I have found out over on WUWT few people get the difference between LST and SAT.
Gtopo is 1000 m if I recall correctly. ASTER is available too at a 30 m resolution. Can’t remember if its got all the land mass covered but i’m pretty sure it has most. It’s freely accessible via NASA and through another site (i’ll have to look it up). Anyways the entire GLOBAL dataset is available and was created using an automated software (SilCAST) I believe. The stuff comes in at 30 m but I have issues with its accuracy in Alpine Environments where much of my work is focused.
There was an interesting study done by some colleagues (Nuth et al. 2011, Cryosphere) which looked at some of the biases in these products and created an algorithm for correcting them if you have a higher quality product but I digress…
Anyways the issues for me would not be issues for you. The dataset is obviously huge but you could take it and automatically resample the whole ASTER gDEM to 100 m or something if space is a real concern. The high latitude coverage is certainly a plus.
It is interesting to see how biomes control these things of course. I think some of the paleo work using Pollen really captures the different dynamics associated with different vegetation types and their climatic sensitivity a bit too. Actually that reminds me of a paper or two i’ll have to look up that are recent and quantify the last 1000 years in NA from pollen (Viau et al 2011 in Climatic Change maybe?)
Anyways we’ve talked before about the MODIS work and I didn’t really get around much to starting at it but in the new year I might because I find that stuff the most interesting. Remote sensing systems with the LST is such a useful tool for all of these analysis. I know that Landsat has its archives accessible through GLOVIS and much of this goes back pretty far but the challenge with landsat is of course that it doesn’t get frequent enough regional passes for the type of work I am suggesting and furthermore that Landsat 7 ETM+ stopped working after 2003 (still works but there’s striping).
There are of course other alternatives that could be considered but the cloud free is definitely a big issue for a lot of sensors. If I recall there is a paper (Duguay et al. 2007 maybe?) which does some interesting analysis on LST using MODIS in Canada.
Re: LST vs SAT
I do wonder about what sort of trend difference you would see though… i mean it is obvious that there would be some but one would think that it wouldn’t be much different from the SAT and SST… perhaps somewhere in between?
Thanks for the histograms Steven. You know that’s how I roll. 😉
There are two problems with coastal locations. One is the direct marine boundary layer issue. The other is that many western US coastal locations have nearby coastal ranges, and at night time there are density currents associated with that…
The mountain valley is a similar problem. Again the easiest solution is probably to make geographical cuts.
The ideal location for a temperature sensor is on a flat plain well away from any discontinuities (regions with “good fetch”). That’s if what you are trying to tease out is UHI as opposed to UHI contaminated with other effects….
Robert
LST versus SAT.
I think the biggest issue would be getting the right time of day. I know in imhoffs study he pick 130am and 130 pm. That rather maximizes UHI and is hard to compare to Tmax and Tmin.
for trends it might be ok? need to think
Steven Mosher:
If it’s not, then the comparison of satellite temperature and ground temperature is compromised.
Gave you a wrong reference earlier Mosher
Hachem S, Allard M, and Duguay C. 2009. Using the MODIS land surface temperature product for mapping permafrost: an application to northern Quebec and Labrador, Canada. Permafrost and Periglacial Processes 20, 4: 407-416.
I just knew duguay was on it but forgot which author he was. Interesting study i’ll shoot you an email with it.
Ok thanks robert
Maybe I could snake his data
Steven Mosher,
Although the question was for Zeke thanks for your answer. What I gather is that you did NOTHING to attribute all or part of the excess trend between stations to UHI. You just ASSume that is what causes it.
Zeke,
Thanks for your answer. It does not tell me how you were able to separate the GHG trend, the solar trend, and the UHI trend plus possibly others.
Is there nothing else??
Kuhnkat
It is obvious.
you have two stations. one is urban the other is rural
The urban warms faster
it is clear that since they both exist under the same sun
and same atmosphere that the urban trend is due
to solar forcing and the rural trend is due to
GHG forcing.
Actually, we interviewed all the molecules and they agreed
So its settled
You can’t talk to skeptics like that Mosher. It is condescending, bad manners, and elitist. You need to encourage skepticism.
Goddard
Stop using Bugs name.
I find this debate particularly non-sensical as “heat” is a thermodynamically ill defined concept.
And a site’s “temperature” is then presupposed to be a measure of such “heat”.
Any study of sites, and larger furcate constructs (earth) should include humidity as well. A humid site heating 0.1 degrees is worse than 10 dry sites heating 1 degree, for Gaiia.
A conclusion on sites behaviour without taking into account their humidity, is like selling a cure against coughing: Arsenicum. It stops the cough, alright!
Have nice day
87009 – Did the molecules at the cooling stations share the same opinion?
phooniethewee–
I’ve let your comment through so I can respond and let you know rules I have made for you.
I find people who use throwaway email addresses like those from http://www.yopmail.com/en/ who try to post scientific sounding comments using phrases like “thermodynamically ill defined concept” generally say things that make no sense. Your comment posted with a throw-away address is no exception.
I suspect you know your comments are technical zeros and that is part of the reason the every changing sock-puppets posting from IPs associated with btcentralplus.com resort to numerous silly sounding pseudonyms and a large variety of throwaway email addresses.
Please do not comment here any longer under your current pseudonym or any other sock-puppet names.
I am imposing a special rule on you: Whoever you are, as a condition for commenting,
* you must use your real name and credentials.
* you must use a real non-disposable work related email address or use an email address associated with a domain name you control that permits me to email you and which you can use to respond. (That is: for you, gmail, hotmail etc. are not permitted. But if you own “phooniethewee.com”, an email ending with “@phoniethewee.com” will be acceptable to me. If you work for @company.com, using your email from that company would be acceptable. I will accept posts from email from your_account@btcentralplus.com if you are retired and have no work related email.)
* I will only let your comments appear if you first send me a message using the contact lucia page and I verify that your email address works and you make a plausible case that your are using your real name to post.
curious. yes.
Phooie. Given that climate science has decided to use an average of air temperatures at 2m as a metric of global warming, and given the fact that cities
to have higher SAT than rural areas surrounding them, the question is how much of this bias is present in the record
We might argue that OHC is a better metric and indeed it is.
We might argue that heat is a better metric and indeed it is.
But that is not the issue we address. We address the UHI bias
in the SAT record. Changing the subject, suggests a lack
of intellectual honesty. Makes me think you are a sock puppet,
who perhaps is hiding in a hole he has doug before.
How many more 0.0C’s or less (november 2011 LTL RSS), do we have to have before most on this site stop being even lukewarmers?
http://icecap.us/images/uploads/AMOPDO_UAH.jpg
Lucia,
I always thought “wild hair” referred to rabs not follicles. ‘been thinking of migrating email address from aol to gmail. You suggest that the sock puppeteer not have an gmail address. Why not?
j ferguson–
Because he’s been a sock-puppet and that ISP has posted numerous comments ranging from obnoxious to just ignorant, I suspect he wants sooooper-doooper anonymity and specifically wishes to avoid people knowing he posts this garbage. So, I prefer an email that makes him potentially or easily tracable. Anyone can set up numerous gmail addresses. So, while gmail addresses aren’t throw away, I prefer he use a traceable one.
BTW: He has submitted another comment without remotely complying to the requirements. Obviously, I’m not going to approve it. Or… maybe I should. 🙂
More generally, I think gmail is great. They have the best spam filtering in the world, and you can even route your domain name email through gmail to get the spam filtering.
j furguson,
“I always thought “wild hair†referred to rabs not follicles.”
.
Well, maybe it refers to hair follicles on a bad-hair day. 😉
Lucia,
“I suspect he wants sooooper-doooper anonymity and specifically wishes to avoid people knowing he posts this garbage.”
Maybe it is Robert on a ‘bad hair’ day. 🙂 Whoever it is, they seem mostly to want to have you waste your time.
I don’t know if anyone else has asked this question on this thread.
Spatially on land Rural has many many times the area than urban areas and yet the “bias” is toward the many more urban records.
Should the overall temperature be weighted to redress the balance?
SteveF:
You need to be careful; there are two Roberts posting here now! The one posting in this thread appears to be the newer one, easily distinguished from the other Robert in that this one is (a) polite and (b) appears to know what he is talking about
SteveF: (the two Roberts) I was really puzzled, but not having read comments of the virulent Robert that carefully, I’d imagined that it might be possible to be technically competent and thoughtful and yet have strident views on some aspects of the debate.
Good to see there are two guys. maybe it’s time for an “andrew” fix.
I’m certain there’s two guys, unless “virulent Robert” has both gone back to school and had a personality transplant.
Robert, the guy you really don’t want to be mistaken for is the first commenter here, and this is typical of him: http://rankexploits.com/musings/2011/ceres-and-the-shortwave-cloud-feedback/
Interestingly you can distinguish the two because “virulent Robert’s” name is hyperlinked to the “idiot tracker” blog…
I can confirm there are at least two Roberts. I thought about asking the recent Robert to add a letter or something, but the other Robert hasn’t been showing up.
Lucia, Phooie who-ever has shown me a new word. “furcate”. I’m going to use it in my day to day conversations. That will prove to every one that I AM smart!
Duh,
Ray
AC
that is the whole point of doing a rural only estimate.
kuhnkat,
We simply look for differences in trends correlated with urbanity, in an attempt to look at urbanity-related biases in the temperature record. Unless solar and GHGs exhibit urbanity-correlated forcings (which would be an odd result indeed), they are not in scope for our current project.
“Our estimate for the bias due to UHI in the land record is on the order of 0.03C per decade for urban stations.”
.
Way too low. This is Zeke pretending to be reasonable by giving up a certain amount of UHI while at the same time maintaining the “AGW is alarming” myth. The basic problem is sorting stations into two bins and assuming you can get the magnitude of UHI by differencing the bins. A much better method is to look at satellite images that show you the difference between an urban area and it’s unpopulated surroundings.
.
From Zang, Imhoff, et al.
.
“Globally, an average of 3.8 °C UHI is found in cities built in biomes dominated by forests; 1.9 °C UHI in cities embedded in grass-shrubs biomes; and only a weak UHI or sometimes an urban heat sink (UHS) in cities in arid and semi-arid biomes. Overall, the amplitude of the UHI is negatively correlated (R = –0.66) with the difference in vegetation density between urban and rural zones represented by the MODIS normalized difference vegetation index (NDVI). Globally averaged, the daytime UHI amplitude for all settlements is 2.6 °C in summer and 1.4 °C in winter. Globally, the average summer daytime UHI is 4.7 °C for settlements larger than 500 km2 compared with 2.5 °C for settlements smaller than 50 km2 and larger than 10 km2. The stratification of cities by size indicates that the aggregated amount of ISA is the primary driver of UHI amplitude, with variations between ecological contexts and latitudinal zones. More than 60% of the total LST variance is explained by ISA for urban settlements within forests at mid to high latitudes. ”
.
http://pubs.casi.ca/doi/abs/10.5589/m10-039
There are still some very distinct issues that are not addressed. “Rural” is a problematic term since most “rural” landscapes are still anthropic landscapes subject to direct anthropic effects that will have direct effects on microclimate. One example is the steady conversion in California of grassland to grazing land to dry crop land to irrigated crop land (from various truck crops to rice paddies) to irrigated orchards and vineyards. This process also involves transport of vast amounts of water from water-rich to water-poor regions, e.g. from the Sacramento Valley to the Southern San Joaquin, which is effectively desert climatically. So, even highly stable stations such as some of Anthony Watt’s examples from the Sacramento Valley that adhere closely to standards are highly likely to have experienced trends measurements imposed by local shifts in agriculture practices and products. We have a two tiered system of classifying weather data but in fact, nearly all stations where such data is collected, rural or urban, are extremely likely to have experienced anthropic “climate” effects.
