Home > Uncategorized > Anthromes and UHI

Anthromes and UHI

With BerkeleyEarth 1.6 posted to CRAN I figured it was time to do some sample programs to explain how the package worked and integrated with other packages. Also, I have some issues to check out with the metadata; and in the long run I want to reformulate my metadata package to include some new resources. There is a great project I found at http://www.worldgrids.org.  When that project completes I think it will make an interesting addition to our understanding of temperature station metadata. As I travelled around the internet in search of more data I happened upon a couple of resources. First a resource that will make a great cross check on Modis Urban extent data: It’s a 300 meter resolution map of the entire globe: http://postel.mediasfrance.org/sommaire.php3?langue=English.  The dataset is behind a registration wall, but getting access to it was easier than getting access to Modis. In a future post I will compare the two datasets. The second dataset I found was referenced at worldgrids: Anthromes. What is an anthrome?  Human beings, like other animals, transform the landscape they inhabit, and anthromes represent the various classes of human transformed landscape.  In a nutshell an Anthrome is a function of population, technology, affluence and the natural landscape. Croplands are an example. So are cities. This method of classifying the lanscape was interesting to me because of the difficulty researchers have had in finding a discernable UHI signal in global temperature records. I won’t recount the list of failed attempts. In simple terms most researchers attack the problem by identifying “rural” sites and  “urban” sites and then looking for differences in trends between the two. That is relatively straightfoward. If “urban” sites are infected with UHI, then we should be able to see that bias by differencing  rural sites and urban sites. That approach has several complications. First and foremost is the problem of how one classifies a site as Urban or Rural. Underneath that is the assumption that rural sites are not biased by land use changes themselves. Anthromes, it seemed to me, might provide a way out of this issue as sites are classified  by basically two dimensions– a land use dimension and a population dimension. The data is open and freely available and can be read into R rather simply after downloading.

Anthrome data is delivered in 5 minute resolution or roughly a 10KM cell –at the equator. In addition, there are historical estimates for 1900, 1800 and 1700. The work depends upon the HYDE 3.1 dataset and Landscan population.  The Anthromes  are  shown below in the legend

I started with a rather simple test. Let’s go the code
require(BerkeleyEarth)
 

Stations <- readSiteSummary(Directory = choose.dir())

Anthromes1900 <- file.choose()
Anthromes2000 <- file.choose()
 

Ant1900 <- raster(Anthromes1900)
Ant2000 <- raster(Anthromes2000)

lonlat <- cbind(Stations$Lon, Stations$Lat)

Stations <- cbind(Stations, Anthrome1900 =  extract(Ant1900,y = lonlat),
                                                        Anthrome2000 =  extract(Ant2000,y= lonlat))

Urban <- Stations[which(Stations$Anthrome2000 == 12 | Stations$Anthrome2000 ==11),]

Pretty simple. I use the BerkeleyEarth function to read in the stations. Then I read in two rasters: one containing Anthromes for 1900 and the other for the year 2000. Then I extract the values for each station given its Longitude and Latitude. Then I select only those stations which were Urban or mixed settlements in the year 2000. That reduces the 36,000 Berkeley Earth stations to around 8000. Their geographic distribution looks like this.

Next, we read in the data, window it to 1900 to 2011 and then we use “intersectInvData()” to make sure that the stations in the inventory match those in the data. After that we calculate weights and then “solve” a least squares problem.

Data <- readAsArray(Directory = choose.dir())

Data <- windowArray(Data, start = 1900, end = 2011.99)

Data <- intersectInvData(Urban, Data)

weights <- inverseDensity(Data$Inventory,Data$Array)

Temps <- solveTemperature(Data$Array,weights)

That’s it!  The variable Temps now holds the temperature anomalies for all the urban anthromes. Next, I decided to segregate the Urban Anthromes into 2 classes: those that were already urban in 1900 and those that were not. So, we have two classes. In one class the stations were urban in 1900 and remained urban in 2000. In the other class, they were non urban in 1900 and become urban in 2000. I solve the least squares equation for both sub classes and difference them.Think about what you expect the answer to be. In short, you will have two curves. The black curve is the temperature anomaly of all the urban stations that remained urban. The red line is the anomaly of the non urban stations that became urban.