From the poster
“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”
Can one of the authors just comment on this? I don’t know the details of the homogenization process but will it always be the case that urban data will move to match rural changes. Is this specifically set in the homogenisation process? Where urban stations dominate why would rural stations not look like the data with the oddity and move to be inline with urban stations?
I had another question.
Do the graphs under the title “Trends in Urban Warming via Other Urbanity Proxies:” suggest that UHI affect is generally higher in areas that have lower levels of urbanisations? Does this not describe a larger portion of the global land surface area?
Tilo Reber (Comment #87051)
Thanks for the link.
“Globally, an average of 3.8 °C UHI is found in cities built in biomes dominated by forests; 1.9 °C UHI in cities embedded in grass-shrubs biomes; and only a weak UHI or sometimes an urban heat sink (UHS) in cities in arid and semi-arid biomes..”
Can someone shed some light on what UHS means in this contex.
Is this trying to tell me that urban cities in arid is cooler than the outlying arid rural?
And that cities with trees are warmer than without?
Zeke,
as I have pointed out to MoshPup a number of times, until you have quantified the amount of UHI for individual stations and correlated it to increases in population/asphalt/concrete/waste heat… you have no idea when a area is actually experiencing what amount of UHI. Additionally you are comparing inhabited to inhabited so you can have a portion of the UHI masked. That is, the non-UHI trend for both stations could be .17c/C, the UHI for the first station .08c/C, the UHI for the second station .05c/C giving you .25c and .22c for a difference of .03c with .05c UHI masked.
As I pointed out over on the Curry thread for Best, the idea of negative UHI is quite reasonable also. Our flat temps could be the result of a lot of stations having minimal to negative UHI due to increases in energy efficiency, slow or no growth in population/area, increased parks/trees/shrubs… We know Europe is negative population growth and the US is only growing due to immigration.
With no way of quantifying the amount of UHI at each station over what period of time you really don’t know what your result is telling you. You can’t even be sure that the underlying trends are the same without having some idea of whether wind/precip/cloudiness is close between the stations.
Ed Forbes:
I think I can.
In some cases, it has been found urban areas are cooler than the surrounding areas. It’s especially (but not universally) observed in arid regions, primarily in the morning. I’m not sure what the physical explanation is for it, but it’s not something which is unheard of.
No. It’s talking about the areas the cities are found in. In other words, cities built in forested areas have about twice as much UHI as cities built in grasslands.
HR: “It is likely that homogenization will further reduce the observed UHI-related bias, as many urbanity biases are detectable”
.
We’ve been over this with Zeke before, HR. He refuses to acknowledge that homoginization doesn’t remove UHI bias, it simply spreads it around.
.
ED: “Is this trying to tell me that urban cities in arid is cooler than the outlying arid rural? ”
.
No, Ed, they are only saying that the UHI difference between cities and rural is smaller in arid areas than it is in forest areas. But in virutally all cases, UHI in cities is positive with respect to the surrounding rural areas. The difference is due to the effect of plant respiration for cities in areas with a lot of biomass.
Tilo
“.
“Globally, an average of 3.8 °C UHI is found in cities built in biomes dominated by forests; 1.9 °C UHI in cities embedded in grass-shrubs biomes; and only a weak UHI or sometimes an urban heat sink (UHS) in cities in arid and semi-arid biomes. Overall, the amplitude of the UHI is negatively correlated (R = –0.66) with the difference in vegetation density between urban and rural zones represented by the MODIS normalized difference vegetation index (NDVI). Globally averaged, the daytime UHI amplitude for all settlements is 2.6 °C in summer and 1.4 °C in winter. Globally, the average summer daytime UHI is 4.7 °C for settlements larger than 500 km2 compared with 2.5 °C for settlements smaller than 50 km2 and larger than 10 km2. The stratification of cities by size indicates that the aggregated amount of ISA is the primary driver of UHI amplitude, with variations between ecological contexts and latitudinal zones. More than 60% of the total LST variance is explained by ISA for urban settlements within forests at mid to high latitudes. â€
”
There are several things you do not understand about Imhoff’s study.
1. In this study he looked at cities that had 10sqkm to over 500 sqkm of built area.
2. The amount of UHI he measured went down linearly as a function of built area. so 3.8 C is an average of urban areas
from 10 sq km to over 500.
3. 50% of our urban stations would not even qualify as urban
for imhoff, that is, 50% had less than 7 sq km of built
area. 75% had less than 30sq km of built area.
4. Imhoff measured LST. LST is higher than SAT and more variable
than SAT.
5. The time of day he took the measures maximizes the UHI
So we show .039C per decade between 1979 and 2010
You do recall that the total bias cannot be more than .08 to .1C per decade.
Kuhkkat
‘as I have pointed out to MoshPup a number of times, until you have quantified the amount of UHI for individual stations and correlated it to increases in population/asphalt/concrete/waste heat… you have no idea when a area is actually experiencing what amount of UHI. ”
And as I told you I’ve done that work.
It’s also important to note there is a difference between bias in temperature offset and bias in temperature trend. We care about temperature trend and not temperature offset with respect to measuring climate change.
If you have Person A analyzing temperature trend bias, then Person B saying “yes but” and pointing to a measurement of temperature offset bias, you have a basic disconnect, usually associated with Person B not understanding math the or science well enough to understand the distinction and usually failing the “know thyself” self-test. (The alternative, that they are being willfully deceptive, is less pretty than just unrecognized self-ignorance.)
kuhnkat:
You’re such an “adult”.
Ed
“Tilo Reber (Comment #87051)
Thanks for the link.
“Globally, an average of 3.8 °C UHI is found in cities built in biomes dominated by forests; 1.9 °C UHI in cities embedded in grass-shrubs biomes; and only a weak UHI or sometimes an urban heat sink (UHS) in cities in arid and semi-arid biomes..â€
Can someone shed some light on what UHS means in this contex.
Is this trying to tell me that urban cities in arid is cooler than the outlying arid rural?
And that cities with trees are warmer than without?”
First off folks should not just read abstracts. Lets see if I can shed a little light here
“Globally, an average of 3.8 °C UHI is found in cities built in biomes dominated by forests; 1.9 °C UHI in cities embedded in grass-shrubs biomes; and only a weak UHI or sometimes an urban heat sink (UHS) in cities in arid and semi-arid biomes..â€
If you build a city in an area dominated by forest that city will have a differential temperature from the surrounding forest of
3.8C. This measure is actually a measure of the land surface temp and not the air temperature. LST can be anywhere from 0C to 11C warmer than the air temperature. We look at air temp at 2meters.
The UHI measured in there study is measured at 130AM and 130PM for 8 days. we look at averages over all the days. For cities located in forests the main driver is the change in evapotranspiration. If you build a city in an arid/semi arid place there is very little UHI. Again, its evapotranspiration and perhaps albedo driven. Same with grassland.
The other thing is that these studies tend to look at urban areas that are much larger than the urban areas in the temperature record. Also, they measure UHI on cloudless, rain free days. Basically the days when UHI is the worst. A few clouds, a little wind ( <7m/sec) and UHI shrinks. Finally we look at trends over 30 year period and not merely the worst UHI days.
Steven Mosher:
It would also be helpful to the general public if there were at least a preprint version of these papers made available. Not everybody has an overhead account to charge these papers that come from a myriad of journals (or a university library account that covers 90% of the journals to start with).
yea carrick, in my case the author was kind enough to supply me with the papers because i was using his data.
“It would also be helpful to the general public if there were at least a preprint version of these papers made available.”
It would be good if there was something like PubMed for climate papers and also if there was funding to make climate change related papers all publically available regardless of funding source.
The other comment I’ll make is a somewhat technical one. LST measured by satellites is available only under clear sky, high pressure conditions. This circumstance is unfortunately the only one under which one expects a strong correlation between LST and SAT to be present.
There have been studies suggesting that SAT can be used as a proxy for LST to “fill in” when LST is not available. I personally don’t find that very convincing… because SAT are LST are strictly related only when the mean-wind sheer is near zero. In other conditions, mechanically driven turbulence can substantially affect the mean SAT, because a) there is usually a nonzero lapse rate (gradient in mean temperature with altitude) and b) because wind-shear causes mixing between the layers, so that wind-shear driven turbulence can be a dominant source of heat exchange compared to local land surface temperature.
In any case, it is important to recognize that the physics that drives land surface temperature are different than the physics that drive surface air temperature, and while the two are related, that relationship “gets complicated” in all but the simplest of cases (e.g., the one case MODIS can actually image).
Nyq:
Given that the funding is in the billions for climate change research, and the constant bitching that the scientists have about the “breakdown in communication with the public” (no lie, you should have seen all of the AGU sessions dedicated to “public relations”—I didn’t because I just don’t like the taste of Pepto-Bismal, a required medication for holding your lunch while watching these types of sessions), you’d think they’d recognize that public access to their research might have some beneficial consequences.
Kukhkat and others.
After assessing the bias between an all rural series and an all urban series, I did some additional work
I’ll show you all some of that work so you can get an idea about how our proxy works ( differently than BEST and others )
1. For a radius of 11km around the site I count the number of
500m grids that are “built”. in the present work this is turned
into a fraction ranging from 0 to 1
2. 0 percent is rural, anything > 0 is urban. The urban thus have
a measure of urbanity up to 1. I can also turn this into sq km
3. We then look at the difference in trend between the rural and
the urban. It’s .039C per decade or about .12C over the
1979 to 2010 time period.
To see if I had something that made sense I did a side study on the
US only. The approach here was as follows.
A. do a rural versus urban
http://dl.dropbox.com/u/52801067/UsaUrbanRural.png
Here you see that the UHI bias in the USA is slightly lower
than the ROW. You also see that the mean “urban percent”
is 9.61% That means the mean urban area is around 40 sq
km for the urban zones.
B. Now I change my definition of urban. Rural is still zero
percent, but urban will require more than just a few
pixels of built. For the next case I require that MORE THAN 1% of
of the pixels in the 11km radius be urban.
http://dl.dropbox.com/u/52801067/UsaUrbanRural01.png
And what we see is that the mean urban area moves up to
16% ( about 60 sq km) and the bias moves up to about .04C
Just to be clear, urban areas that have less than 1% of the pixels
as built are dropped out of this approach. We are comparing rural to stations that are more urban.
Next urban was defined as having more than 5% of pixels in the zone as built
http://dl.dropbox.com/u/52801067/UsaUrbanRural05.png
Here we are comparing rural to urban stations that have more than
5% of the pixels in the 11km radius as built. As you can see when I set a 5% threshold the average station has 28% of the pixels as built
or around 108sq km
If you want to see what that looks like
http://dl.dropbox.com/u/52801067/StationMap05.png
Next we move up to 10% threshold. here rural is 0%
between 0 and 10% is dropped
10% and greater is defined as urban
http://dl.dropbox.com/u/52801067/UsaUrbanRural10.png
What we see here is that the mean urban site is now surrounded
by pixels that are 36% built. The bias goes up. At this level
urbna area is around 140 sq km.
Changing the cut off to 15%
http://dl.dropbox.com/u/52801067/UsaUrbanRural15.png
And what you see is that the bias thresholds. That is we increased the urban area to a mean of 43% ( 166 sqkm) and the bias trend did not increase. This is consistent with the results of Imhoff who
found that UHI increases as a linear function of urban area up to a threshhold and then stopped.
I trust that is all clear.
When we compare rural 0% built pixels, to urban, where urban is
defined as having ANY built area within 11km we get a bias of
.024C per decade ( in the US ), If I tihten the filter for urban and require that a urban area have more built pixels, the bias goes up.
Note, I do not change the definition of rural. “periurban” sites
are just dropped, although I could analyze them separately.
At the limit if we compare the most rural to the most urban ( see other sensitivities) you are talking about a .1C difference in trend per decade. However, the bias IN THE RECORD is less than this.
Its less than this because the record has rural (0%) and a whole range of urban stations from small (1 sq km) to large 400 sq km+
50% of the urban stations have a small amount of built area (<7sqkm) 75% have less than 30sq km of built area.
This fact leads to some pretty gross misunderstanding because most people compare the two extremes the most urban versus the most rural. That gives you a look at the peak UHI but not the UHI actually present in the record. the UHi actually present in the record is the average of small area UHI and medium area UHI and large area UHI.
Carrick will note that the urban stations in the US tend to be located on the Coast. That also has the effect of lowering the
trend. Also, the west coast urban tends to be located in semi/arid, grassland areas, so you have lower UHI as a result.
Re: 87095
Since you did not provide labels, I am assuming that red is urban and blue is rural. If so, why is urban colder before the early 1980s and then warmer afterward? What caused urban behavior to change?
Tilo:
“No, Ed, they are only saying that the UHI difference between cities and rural is smaller in arid areas than it is in forest areas. But in virutally all cases, UHI in cities is positive with respect to the surrounding rural areas. ”
This is actually not the complete story. It depends upon the season ( they only looked at summer winter) and the sky conditions ( they can only look at cloud free days) and it depends upon the city size. in some areas there is actually an urban heat sink effect where urban is cooler than rural. Again, it helps to actually have read the papers. The most important things that come out of these papers is not the actual estimate in differences between LST, but rather the following.
1. You have to control for the embedding Biome.
2. You have to look at area. population is the wrong
metric ( unless its population per km )
3. UHI varies linearly with area of built area up to a threshold
4. UHI falls off with distance from the urban core in a linear
fashion.
I’m still working on meshing our analysis with Imhoff, but at first glance his definition of rural is going to be consistent with ours.
That is if I call it rural so will . he defines rural as being 15km
from the 25% ISA isoline. that is probably less strict than my rule.
JR.
Our analysis is focused on one question.
1. Does an all rural database warm less slowly than an all urban
database.
The answer to that question is yes. Rural warm less by about .03C per decade.
Your question is about determining the causation in subsets of the data. Totally different issue. Sorting that out is a very complicated issue’ So, if people have theories they would like to test that’s the approach I would take. Mining the data for
explanations willy nilly is not a good approach.
Steven Mosher,
A pixel is 500m x 500m? This is the resolution of MODIS I understand? For a pixel is counted as built, what is the real percentage of urbanized area required on a pixel?
Carrick,
“Pepto-Bismal”
.
Is this the deep-south pronunciation of ‘Pepto-Bismol’? 😉
.
Actually, I don’t mind the taste so much…. better than the alternative.
Thanks for the comments on Zang, Imhoff, et al.
So if either the urban or the outlying area undergoes a change, there will be a change in bias vs UHI.
.
the records of the Calif central valley, Arizona, Texas, others…. all undergo major changes in time as the cities grew fast and 100’s of thousands of arid lands were coveted into irrigated crop lands with the dam building spree in the west. Taking the CA central valley as an example, it also does not take in the change in crop use going from cotton and grapes to almonds and tree fruit. This area has converted 10’s of thousands of acres of desert into “forest” over time. Desert to shrub (cotton, grapes) to forest ( almond, tree fruit).
.
As this started mainly in the 1950’s through the 1970’s at least for the CA central valley, it could not be as simple as adding a fixed number to adjust.
.
Will be interesting to see how all this shakes out in the end.
Re: 87099
My comment was posted in reference to your comment 87095, in which you state At the limit if we compare the most rural to the most urban ( see other sensitivities) you are talking about a .1C difference in trend per decade.
If you don’t know why urban vs. rural behavior changed, then you don’t know if your trends are good. If you don’t know if your trends are good, then you don’t know what the actual difference in trends is.
Phi,
to count as built a pixel must have 50% of its area as built
and to count as urban you must have two contiguous built pixels.
Like all classifications the modis classification has two error rates.
A commission error and an omission error. The commission error rate is the rate at which Modis says the pixel is Urban, but it is actually not. This error rate is 2%.
The omission error rate is the rate at which Modis says a pixel is not urban and it actually is. This is the omission error rate. The omission error rate is the error rate that I control for by using multiple proxies from different sensors. If you merely use Modis ( like BEST ) then you will find that some urban creep into your rural class. ( as I recall maybe 10% ) For example, there were about 200 airports that made it through the Modis filter. Those are removed. In addition you will have sites where there are sparse small buildings, all less than 500m. These however show up in other proxies: they show up in nightlights and they show up in ISA. So,
they are removed from the rural class.
So, if one just uses Modis Built as a filter you will get a leakage of
some urban sites into the rural. These are removed by additional
filters we employ ( no airports, nightlights < 30, ISA 0) have a bias relative to the stations with no population, then you see that they dont. In short the trends at rural locations is explained by other variables– like latitude and distance from coast and land cover.
Finally, we see, like imhoff, that there is a linear relationship
between the urban area and the bias. Imhoff sees this as
.11C per every 10% of ISA. 75% of our rural stations have an
ISA figure that is less than 1%
Note that this ISA figure is derived from 30 meter data and landscan population.
JR:
“If you don’t know why urban vs. rural behavior changed, then you don’t know if your trends are good. ”
that clearly doesnt follow logically. The trends are merely the result of a calculation. The question is does an all rural network have a lower trend than an urban network.
At any given time since the difference in trend is small you can see random effects. In fact, we EXPECT that there will be periods were one warms more than the other and vice versa. The small size of the effect makes this highly likely. Now if the effect was large and we saw reversals, then you might require an explanation. But where the effect size is small and where there is a random component to the trends, you can expect reversals that are fundamentally random in nature.
In short. If we compare the satellite trends over land to the surface record trends we can see that the trend we show for an
all rural network is very close depending on your choice of
satellie records ( within .02C to .05C) per decade, not counting
error bars. That pretty much answers the question we set out to answer. UHi exists, its not that big. If you want to deny that UHi exists, have at it. If you want to argue that UHI is more than
.1C per decade, show your work, you’ll have to face the satellite record and show why that is wrong as well.
Steven Mosher,
Thank you for the explanation. Don’t you think missing disturbances from nearby urbanization, typically from thermometers shelter to a few hundred meters ? Do you plan to address this issue ?
Phi
“phi (Comment #87107)
December 13th, 2011 at 3:31 pm
Steven Mosher,
Thank you for the explanation. Don’t you think missing disturbances from nearby urbanization, typically from thermometers shelter to a few hundred meters ? Do you plan to address this issue ?”
If you are talking about micro climate issues then I would say the following
First, if we take the satellite trend over land at this time ( .18-.22c per decade) to be a bias free estimate, then can compare the land record to that and see that there is a possibility of bias in
the land record of something on the order of .04C to .1C per decade. This possible bias in the land record will be the sum of
all biases:
1. Microsite
2. UHI
3. Sampling
From looking at a variety of studies ( regional ) to our global
study, we can see that something on the order of .04C of this
bias can be attributed to UHI. For microsite, I would say the effect
if it is ever found will be smaller than UHI. Fall et al (Anthonys paper ) confirmed what I’ve been saying for the past years. Micro site bias is going to be very hard to find. Its almost an order of magnitude smaller than UHI. If I had to focus on anything it would
be carricks observation. The latitudinal bias is on the order of the UHI bias. Simply, trend changes with latitude. This is polar amplification. as you move from the south pole up trends get larger. The stations are overweighted at 30-60 degree north.
That means the distribution of stations with latitude does not match the distribution of land area with latitude, consequently there is a bias toward higher trends being in the record. Funnily the warmth you see in the 30s-40s is driven by sampling bias with northern latitudes being overweighted.
i suspect that if we we could do a temperature series that was
corrected for this bias and the UHi bias that we would match
the satellite record to a nats ass. or, if we can build a record that has stations that have the correct frequency distribution of latitude you would see the same thing.
But the bottom line is this. The record post 1979 is pretty darn good despite all its flaws.
Ed
‘
As this started mainly in the 1950′s through the 1970′s at least for the CA central valley, it could not be as simple as adding a fixed number to adjust.
.
Will be interesting to see how all this shakes out in the end.
###########################################
to address the land use change I’m also looking at a variety of data that cover the change of land into crop land.
This currently outside the scope of what we are looking at, but
I have the data so I play around with various hypothesis.
For example. I take all urban stations. I reduce this to only
long stations ( 30 years of data ) Then I calculate trends.
Then I do a series of regressions to see if I can explain the trends as a function of other data : latitude, coastal location, urban area,
biome, area of cultivated land, change in cultivated land, amount of blue water irrigation, etc etc.
here is what I generally find. You can pick extreme examples
( like central california ) and find small effects. When you try to
find the global effect these factors fail to have any explanatory
power. Lots more work to due there and If I was done with it it would be a few papers worth of data. For now, nailing down
the UHI is enough of a job.
HR
‘Do the graphs under the title “Trends in Urban Warming via Other Urbanity Proxies:†suggest that UHI affect is generally higher in areas that have lower levels of urbanisations? Does this not describe a larger portion of the global land surface area?”
No the purpose of those charts is as follows.
We selected Modis as a measure of urbanization. hansen uses nightlights, Imhoff uses ISA, others have suggested population growth as proxies.
I have criticism of all those proxies which is why we selected Modis. However, We found it important to place our results in the context of other proxies. What if I used Hansen’s Nighlights as my proxy, how does that change my answer? what If I used Imhoffs
Isa? what if I used population growth. What those charts show you is that the answer is relatively insensitive to the selection of your proxy. That doesnt mean I think Nightlights is better than Modis, I think Modis is better for a variety of reasons, but by comparing it to other proxies I’m showing you that my slection of a new proxy is not driving the answer. That is what all the sensitivities do.
So I come up with a proxy; 0 modis pixels within 11km
And then Zeke says; Show how the answers changes if you change the following
1. change 0 by increasing it
2. change 11km and vary it from 5 to 20
3. change the proxy you choose and then do sensitivities
on those.
Anyway, so that is the purpose of those charts, to illustrate that this new proxy is not driving the result in any crazy manner.
If I had to pick an alternative proxy it would be ISA. Nightlights
is tough to work with because there is no clear way of tying
the light level to anything else with physical units.
‘As I pointed out over on the Curry thread for Best, the idea of negative UHI is quite reasonable also. Our flat temps could be the result of a lot of stations having minimal to negative UHI due to increases in energy efficiency, slow or no growth in population/area, increased parks/trees/shrubs… We know Europe is negative population growth and the US is only growing due to immigration.”
This is wonderful speculation that inst really tied to any facts.
The issue is not population growth at a national level. In fact,
of all the things we look at population growth is the least interesting.
population doesnt cause UHI. changes people make to the surface cause UHI. Since I have stations that show negative population growth it’s pretty easy to test this. If people change the land and then leave, the effect their land change doesnt go away merely because people left. Now if waste heat was a huge part of UHI that might be different. But waste heat is not a principle driver.
Steve–
What is the principle drive? Asphalt? Concrete buildings? Etc? (Real question.)
I think conceptually, no matter what the driver, negative UHI trends due to depopulation are possible. For example: in Planet of the Apes, all cities eventually became overgrown with vegetation. I’m sure I could find other examples from dystopian Sci Fi. But I suspect there aren’t many places where that has actually happened.
lucia
as a possible example, in the aftermath of WW2, many cities in Europe were bombed to pieces…..afterwards, buildings were erected to create a highly urbanised environment with a lot of empty plots of rubble, which encouraged the growth of hardy plants such as buddleia. Now trees and parks are encouraged to soften the urban impact – as in central London, Cologne, Berlin, St Petersburg, Budapest, Prague ….together with irrigation of those green spaces where necessary. It is possible to conceptualise that after 1945 there would have been a strong increasing temperature trend that then faded once buildings were established and irrigated planting took hold. but this is just a hypothesis based on visual memories.
Lucia:
Surface properties, changing the heat capacity of the surface
material and then probably building height which is going
to change turbulant mixing. ( there is also a height/width
issue)
yes, if the cities turned back green then you would see the effect
Cities tend to have lower albedo than natural landscapes.
I might play abound with some urban energy balance models some day, but for now the plate is pretty damn full
Steve/Lucia – regarding principal drivers of UHI effect, it seems that surface and/or roof properties have a very large effect along with how much vegetation is left or allowed in place. Darker roofs/surfaces = greater UHI, less leafy vegetation (compared to surrounding undeveloped areas) =greater UHI. Interestingly, one observation I made from IR satellite images of NYC is that taller buildings appear to reduce UHI. Brooklyn, particularly areas of residential row homes, show much hotter IR levels than the high rise sections of mid-town and downtown Manhattan. The lower building heights in soho, the village, etc. also show greater IR levels than mid-town or downtown. The various NYC airports were way off the hook. It would be interesting to see if this observation held for other cities but I personnally haven’t looked into it.
lucia:
Irrigation has a big effect too.
This might be an interesting read.
Another interesting read:
Aerodynamic Properties of Urban Areas Derived from Analysis of Surface Form by Grimmond and Oke (1999).
Referenced by the prior article.
Mosher: “This is actually not the complete story. ”
This is the complete story:
“Globally averaged, the daytime UHI amplitude for all settlements is 2.6 °C in summer and 1.4 °C in winter. ”
Mosher: “1. You have to control for the embedding Biome.”
No, you don’t. You simply take all the cities in all the biomes, difference them with the surrounding rural area and average them.
Mosher: “2. You have to look at area. population is the wrong
metric ( unless its population per km )”
No, you don’t. Again, the idea was to get the average UHI effect for cities in a certain population range. Knowing that number we can get to a baseline UHI effect even if all other lesser population centers have no UHI effect at all.
“4. UHI falls off with distance from the urban core in a linear
fashion.”
Yes, so what?
Mosher: “It depends upon the season ”
No, it doesn’t. The summer UHI numbers are large. The winter UHI numbers are large. The spring and fall will be somewhere in between – all large.
Mosher: “3. UHI varies linearly with area of built area up to a threshold”
Where do you get that?
Mosher: “This measure is actually a measure of the land surface temp and not the air temperature. ”
Where do you get that?
“There are several things you do not understand about Imhoff’s study.”
We have already been over this issue, including the population issue. The result was that there is about .5C of UHI built into the global instrument record. If you can’t remember it, go back and read it. But don’t bother me with the same points that we covered before.
Carrick: “If you have Person A analyzing temperature trend bias, then Person B saying “yes but†and pointing to a measurement of temperature offset bias, you have a basic disconnect, usually associated with Person B not understanding math the or science well enough to understand the distinction and usually failing the “know thyself†self-test.”
Then again, maybe the person that knows math is just too dumb to realize that you get to a temperature offset by way of a temperature trend.
Tilo:
Well. You’re groking they aren’t the same thing. That’s a start.
For extra credit tell us what the trend is, for a period when the offset is constant.
Tilo:
Maybe Tilo is so busy being really clever that he doesn’t have time to look up what MODIS LST is.
Carrick: “for a period when the offset is constant.”
For extra credit, tell me why the guy that knows math would posit problems that have nothing to do with reality.
Here is another effort at determining UHI.
http://www.co2science.org/articles/V14/N50/C2.php
While still not definitive, as it doesn’t try and quantify the effects making up the statistical UHI it reports, I would say it shows that there is a lot more work to be done before anyone can claim they know how much UHI is in the record.
BobN,
yes there are many factors. If you look at studies of portand the number one factor was canopy cover which leads to urban cooling.
But in general the big players are surface properties and building height.
Moshpup,
thank you for the more detailed information on what was actually done.
Have you considered using shorter periods based on population growth statistics? That is, sectioning the record to see if there are high trend periods imbedded in the long term trend and how they correlate with population numbers?
You mention using pixels to determine buit area. Was this done for the beginning of the record and comapred to the end of the record to understand how much build out happened in the period??
I apologize if you said you did, I am in a hurry and am just skimming.
With anomalization, there’s always an offset by the way, Tilo. So you don’t need a trend to get to one.
I see kuhnkat thinks he’s still in grade school. What a clever boy.
Tilo
“Mosher: “This is actually not the complete story. â€
This is the complete story:
“Globally averaged, the daytime UHI amplitude for all settlements is 2.6 °C in summer and 1.4 °C in winter. â€
Mosher: “1. You have to control for the embedding Biome.â€
No, you don’t. You simply take all the cities in all the biomes, difference them with the surrounding rural area and average them.”
####################################
well even here its obvious that you have not read Imoff.
Imhoff reduced the biomes to 4 classes and disregarded
certain classes. basically because he was not studying UHI in
the record. He also disgarded cities that are in multiple Biomes.
So when you read the papers let me
know.
“Mosher: “2. You have to look at area. population is the wrong
metric ( unless its population per km )â€
No, you don’t. Again, the idea was to get the average UHI effect for cities in a certain population range. Knowing that number we can get to a baseline UHI effect even if all other lesser population centers have no UHI effect at all.”
Again this is wrong. The cities are selected based on the size of the urban areas and the ability to get enough passes from Modis.
“A minimum area of 10 km2 is used when screening the
global urban settlements to account for the potential errors
introduced by theNightlight ISA.”
“4. UHI falls off with distance from the urban core in a linear
fashion.â€
Yes, so what?
############################
Again, this matters when you start to classify what is meant by rural.
“Mosher: “It depends upon the season â€
No, it doesn’t. The summer UHI numbers are large. The winter UHI numbers are large. The spring and fall will be somewhere in between – all large.”
Again you show that you didnt read the papers. Spring and fall where not even tested.
The average amplitude of the UHI is remarkably asymmetric with a 4.3 °C temperature difference
in summer and only 1.3 °C in winter. In desert environments, the LST’s response to ISA presents an
uncharacteristic “U-shaped†horizontal gradient decreasing from the urban core to the outskirts of the city
and then increasing again in the suburban to the rural zones. UHI’s calculated for these cities point to a
possible heat sink effect. These observational results show that the urban heat island amplitude both
increases with city size and is seasonally asymmetric for a large number of cities across most biomes. The
implications are that for urban areas developed within forested ecosystems the summertime UHI can be
quite high relative to the wintertime UHI suggesting that the residential energy consumption required for
summer cooling is likely to increase with urban growth within those biomes.
“The amplitude of summer daytime UHI appears to be
related to several factors, including surrounding biomes, latitude,
and settlement size. On average, the summer daytime
UHI for global settlements located in forest biomes is 3.8 uC,
which is significantly larger than that in short vegetation
biomes such as grass–shrubs (1.9 uC) and in arid biomes
(0.2 uC) and semi-arid biomes (0.0 uC). In comparison, the
median size of the global settlements in forests, grass, arid,
and semi-arid biomes is 25.8, 27.3, 22.8, and 23.0 km2,
respectively. Thus the UHI differences are a result of differences
in the way in which vegetation regulates the water
balance between the atmosphere and the surface through
crowns and roots.”
“Mosher: “3. UHI varies linearly with area of built area up to a thresholdâ€
Where do you get that?”
Figure 3.
“We also note that the rate of change in LST as a
function of ISA is different for urban areas in forest biomes
versus those characterized by short vegetation (e.g., grass–
shrubs). For example, the slope of the linear fit is 0.11 (an
increase in ISA by 10% equals a 1.1 uC increase in LST) for
urban areas in forests, whereas it is only 0.07 (an increase in
ISA by 10% equals a 0.7 uC increase in LST) for urban areas
in grass–shrubs.It should be noted that the linear relationships between
LST and ISA are almost identical for the two sets of ISA
products in terms of slopes and explained variances
(Figure 3). However, the scatterplot indicates that the effects
of ISA on LST tend to saturate in dense ISA areas, and such
saturation is more obvious in the Nightlight ISA product.
Previous research has indicated that heat island intensity
increases with an increase in the height to width ratio, following
a nonlinear relationship in which the increasing rate
decreases as the height to width ratio increases (e.g., Oke,
1981). Dousset and Gourmelon (2003) also show a similar
trend of saturation between AVHRR LST and the percentage
of build density derived from 20 m resolution Satellite
pour l’Observation de la Terre (SPOT) imagery.
################
Mosher: “This measure is actually a measure of the land surface temp and not the air temperature. â€
Where do you get that?
from understanding what LST stands for and from the paper.
I’ve explained this before but you dont read.
L = Land
S = Surface
T = temperature
This is different more variable and HIGHER than SAT
S= Surface
A= Air
T = Temperature
“To characterize the surface temperature and the presence of
vegetation within the ISA zones, we use MODIS-Aqua Version 5, 8-day
composite (MOD11A2) LST with high quality control (Wan et al.,
2004) and 16-day composite NDVI (Huete et al., 1994, 1997) at
1km×1km resolution (both covering 2003 to 2005). LST’s from
MOD11A2 are retrieved from clear-sky (99% confidence) observations
at 1:30 PM and 1:30 AM using a generalized split-window algorithm
(Wan & Dozier, 1996). The coefficients used in the split window
algorithm are given by interpolating a set of multi-dimensional lookup”
“Many observational studies have estimated the magnitude
of UHI by comparing ground-based observed air temperature
in urban and rural weather stations (e.g., Oke, 1973).
In general, the air temperature defined UHI is found to have
a strong diurnal cycle and is more important at night as
compared with ground surface temperature defined UHI,
which is more important at midday and is postulated to be
primarily due to the difference in heating and cooling rates
between air and ground surfaces (e.g., Roth et al., 1989). In
general, the UHI calculated from weather stations is strongly
dependent on the selection and location of the weather stations
relative to the spatial distribution of urban centers.
Besides air temperature observations obtained from weather
stations, field campaigns using digital thermohygrometers
mounted on cars or bikes have been extensively employed
to study urban temperature variations (e.g., Torok et al.,
2001; Park, 1986; Deosthali, 2000; Rozoff and Cotton,
2003; Jonsson, 2004; Moreno-Garcia, 1994). However, most
field campaign UHI studies have been limited to typical conditions
that are optimal for the development of urban heat
effect, such as during stable anticyclonic events with clear
skies and calm conditions (e.g., Park, 1986).
The urban heat phenomenon can also been characterized
by surface temperatures. Although surface temperatures can
be both higher and more variable than concurrent air temperatures
due to the complexity of the surface types in urban
environments and variations in urban topography (e.g.,
Nichol, 1996; Streutker, 2002),
Kuhnkat,
have I considered using shorter periods? sure, but I try to test things that will actually have a result and not just noise.
“Have you considered using shorter periods based on population growth statistics? That is, sectioning the record to see if there are high trend periods imbedded in the long term trend and how they correlate with population numbers?”
The problem is that population growth is colinear with the things that actually cause UHI. Very simply, to support a bigger population you can grow out or grow up. increase urban AREA or increase height. It turns out that urban area is actually a pretty damn good predictor, and it so happens that population tends to
be co linear with that. As urban area grows other land uses shrink
so there is also co linearity there.
The other issue is this.
Suppose I tell you the density is 1000 people per sq km.
What really matters is how many sq km that density covers.
so again, its back to area. At some point density “might” be a good proxy for building height.
Kuhkkat
“You mention using pixels to determine buit area. Was this done for the beginning of the record and comapred to the end of the record to understand how much build out happened in the period??’
This is a smart question.
The modis built pixel is taken at a point in time say 2006.
If an area is “unbuilt” in 2006, we assume it is “unbuilt” in 1979.
I test that assumption several ways.
1. I actually go an look at sites. Nick Builds a google earth
tour and I look at them. Sometimes GE has historical
photos of areas.
2. I check for historical changes in land use data
3. I check for changes in population ( like de population)
The Built areas will fall into two classes. Those that started urban and those that started rural and became urban. I’m all set up to begin that analysis, but I’ve got some documentation to complete first. The first think to do is to build a transfer function of sorts
from Modis to other proxies for which we have historical data.
That is a much much longer project, but there are some examples of how to do that.
Carrick
yes irrigation can have a big effect. I’m looking for this Oke Grimmond study ( 2002 I think) in which Oke argued that the water differential was most important in certain cases. Especially if the city
sucked water from the rural surrounding. Then the water is released in the city ( watering, etc ).
I have bluewater irrigation data. I figured that would be most critical as opposed to lands that are irrigated by rainfall.. I havent looked into that in any depth.
The other thing I can do is just remove any stations that are in any areas where you have crops and irrigation. Lots to do.
More could be done, but for now explaining the difference between a trend analysis of SAT versus a static comparison of LST is already a bit too much for some folks…
“Maybe Tilo is so busy being really clever that he doesn’t have time to look up what MODIS LST is.”
Are you of the opinion that Land Surface Temperature means that MODIS is reading thermometers stuck into the ground?
NASA definition:
“Land Surface Temperature
The ambient temperature measured directly above the land surface.”
http://disc.sci.gsfc.nasa.gov/giovanni/additional/users-manual/G3_manual_parameter_appendix.shtml
Does the guy that knows how to do math notice that tricky little word, “above” in the definition.
Carrick: “With anomalization, there’s always an offset by the way, Tilo. So you don’t need a trend to get to one.”
Again, what does that have to do with the real world? In the real world population grows; buildings get built; infastructure gets built. In the real world Philidelphia or Tokyo don’t pop up on the surface of the earth in one day. Urbanization and UHI are trends. At the beginning the offset is small. As urbanization and UHI increase the offset get’s larger. The final offset is a result of the trend that has happened before. So that when a guy, who knows how to do math, claims that offset and trend are two seperate issues, when we are talking about urbanization, population and UHI, where offset is the direct end product of trend; then that guy has made a really stupid distinction – don’t you agree.
http://modis.gsfc.nasa.gov/data/atbd/atbd_mod11.pdf
http://www.infim.ro/rrp/2010_62_01/art17Diamandi.pdf
MODIS LST is different from SAT.
SAT is the air temperature at 2 meters
LST is the land Surface temperature.
So for example you will get studies that try to relate LST to SAT
http://ieeexplore.ieee.org/Xplore/login.jsp?url=http%3A%2F%2Fieeexplore.ieee.org%2Fiel5%2F9906%2F31478%2F01469844.pdf%3Farnumber%3D1469844&authDecision=-203
http://www.gri.msstate.edu/publications/docs/2006/03/3986GIScience_pub_2006.pdf
http://www.eproceedings.org/static/vol06_1/06_1_colombi1.pdf
“This study is aimed at demonstrating the feasibility of the MODIS Land Surface Temperature (LST) product as a source for calculating spatially distributed daily mean air temperature to be used as input for hydrological or environmental models. The test area is located in the Italian Alpine area. The proposed procedure solves, by empirical approaches, the problem of relating LST to the Air Temperature (Tair) and instantaneous Tair values to daily mean values, exploiting ground data weather station measurements as a reference. The relationship between LST and Tair is deter-mined by correlation analysis and equation generalisation for spatial distribution. The extrapolation of daily mean values of Tair from instantaneous values is addressed again by correlation analyses taking into account the altitude variability and exploiting historical series.”
Mosher: “Imhoff reduced the biomes to 4 classes and disregarded
certain classes. basically because he was not studying UHI in
the record. He also disgarded cities that are in multiple Biomes.”
Mosher, all of the stuff that you bring up is totally irrelevant. It is simply details and none of those details have a thing to do with the fact that the UHI is very large.
For example, you worry that they only used winter and summer. Do you think that the UHI effect disappeared in the spring and the fall? It’s a pretty good guess that the UHI effect is somewhere between summer and winter. It doesn’t matter if it leans a little more to the summerish numbers or to the winterish numbers it still gives you a huge UHI effect.
Yes, this varies, that varies, the other thing varies. Imhoff knows it. SO WHAT? None of that makes a bit of difference to the bottom line numbers. You can’t babble on like an idiot about the complexity as though the complexity was going to magically get rid of the UHI effect for you. It doesn’t matter if certain kinds of settlements or certain time periods or certain seasons were left out. Unless you have a good reason to think that the result of all that would be a UHI number that would be reduced by 50% or more, you are just babbling as usual.
And if you want to go on ranting about reading the whole paper, then tell me how the whole paper changes the conclusion. It doesn’t change it a bit. The whole paper is only the details behind the conclusion.
Now, getting back to our agreement of .1C per decade limit to the UHI effect to cover the difference between satellite data and surface data, Zeke’s 0.03 per decade isn’t even close to the ballpark, since his number covers only urban areas and only the earth’s land areas. Whereas the .1C per decade limit we were talking about is a global trend number.
http://www.tandfonline.com/doi/abs/10.1080/01431161.2011.560622#preview
LST is not the same as SAT Tilo. do some reading
https://secure.ntsg.umt.edu/publications/2011/MZR11b/Mildrexler_et_al_JGR_2011.pdf
“Most global temperature analyses are based on station air temperatures. This
study presents a global analysis of the relationship between remotely sensed annual
maximum LST (LSTmax) from the Aqua/Moderate Resolution Imaging Spectroradiometer
(MODIS) sensor and the corresponding siteÂ]based maximum air temperature (Tamax)
for every World Meteorological Organization station on Earth. The relationship is
analyzed for different land cover types. We observed a strong positive correlation between
LSTmax and Tamax. As temperature increases, LSTmax increases faster than Tamax and
captures additional information on the concentration of thermal energy at the EarthÂfs
surface, and biophysical controls on surface temperature, such as surface roughness and
transpirational cooling. For hot conditions and in nonforested cover types, LST is more
closely coupled to the radiative and thermodynamic characteristics of the Earth than the air
temperature (Tair). Barren areas, shrublands, grasslands, savannas, and croplands have
LSTmax values between 10‹C and 20‹C hotter than the corresponding Tamax at higher
temperatures. Forest cover types are the exception with a near 1:1 relationship between
LSTmax and Tamax across the temperature range and 38‹C as the approximate upper limit of
LSTmax with the exception of subtropical deciduous forest types where LSTmax occurs after
canopy senescence.”
Mosher: “MODIS LST is different from SAT.
SAT is the air temperature at 2 meters
LST is the land Surface temperature.”
Land Surface Temperature means temperature just above the surface Mosher. You can use it to difference urban and rural temperatures. There is no reason that the difference should be significantly different from SAT differences. Stop grasping for straws.
Tilo:
I’m of the opinion you should learn something about which you speak before you open your mouth and expose your ignorance.
If you really don’t know, then perhaps you should learn before you open your mouth and expose your ignorance.
Tilo:
MODIS LST is measured by inferring the temperature from the emissivity of the surface, not the air. It’s a complex process, but doable using multiple wavelength measurements. Measuring SAT is virtually impossible from space-based measurements, because it is swamped by LST irradiance.
Puzzle us this Tilo, if it were SAT, why do you think they call it “LST” (which in meteorology refers to the temperature at the surface of the ground not the air temperature) instead of the well-defined SAT?
Here’s a bit of help, though I suspect it’s wasted on you.
Steven Mosher:
Which is why it gets tricky, because rural sites near agriculture could be showing an anomalous cooling in addition to cities showing an anomalous warming.
One thing that is interesting there is the switch to higher water-demand crops like corn and rice has been a recent trend in the US. For urbanization, I believe for many sites, we have been near saturation of the UHI effect for at least 50 years. For rural irrigation, this may be a more recent corruption of the temperature record (and could be substantively influencing the measured “rural” SAT).
I think this is why you’re on the right track in getting your land types classified. It’s more complicated as you know than just urban versus rural. Latitude is probably the biggest predictor of trend, then coast, and probably third whether you’re in a region that have recently adopted “intensive agricultural practices.”
Here’s a discussion of the MODIS-LST algorithm.
The first link was to the AASTR (Advanced Along-Track Scanning Radiometer) algorithm. For that one, they use the 11µm and 12µm wavelengths to obtain the temperature of the surface.
For MODIS LST, they have three different algorithms, including a single-channel measurement that using an emissivity model, the two channel method developed by Wan and Dozier, and a seven channel method described here developed by Wan and Li.
Just one note: it is extremely obvious from reading the documents that land surface temperature is measuring the temperature at the surface and not the atmospheric temperature above it. If I get something mixed up, I admit to it as quickly as possible. Let’s see if Tilo can rise to the challenge, or chooses to double down on his wrong-horse bet.
Steven Mosher, I thought this was kind of interesting:
Population Trends in New York State’s Cities
Especially this graph which shows that many urban areas peaked in population in the 1930-50 time frame. (SInce then many urban areas in the US and Europe have been experiencing a well-documented deurbanization.)
So doesn’t this suggest that really you’d expect for large urban areas a slight decline in UHI effect?
I say “slight” because e.g. studies like Spencer’s UHI analysis doesn’t suggest a huge effect from UHI in changes in population density, once you get over a certain size.
Of course, once you have deurbanization along the other big curve ball, which is station moves to more rural sites, without appropriate homogenization of data, you’ll likely get large spurious negative temperature trend biases.
(Just for fun and to give a reference frame to Spencer’s abscissa axis–population density–here is a list of cities by population density.)
How does aerosols affect UHI? Air is nowadays cleaner in cities than in 70’s. Could that be reason, why urban areas are warming more than rural areas during the last few decades? Perhaps there is more insolation on the ground in urban areas nowadays than was , say, forty years ago.
Carrick: “Just one note: it is extremely obvious from reading the documents that land surface temperature is measuring the temperature at the surface and not the atmospheric temperature above it.”
Then you’ll have to get NASA to change their definition. And you’ll need to explain what you meant in #85495 when you said this:
“Satellites don’t measure temperature in the surface boundary layer so it is worthwhile looking at surface temperature measurements too.”
Mosher:
The .03C per decade UHI trend (urban and land only) is far from the roughly .1C per decade difference that we found between satellite global temps and surface station global temps. We agreed that most of that difference was likely UHI and, again, that is a global number.
The .03C per decade trend is far from the average differences between urban and rural areas that were found by the Imhoff study, where the average between winter and summer is 2C per settlement. Using your trend (if it were constant) would require about 700 years to get that offset. But of course the trend is not constant, since the great majority of the world’s infastructure was built in the last 200 years.
And the .03C trend also disagrees with the magnitude of rural/urban differences that Spencer found in his study. Spencer’s numbers come much closer to Imhoff’s.
As we have seen before, children and their fathers are easily able to find the UHI difference between rural and urban by simply comparing urban records with nearby rural records. And the trend difference is much larger than .03C per decade.
People driving from rural areas into cities can see the difference on their car thermometers.
People who live in the suburbs can see the crocuses in the city blooming a week to two weeks earlier than their own.
The .03C per decade number (for urban areas only) is simply BS. It is a hard fail. Like a Michael Mann proxy reconstruction, it is junk science and it belongs in the garbage bin.
Tilo:
I’m not sure why they choose the language “just above” the surface, when they are measuring the radiance from the surface itself and inferring temperature from that. I’m also not sure why you think it’s my responsibility to get NASA to correct their language. However, I supplied you with the mathematical definitions Tilo from the the relevant technical documents. If you can read the maths (I can help if you need), the interpretation is not ambiguous.
Again “land surface temperature” and “surface air temperature” have specific nomenclature in the met community (undoubtedly less so for number crunching types in the climate community who’ve never collected a data point in their life).
Here is an example where the two are clearly delineated.:
Tilo:
I thought from the context it was obvious: “surface air temperature” or SAT. Note in this same sentence, “temperature in the surface boundary layer” appeared, so it’s not like you had to go searching for context.
You may find a discussion of the (planetary or atmospheric) surface boundary layer of interest, and the Wiki does an OK job here.
Tilo:
I of course think this claim is nonsense. But if it were true, and assuming you are not technically inept, you could demonstrate this point, instead of just claiming it were true.
We look forward to your detailed analysis, which hopefully involves something more than resorting to salesman-pitch anecdote.
A bit more text for Tilo, who I suspect can only read abstracts:
Any questions Tilo on the interpretation of LST at this point?
For Steven, here’s an interesting figure from the paper.
Let me know and I’ll sling you a copy.
(I’ll note that there are land-mines in the comparison of SAT to LST having to do with the near-surface variation in the vertical air temperature profile , and leave it at that.)
Regarding the effects of farming and irrigation, here is a lecture by Christy that addresses that subject.
http://www.youtube.com/watch?v=-WWpH0lmcxA
This is a long lecture. To get right to the part about irrigation, go to time 13:40.
Steven Mosher comment # 87108:
“i suspect that if we we could do a temperature series that was
corrected for this bias and the UHi bias that we would match
the satellite record to a nats ass. or, if we can build a record that has stations that have the correct frequency distribution of latitude you would see the same thing.”
(G)nats ass is close enough. Like I said, go with the satellites. Move on to ocean heat content. The inadequate surface station record has been flogged, and reflogged enough already. Those are my orders. I’ll be back.
Carrick: “But if it were true, and assuming you are not technically inept, you could demonstrate this point, instead of just claiming it were true.”
You are obviously reading impaired. I don’t simply claim that it is true, I reference the work of others that show that it is true.
Carrick: “I’ll note that there are land-mines in the comparison of SAT to LST having to do with the near-surface variation in the vertical air temperature profile”
It’s irrelevant for the most part, since Imhoff’s comparisons are not SAT to LST, but urban LST to rural LST. Spencer uses only SAT and his results are reasonably close to Imhoff.
The stations are overweighted at 30-60 degree north.
That means the distribution of stations with latitude does not match the distribution of land area with latitude, consequently there is a bias toward higher trends being in the record.
Is there any possibility of constructing a “30-60degN only” satellite record, just to compare with the surface record in that zone? That should help settling this problem.
Tilo:
That you made this comment says more about you than it does about me.
Actually the “work of others” doesn’t say what you think it says. Like with the LST versus SAT issue, which you’ve still failed to admit error to btw, you really need to read more than “top of the envelope” information.
It’s not irrelevant, they are different physical quantities, and they have different responses to the UHI effect. I would explain but don’t think a “reading impaired” person like myself could possibly explain it in terms that are clever enough for a bright young chap like yourself.
Carrick: “It’s not irrelevant, they are different physical quantities”
Again, you are reading impaired. What does it matter if they are different physical quantities when it is not what is being compared to produce the UHI difference between urban and rural. Try to understand this now. We are not comparing rural LST to urban SAT or urban LST to rural SAT. We are comparing rural LST to urban LST. So it is irrelevant. Now you can claim that the delta’s that are given for LST differences would be different from the delta’s that are given for SAT differences, but that doesn’t seem to hold water either, since Imhoff’s LST only results are in the ballpark with Spencer’s SAT only results – and the 0.03C per decade trend is out of the ballpark with both of them.
Tilo, seriously, I’m not reading impaired. Can you stop with the kindergarten tactics now?
It’s also unfortunate, but you really don’t have the slightest idea what you’re talking about and you’re way in over your head.
The metrics MODIS LST and ground-instrument measured SAT show different responses to UHI effect. Steven discussed some of the reasons for that above.
Having said all of that, let’s address this:
Since I obviously am reading impaired, point us to the specific place within Spencer’s or Imhoff’s analyses where they published the numerical value for their estimates of the bias from UHI from say 1979-now in global mean temperature trend (obtained using any of the standard global temperature reconstruction algorithms such as GISTEMP or CRUTEM3).
If you can’t point to such a place in either analysis, you can’t just claim “0.03C per decade trend is out of the ballpark with both of them,” and it will require further analysis on your part to derive what bias in temperature trend would be expected from their results, in order to demonstrate that “the 0.03C per decade trend is out of the ballpark with both of them.”
And it also implies that the phrase “reading impaired” apparently doesn’t mean what you seem to think it means.
Carrick: “It’s also unfortunate, but you really don’t have the slightest idea what you’re talking about and you’re way in over your head. ”
You keep repeating that, but every time you open your mouth you show that it is you who has that problem.
Carrick: “Can you stop with the kindergarten tactics now? ”
I can if you can.
Carrick: “If you can’t point to such a place in either analysis, you can’t just claim “0.03C per decade trend is out of the ballpark with both of them,†”
Try your own experiment. Step 1 – how long would it take, at 0.03 per decade, to get the kind of offsets that Imhoff and Spencer show? We’ll do some others once you get close on that one.
Tilo,
If you cannot do what Carrick just suggested, you should stop.
Tilo:
If you’re conflating me having mental deficiencies (reading problems) with my commenting that I think you in are way over your head (a issue with lack of knowledge), then chalk up “kindergarten tactics” as another phrase you could use help with.
In any case, the question at hand is what is the effect of UHI on temperature trend estimated from global mean temperature reconstructions, and by that we are referring to data products such as NCDC, GISTEMP and CRUTEM3.
Steve and Zeke’s poster and Steve’s additional results posted on this thread provides such an analysis, as well as a concrete estimate of the effect.
Let’s agree to stipulate that neither of the analyses you pointed to contain such an analysis.
BZZ. What plain silliness. Sorry I don’t have time for a game of pea and thimbles right now.
Remember you’re the one who claimed I was reading impaired because I couldn’t see how the papers were could be directly related to Steven and Zeke’s results.
What I challenged you to do was, if you thought these papers were relevant to the demonstration that the magnitude of the result reported by Zeke and Steven was too small, to either locate the place in these published analyses where an estimate of temperature trend bias was made, or to produce such a number using your own analysis.
If you are incapable of producing an analysis, I suggest you a) admit you are not capable of it and b) admit your comments should be disregarded because you weren’t knowledgable enough to have made them.
So we’re left with this Tilo, what is the temperature trend bias in global mean temperature reconstructions implied by Spencer or Imhoff’s analysis?
This is a straightforward question, you’ve assured us the answer is very easy for anybody who isn’t “reading impaired”, and since you are asserting that I’m “reading impaired” and the one “way in over my head”, I think the onus is on you to produce this result yourself.
Tilo:
You need to cast a broader net. As much as I admire Dr. Spencer, I think his finding that half of the warming trend may be due to UHI effects seems off the mark.
If, for example, you read Roger Pielke Sr.’s estimable blog for temperature record issues (height off the ground, UHI, time of day bias, changing locations, wet bulb/dry bulb, surface conditions, etc.) he and those he cites have sliced and sifted this stuff as many ways as possible and find that yes, the temperature record has an small but significant upward bias for methodological reasons but certainly not in the 50% range.
Given the limited quality of metadata (i.e., we do not know the detailed history of the plot of land surrounding each station) I think the work done by Mosher, Zeke et al is as good and as thorough as we can get. This is not their first pass on this stuff.
They may have come up with a low-ball measure but I doubt there is a substantive way to generate a better (scientifically & mathematically speaking) number. Unless and until there is a better substantive answer I see no reason not to accept the 0.03 UHI estimated rate.
I think Spencer’s attempt to generate a population density corrective constant might be interesting if a historical study could be done to capture the effects of changing population. But even that should be pretty rough measure given that similar population density does not guarantee similar land use etc.
So even allowing for the complexity and limitations of the data issues, I think it is reasonable to assume a smaller rather than a larger UHI effect. Assuming a contrary position now comes with a burden of proof.
Tilo,
If you want to see comparisons of LST and SAT and how different they are, I suggest you read this
http://land.umn.edu/documents/Urban_heat_island–Impervious__RSE_paper.pdf
Second. The LST Imhoff uses is taken at 130PM and 130AM
Several things
1. The samples are 8 days worth of data
2. These days will represent maximum UHIs all other
days will have less UHI, so his figure is NOT an annual
average or Monthly average. It is the average of 8 cloudless
days. Also see his comment about anti cyclonic conditions.
3. Take a measure at 130AM maximizes the difference with
rural stations. It doesnt measure Tmin.
If you want to see how the times he uses maximize UHI have a look at UHI versus Time for these cities
http://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=10&ved=0CHEQFjAJ&url=http%3A%2F%2Fams.confex.com%2Fams%2Fpdfpapers%2F126615.pdf&ei=D3biTpmuLaSZiQLB0eHdBg&usg=AFQjCNFXuXjv6LbLu65fEGqPkexcQw_Leg&sig2=yqwoIeQP8PU_vkIX_yaduA
Carrick: Sorry, ranting won’t work here. You get an F for failing to answer a very simple problem.
As Imhoff states:
“Globally averaged, the daytime UHI amplitude for all settlements is 2.6 °C in summer and 1.4 °C in winter.”
And Spencer is in that same ballpark.
So, we get roughly 2 C UHI amplitude for all settlements. I went through what that means for the population density numbers before in Zeke’s post and I’ll bring it back shortly. So, to get 2C of UHI warming for all settlements, using a trend of .03C per decade, it would take about 700 hundred years. I would have given you a pass on anything within a hundred. But, oh well. Now, of course that only applies if the UHI trend was constant. But the reality is that population growth has been an accelerating phenomena. And infastructure growth has accelerated even faster than population growth. So the .03C won’t be a constant, rather it will be the steepest part of the UHI trend right at the end, for the last 100 years. So, now you are off the charts for how long it would take to get 2C of UHI when the steepest end piece is .03C.
I know that you would like to have an exact trend number provided by Imhoff and Spencer, but only a tiny amount of thinking will show you that you can’t bridge from Zeke to either of them in any way that looks reasonable.
Mosher: “If you want to see how the times he uses maximize UHI have a look at UHI versus Time for these cities”
So, looking at Dallas, I see an hourly variation between 1C and 2C. The average looks to be roughly 1.5C. Given that Dallas is arid, that number still fits into Imhoff’s 2C for all settlements.
George Tobin,
I also believe that Zeke, Steven et al have done what can be done with the surface station record. But knowing the record is almost devoid of truly rural stations, isn’t it plausible to assume that the UHI effect is higher than .03 per decade? As I stated in a previous comment:
The point I was attempting to make is this: “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.†and “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.†Leading to my naive theory that if the unpopulated expanses of the land surface had been properly sampled over the period of your study, the UHI bias might be estimated to be in the .06 to 0.1C per decade range, and would likely be more than .03.
I would draw Tilo’s attention to the following
See figure 8 in the following
http://land.umn.edu/documents/Urban_heat_island–Impervious__RSE_paper.pdf
You will all note how LST varies in a linear fashion with % of ISA
Lets take an example
plate b
In the summer the temperature at 100% ISA is 312K
the temperature at 25% ISA is ~305K
the temperature at 0% ISA is ~ 303K
Imhoff takes his measure of LST at the urban core where
ISA values by his definition are 75%to 100%
the rural measure is taken beyond the 25% contour line.
For this example you would have a value of around 310K
for the urban core and then around 303K for the rural LST.
Or 7K.
Now, how does that relate to what we did.
Well, the first thing you have to ask yourself is where are stations ACTUALLY located. For our urban stations the average
ISA is …….. about 25%. Simply, Imhoff measures SUHI at
its MAXIMUM. Maximum in terms of picking places where ISA is HIGHEST. Maximum in terms of weather conditions. MAximums in terms of season. Maximum in terms of time of day. Maximum in terms of what is measured– the inferred temperature of the
SURFACE– not the air 2 meters above the surface
This UHI never gets into the record. That is, the SUHI is not
put into the record that we analyze
1. we analyze SAT. as you can all see from the table in the paper
the comparsion between LST and SAT varies widely. with
LST being anywhere from 0K to 11K greater than Tmean.
So, Tilo is wrong. LST is different than SAT. It is higher
and more variable than SAT
2. because the actual stations are not located in the same areas
that Imhoff uses. for all urban stations the ISA figure is
around 25%.. NOT the 75-100% that Imhoff uses. In the example
above this would mean a temperature of around 305K.
basically, you cannot compare what Imhoff did to what we did.
1. he took a static measure
2. he used a metric (LST) that is higher than SAT
3. He measured the peak UHI, we looked for the trend.
Further. From 1979 to 2010 the trend from satellites
is around .20C. Our rural only trend ~ .25C ; our urban only
trend is ~ .28C.
As we argue, there is a “budget” for bias over this period of
around .1C. The sum of all biases in the land record will
be less than .1C per deacde. We identify about .04C of that
as being attributed to UHI.
Tilo,
Oddly enough, many currently urban-located instruments were originally installed in urban areas and there are (unfortunately) few long continuous station records (with no moves) outside of the US, Australia, and Europe. Your 700 year number makes no sense unless you assume that all currently urban stations start off as pristine rural stations and manifest the absolute UHI in the trend, something that is the exception rather than the rule.
This paper is rather old ‘http://www.metlink.org/pdf/articles/urban_heat_island_-_birmingham.pdf’, but its conclusions seem to suggest that UHI is rather more complicated than just ‘UHI is real. We do not accept this on faith but we accept it based on many studies that show that urban areas are warmer than their rural surrounding.’ In case people don’t want to spend time reading it, those conclusions are:-
‘In so far as it is revealed by an urban rural comparison of nocturnal minima, heat island development is a fairly frequent feature at Birmingham, especially during spring and autumn in settled anticyclonic conditions when city temperatures can be as much as 5 K greater than those in the rural surroundings. During the day-time there is a less well marked tendency for the city to be colder than its surroundings. These negative temperature anomalies are associated with disturbed airflow types, suggesting that vertical mixing is an important factor in their development. In common with summer heat islands reported from other mid-latitude cities, Birmingham’s heat island is predominantly of type 1, that is the city to be colder than its surroundings, especially in spring and high surroundings[sic].’
I think that last is probably a typo.
Is there a chance that UHI is being over-simplified?
Tilo:
Ugh. More dreck.
I don’t get any grade because I refuse to participate in your nonsensical proposal that I do any stage 1, stage 2, etc. silliness.
You’ve made a number of claims, at this point, I’m just stating for the record that you have categorically failed to live up to any reasonable expectation of supporting those claims, you’ve made numerous mistaken statements, and you’ve never gone back and corrected a single one.
IMO you can either a) admit you don’t know anything and your opinions are therefore virtually meaningless with respect to UHI contamination or b) demonstrate by substantiating your own claims which you insinuate are so obvious from the references you given that one must have reading comprehension issues to not immediately grasp them, or c) cast off what tiny shred of credibility you might have remaining.
This is a hole that you’ve dug entirely on your own, and the choice of what you do now is also entirely your own.
Ciao.
toto:
I think this is an interesting question.
A related one is “what should the temporal effect of UHI on anomaly offset really look like”? In other words, can we model how we might expect it to vary with time, and use a Monte Carlo based approach to insert realistic UHI signals (with any geographical distribution we choose), to compute how much bias in trend is generated by that UHI signal.
Roger Smith:
I think people understand that it’s pretty complex. The question is more of “can we set an upper bounds” on the influence of UHI on global mean temperature trend than how well we can categorize UHI.
Carrick: “You’ve made a number of claims, at this point, I’m just stating for the record that you have categorically failed to live up to any reasonable expectation of supporting those claims,”
What you are stating for the record is an ignorant opinion that is based on absolutely nothing. I have given you the support for the claims and you have yet to contradict or in any way challenge the support. Your only response has to do with LST vs SAT which is basically irrelevant to the argument in any case since the two methods of measuring are not mixed. Your argument that there is a UHI offset, but little or no trend to get you to that offset is just plain dumb. This shows that either you don’t understand the arguments that I have made and are trying to cover it up, or you do understand the argument and you are just flapping your lips to save your overblown ego. In either case, I don’t care.
RSS has a land only 20-82.5.
That might be close enough.
http://www.remss.com/data/msu/monthly_time_series/RSS_Monthly_MSU_AMSU_Channel_TLT_Anomalies_Land_v03_3.txt
Certainly looks like its possible to get any lattitude range, but probably not by you or me.
If you can live with land + ocean, you can get an idea from figure 3 here.
http://www.remss.com/msu/msu_data_description.html
Zeke: “Your 700 year number makes no sense”
It’s not my 700 year number, Zeke. It’s how long your number, 0.03C per decade, would take to produce a 2C UHI offset. That being the offset that was found by Imhoff. That’s why your number doesn’t make sense. But since UHI is an accelerating phenomena, due to accelerating population growth, and an even more rapid infastructure growth, your 0.03C makes even less sense because it would be the most rapidly increasing portion of the UHI curve. This then means that the time it would take to get 2C of UHI offset, using your 0.03C number as then end of an accelerating UHI curve, is simply off the charts. Let me interpret. If Imhoff is right about there being and average of 2C of UHI effect per settlement today, then we can’t get there with your 0.03C per decade of UHI trend. One of the numbers has to be wrong. Given Spencers work on UHI and given the difference between satellites (no UHI) and ground stations (with UHI) of about .1C per decade on a global scale; my vote is that your 0.03C per decade of UHI effect for urban areas only, is wrong.
John M:
John, are you aware of any progress being made in sorting out the divergence between UAH and RSS?
Tilo, lol. whatever.
Tilo,
I’m not aware of progress I’m making on my own job let alone someone else’s.
Toto:
Here are land+sea averages from +30 to +60:
UAH
HADCRUT3
I don’t think it’s wise to try and apply land masks without really careful consideration, so I haven’t tried that.
The trends are 0.276°C/decade (HADCRUT) and 0.312°C/decade (UAH). The uncertainty in trend associated with climate fluctuations is roughly ±0.025°.
Anyway, this gives trend_HADCRUT – trend_UAH = -0.036 ± 0.033 °C/decade.
I’ve just read the poster and join with Lucia in congratulating all the author’s in putting together this excellently researched and well presented document.
Forgive me if this question as been covered in the comment’s as I’ve just briefly scrolled through them. Under the discussion you state: –
“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”.
I understand the point about being problematic from the spatial coverage aspect but isn’t this larger bias of 0.06 to 0.1C per decade an indication that, even with the stations categorised as rural, possibly there may have been a significant amount of dUHI over the 31 year period studied?
Carrick,
“I don’t think it’s wise to try and apply land masks without really careful consideration, so I haven’t tried that.”
Any mask would be better than none.
phi:
Not if it isn’t applied consistently. HADCRUT is a blend of two products, and you can get to CRUMTEM3 directly. Getting it to match up with UAH is really tough though, and the error you introduce from the mask can be huge, if you don’t get the mask exactly the same.
The reason is ocean behaves very differently (no latitudinal dependence), see e.g. this.
Any contamination from the ocean portion of the Earth’s surface in your mask is going to really screw with your land-only reconstruction.
We found this when trying to match CRUTEM3 and GISTEMP to BEST land-only.
Here are some examples, three different land-masks for clear climate code (courtesy of KevenC at CCC):
These are all 1979-now and they apply to the global temperature (getting the +30 to +60 degree reconstructions would take a bit more work than I’m willing to do right now) so their trends are obviously going to be smaller than the trends for UAH and HADCRUT (30-60 only).
giss.land+ocean.txt 0.163
giss.mask1.txt 0.240
giss.mask2.txt 0.219
giss.mask3.txt 0.191
As you can see, the choice of the land mask introduces much more uncertainty than the small difference seen between UAH and HADCRUT3. The take home, is there isn’t a “land-only” Earth, and there is a substantial amount of variability introduced by how you try and mask the ocean to create one.
In any case, I don’t see any reason why you shouldn’t be looking at the effect of UHI on global temperature reconstructions in any case (not just the land only portion), if what you want to look at is “global warming”. Obviously you could optimize it to see more UHI effect, but what’s the point?
What you really care about is global mean temperature for the real Earth, and the putative influence of different systematics on that measure, and not the influence of different people’s choice of hypothetical land-only planets.
OK, thanks guys.
The question is whether satellite-surface difference are reduced when you consider only the extra-tropical, extra-polar northern hemisphere (which would indicate a temperate-NH bias of the surface record, confirming Mosher’s hypothesis).
RSS land only, latitudes 20-82.5 (from John M’s link), gives me .2517 / decade. But I couldn’t find any surface data to compare it to. Crutem NH has .305 / decade (as per woodfortrees), but that includes the 0-20 latitude band, which presumably drags down the value.
I guess the folks who have built actual gridded reconstructions could do a better comparison (land-masked stations vs RSS land-only on the 20-82.5 latitude range).
Of course I suspect this has already been done ten times over by smarter people :p
Geoff:
Really, we don’t know if the difference is entirely due to urban heating or not. There’s the effect of irrigation on “rural” areas that could lead to net cooling of sites located in agricultural regions. Then there’s the effect of a high fraction of urban sites being located near oceans, and the contamination of marine boundary layer on your land surface air temperature measurement, this could be shifting the urban-only sites to a lower mean trend than they might have had, were UHI the only systematic influence present.
Anyway, I’d argue that you have to address what “significant” means before you can go much further. I think the appropriate thing to be comparing is influence of different systematic effects on the various global mean temperature reconstructions. Some reconstructions, like CRUTEM3, ignore UHI entirely. GISTEMP tries to address it by assuming any difference between rural and urban is UHI, but this may be over-correcting the urban site.
In either case, the effect on global mean temperature isn’t that large, and ironically the reconstructions that have the UHI correction ends up with a larger trend than the one that doesn’t (these are 1979-2009 because I haven’t bothered to update every series to current):
ecmwf 0.164
ncdc.temp 0.164
hadcrut3gl 0.160
giss 0.165
But the difference is only 0.005 °C/decade, and probably not driven by UHI but other effects such as differences in geographical coverage, and e.g. differences in how the various algorithms infill missing station data.
Notice that Tilo has still not got it
1. Imhoff measured UHI at the urban core( high ISA)
2. Urban Stations are located on average in zones
where ISA is 25%
3. The fall out in temps between these two zones ( see above )
is huge.
4. Imhoff measured the peak UHI
5. LST which he used is always warmer than Air temps.
Questions to Tilo.
1. Do you see how LST falls off with the % of ISA
2. Do you see why Imhoff’s measure would be the peak?
3. Do you finally understand the difference between
the surface temperature of the land ( LST ) and the
air temperture 2 meters above the land.
I know other people get this.
At this point, I think most people here also get that Tilo doesn’t get it.
Congratulations, gentleman.
Wish I could be there.
I suspect that this thread might be essentially dead, but I did some follow up work on urban to rural USHCN stations that might bear on this paper or future work that these authors might be contemplating doing. If I receive no responses here I will recast it at another thread at another time.
Here I was attempting to get a feel for the CIs that would be appropriate for the part of the paper/poster under discussion in this thread where the authors used urban-rural pairs. In the process I came up with what I considered some interesting results.
I used the rural and urban designations from GHCN V3 and applied it to the USHCN stations. I also used the station altitudes and proximity to water from GHCN. I looked at the trend differences for an urban station to its 7 nearest rural neighbors and calculated trends of the difference series and correlations of the series of the urban to rural stations. In this exercise I used the monthly maximum and minimum adjusted temperatures from the USHCN for two periods: 1979-2011(Sept) and 1920-2011(Sept).
Further, I used for my comparisons only the station pairs that were no more than 300 meters different in altitude and both had the same water proximity designation. The results are listed below for the two time periods for the average trend of the difference series of Urban – Rural:
1979-2011 Maximum Temperature: -0.0047 +/- 0.022 Degrees C per Decade
1979-2011 Minimum Temperature: 0.017 +/-0.0181 Degrees C per Decade
1920-2011 Maximum Temperature: 0.0305 +/-0.0138 Degrees C per Decade
1920-2011 Minimum Temperature: 0.0561 +/-0.0134 Degrees C per Decade
What is notable with the results is that for the 1979-2011 period one cannot reject the null hypothesis that the trend in the difference series of Urban – Rural is not zero while for the period 1920-2011 one can for both maximum and minimum temperatures and with the difference being positive. I need to quickly note that while this finding will effect little the overall trends 1920-2011 (because the rural stations in the analysis numbered 7 times more than the urban ones), it does have possible implications for the USHCN adjusted data for the 1920-2011 period in that the USHCN breakpoint algorithm evidently was not able to homogenize out the “UHI” effect.
I studied the 1920-2011 period because I consider that a good compromise for looking at longer term trends in the US between available and reliable data. I also think that since we have satellite data from 1979-2011 to more or less confirm the earth bound station data we need to look further back at station data to analyze long term temperature trends. I also think that if any likely UHI effect was measurable it would be more likely found in a longer series where the urban environment would more likely have changed (sufficiently to affect a temperature trend) than it would have in more recent times like 1979-2011.
I used the 300 meter altitude difference and water designation based on the comparison of all stations with their nearest 7 neighbors and looking at the correlations and the variations in correlations and trends where the altitude and water proximity were used as factors.
I should also note that the CI limits I calculated are not the most conservative I could have used – I think. I used pairs such as U1-R1 and U1-R2 and counted these a distinct pairs but I am not certain that that is correct. The DOFs I do not think would imply only one pair but they might not imply two either. Even so, with more conservative CIs, the null hypothesis for the 1920 period should be rejected.
Thanks Ken.
I suspect if you convert those 1920-2011 numbers into a bias on global anomaly, even those they appear significant, you’d find the bias error to be smaller than the error in reconstructed global temperature. Do you agree with this?
Also, your approach would seem to overestimate the number of DOFs, is that correct? Or do you think it underestimates it? I wasn’t sure which you were saying was the case.
Nice work!
“Also, your approach would seem to overestimate the number of DOFs, is that correct? ”
It would not under estimate the degrees of freedom – of that I am sure. I do not know how to handle the U1-R1 and U1-R2 for determining DOFs. All my pairs are unique but I do use the same station in more than one of the pairs and thus at least some of the pairs are not all completely independent. I’ll have to research on how to handle that in calculating DOFs.
I have only analyzed the US data, but I would guess that for the overall global temperature trends that the UHI (by the methods I used) would have only a small effect.
The two major points of my analysis were to test whether the USHCN algorithm for homogeneity adjustment would compensate for the “UHI” effect and whether the effect would be different for the 1979-2011 and 1920-2011 time periods.
I had also noted in an earlier post that in Zeke’s poster there were no CIs listed for the paired comparisons and on thinking about it I thought they might have had a similar problem to what I face with using the same station in some of the pairs. In Zeke’s paper the pair-wise comparisons also pieced together several rural stations to compare to one urban station and I was wondering how that would affect the DOFs.
Also in Zeke’s paper they used different criteria for determining rural and urban stations than I used in my analysis and I am guessing that my 7 to 1 ratio of rural to urban stations would be greatly lowered by Zeke’s criteria. I can thusly see a larger UHI effect that translates back to a much smaller effect on the overall trend because that overall trend is based mostly on stations designated rural.
I have found an error in my calculations and I am now hoping that anyone who might have seen my initial post reads this one as it does change the conclusions. Below is the corrected results along with the original ones with the errors and results using an alternative method for the calculation.
Original results which are in error:
1979-2011 Maximum Temperature: -0.0047 +/- 0.022 Degrees C per Decade
1979-2011 Minimum Temperature: 0.017 +/-0.0181 Degrees C per Decade
1920-2011 Maximum Temperature: 0.0305 +/-0.0138 Degrees C per Decade
1920-2011 Minimum Temperature: 0.0561 +/-0.0134 Degrees C per Decade
Corrected results using the same calculation methods:
1979-2011 Maximum Temperature: -0.0047 +/- 0.022 Degrees C per Decade
1979-2011 Minimum Temperature: 0.0228 +/-0.0173 Degrees C per Decade
1920-2011 Maximum Temperature: 0.0070 +/-0.0097 Degrees C per Decade
1920-2011 Minimum Temperature: 0.0292 +/-0.0114 Degrees C per Decade
Corrected results using alternative method:
1979-2011 Maximum Temperature: -0.0145 +/- 0.0377 Degrees C per Decade
1979-2011 Minimum Temperature: 0.0194 +/-0.0231 Degrees C per Decade
1920-2011 Maximum Temperature: 0.00123 +/-0.0171 Degrees C per Decade
1920-2011 Minimum Temperature: 0.0277 +/-0.0231 Degrees C per Decade
The alternative calculation was made by taking an average of the trends of difference series of the Urban-Rural nearest neighbors pairs for each Urban station and using the total Urban stations to obtain the degrees of freedom for calculating CIs.
While some of the significance is retained after the correction and particularly for the Minimum temperature using the original method, the use of the alternative method reduces the significance of the 1920-2011 Minimum temperature to a bare minimum.
These results then show that for the GHCN rural/urban designation it is difficult to show the UHI effect with the USHCN adjusted (homogenized) temperatures and even for the 1920-2011 series for Minimum temperatures.
Thanks for the update Ken…
The weakness of the UHI signal must continue to be a surprise to people who have in the past seemed convinced that it represented a substantial bias in the observed trend. I think your approach of removing coastal sites from the study and keeping roughly the same elevation are both good practice, since they isolate the UHI from many other potential confounding effects.
My final results change once again after I went through my R code line by line to pin down the (dumb a–) error and it applies to all results and shows that the UHI effect as I determined it in the preceding posts is not statistically significant in either the 1979-2011 or 1920-2011 periods for the USHCN Adjusted temperatures. The error resulted in comparing Urban-Rural station pairs that were not nearest neighbors. My calculations here are with the alternative method from my previous post. I considered that approach the more conservative and proper method.
My final results are listed below:
1979-2011 Maximum Temperature: -0.0119 +/- 0.0298 Degrees C per Decade
1979-2011 Minimum Temperature: 0.0018 +/-0.0323 Degrees C per Decade
1920-2011 Maximum Temperature: 0.00732 +/-0.0718 Degrees C per Decade
1920-2011 Minimum Temperature: 0.0162 +/-0.0166 Degrees C per Decade
It would appear that the USHCN algorithm used to homogenize the TOB data could well compensate for any UHI effects in the TOB data set. I will attempt to next determine whether any UHI effect appears in the TOB data using my criteria and calculations that I used for the Adjusted series.
An interesting aside and result was that for the Urban Rural stations that failed my criteria of no water proximity and altitude differences between pairs greater than 300 meters. Those results are listed below and do show some significant differences – although for my analysis and based on my a prior criteria for selecting Urban_Rural pairs it means nothing other than explaining at least in part the direction of my error in my previous posts.
1979-2011 Maximum Temperature: -0.0475 +/- 0.0550 Degrees C per Decade
1979-2011 Minimum Temperature: 0.0545 +/-0.0421 Degrees C per Decade
1920-2011 Maximum Temperature: -0.0012 +/-0.0258 Degrees C per Decade
1920-2011 Minimum Temperature: 0.0252 +/-0.0230 Degrees C per Decade
“The weakness of the UHI signal must continue to be a surprise to people who have in the past seemed convinced that it represented a substantial bias in the observed trend. I think your approach of removing coastal sites from the study and keeping roughly the same elevation are both good practice, since they isolate the UHI from many other potential confounding effects.”
The signal is no doubt weak and that is why I wanted to test whether the USHCN algorithm using breakpoints could remove it. I started with the Adjusted data and thus now I need to look at the TOB data series to determine whether it shows a statistically significant UHI effect. It may well be that the TOB series will not show it but if it did and the algorithm removed I would have to give Menne and company a hats off.
My results for determining a “UHI” effect in the unadjusted TOB series are listed below:
1979-2011 Maximum Temperature: -0.0312 +/- 0.0627 Degrees C per Decade
1979-2011 Minimum Temperature: 0.117 +/-0.0751 Degrees C per Decade
1920-2011 Maximum Temperature: 0.0555 +/-0.0229 Degrees C per Decade
1920-2011 Minimum Temperature: 0.0430 +/-0.0353 Degrees C per Decade
The TOB series in USHCN do show significant “UHI” effect with a lot of noise included. The TOB series have not been in-filled like the Adjusted series and thus when I use a period like 1979-2011 that is only a nominal one as the missing data in either station in the Urban-Rural pairs results in that month not being used in the difference series. That could account for the larger CIs for TOB than Adjusted series in the 1979-2011 period. I also think that the in-filling and homogenization processes tend to “smooth” out differences in the Adjusted series – maybe it over smoothes the data.
Regardless of the conjecture on how it happens, the Adjusted and TOB series produce very different “UHI” results.
My concern now would be whether the homogenizing algorithm used in USHCN legitimately removes the UHI effect or whether it is merely overly smoothing the TOB data so that differences in the Adjusted series becomes difficult to find. I continue to remind that even the effect found in the TOB series would not have much effect on the overall temperature trends because the cooler rural stations outnumber the Urban ones by 7 to 1 in my analysis. In addition there are the effects of altitude and water proximity on temperature trends that are not small and add to the noise of the overall temperature signal.
For completeness I used the TOB months for calculating the Adjusted trends of the difference series of the Urban station with its 7 nearest neighbors as I did in the previous posts. My comparisons are now improved since I have used the exact same months in both the TOB and Adjusted calculations. TOB has fewer data points since that series has not been in filled for missing data.
The Adjusted trend averages and CIs using the TOB months of available data are listed below:
1979-2011 Maximum Temperature: -0.0206 +/- 0.0411 Degrees C per Decade
1979-2011 Minimum Temperature: -0.0002 +/-0.0429 Degrees C per Decade
1920-2011 Maximum Temperature: 0.0004 +/-0.0022 Degrees C per Decade
1920-2011 Minimum Temperature: 0.0030 +/-0.0041 Degrees C per Decade
A comparison with the previous data for Adjusted and TOB temperatures shows that accounting for fewer data points in the TOB series does not change the view that the TOB and Adjusted data are different with regards to showing a significant UHI effect – or better a difference between Urban stations and the nearby rural stations.
The comparison in my previous post concerning the trends of difference series between the Urban and 7 nearest neighbor Rural stations using both the TOB data series and the Adjusted data series (that was limited to the same months of available data in the TOB series) is not complete without a statistical test comparing the TOB and Adjusted results. To that end I have posted the p values for these comparisons for both one and two sided tests below:
1979-2011: Max Temp 1 sided=0.368; 2 sided=0.735 and Min Temp 1 sided=0.0039; 2 sided=0.0077
1920-2011: Max Temp 1 sided=0.0000; 2 sided=0.0000 and Min Temp 1 sided=0.0127; 2 sided=0.0254
The differences between Adjusted and TOB USHCN temperature series, as applied to the Urban Rural UHI effect, are statistically significant in all cases except for 1979-2011 maximum temperatures.