The black line represents those stations that began as Urban and ended as Urban. The red line is stations that began as non urban and became urban. There is a tiny positive trend (.0007)  in this difference. What does that mean?  You tell me. Did you expect stations that went from non urban to urban to show more warming? I did. Can this negative result be explained by appealing to a variety of suppositions? Sure. As always with these things more research required.

About these ads
Categories: Uncategorized
  1. March 13, 2012 at 2:13 PM | #1

    Is the data available to do the same differencing but changing the to stations that were rural in 1950 instead of 1900 and urban in 2000? And is it a lot of work? I’m not sure why, but I think this might capture more modern growth effect better than your initial choice.

    • Steven Mosher
      March 13, 2012 at 8:54 PM | #2

      century scale data. but I have source data and can do 1950. urban growth actually starts after 1950. in china even later

  2. DocMartyn
    March 13, 2012 at 7:08 PM | #3

    This either means that there is no UHI effect or that the changes in the rural environment over the past 50 years are very similar to the UHI effect.
    This is a very nice internal control.
    Let us for the moment simply assume that Tmin is a property of AGHG and Tmax is a property of Sunlight + AGHG.
    A plot of Tmax/Tmin should curve downward as AGHG increases. Moreover, as one marches from the equator to the poles one should get increased curvature, given that the Sunlight:GHG ratio falls.
    If the rise in U-U, R-U, and R-R all give the same curves at the same latitude, then you will have found the ‘GHG Fingerprint’.

  3. steveta_uk
    March 13, 2012 at 7:25 PM | #4

    Roy Spencers work indicated that the UHI effect tapered off very slowly indeed when moving away from popultation centers. So much so that for example there is almost nowhere at all in the UK which could be considered rural.

    Perhaps you should try the same with rural covering Seminatural and Wild only – assuming there are any temp readings from those locations at all.

    • Steven Mosher
      March 14, 2012 at 12:59 AM | #5

      roy’s work cannot be replicated. Sorry. At some point I will have to write up a study on that,
      but his work wasnt done very well or documented properly.

  4. Frank
    March 14, 2012 at 12:31 AM | #6

    Steve: Is the blue line the stations that started rural and ended rural? Your definition of starting “rural” (no UHI) and ended urban (some UHI) could easily be misleading. Red could be marginally urban getting more urban (marginal to substantial UHI) and black could be clearly urban getting more urban (substantial UHI getting bigger).

    I doubt anyone is going to understand the influence of UHI on the historical climate record by arbitrarily dividing stations into rural and urban.

    Suppose we actually studied UHIs carefully: We could place properly-sited stations at various distances from the urban center of relatively isolated cities of various sizes ranging from 1,000 to 1,000,000. We’d want unambiguously rural stations north, south, east and west of each city, stations near and in the edges of the city and some closer to the center. We’d also want to put several stations around any airport (since that are such an important part of the record). Perhaps 15 stations altogether. We’d monitor the temperature and weather (especially hours sunshine and wind) hourly for a year. Then we might have enough information to estimate how UHI varies with population density and proximity to high population density and some idea of how much different other cities might be due to differences in climate (wind and cloudiness).

    From the limited information available, we might find the average station in a city of 1,000,000 has an average UHI of 1.5 degC; 100,000, 1.0 degC; 10,000, 0.5 degC; and 1,000, negligible UHI. (0.5 degC per 10X increase in population.) World population has grown about 4X since 1900 and the percentage of urban population has increased from 13% to 50% according to the UN (Wikipedia), a 15X increase in urban population. That would mean that the average station near a town of at least 1,000 people in 1900 nearby has grown 15X and its UHI increased about 0.6 degC. Perhaps UHI is half this big and drops off faster with population.

    Any scheme for dividing stations into rural and urban should be tested against possible models of how UHI varies with population and how population has changed with time. It’s easy to see how division into two groups could miss most of the impact of UHI.

    • Steven Mosher
      March 14, 2012 at 1:49 AM | #7

      Steve: Is the blue line the stations that started rural and ended rural?

      Blue is the difference

      “Your definition of starting “rural” (no UHI) and ended urban (some UHI) could easily be misleading. ”

      Actually Not. There are two classes of stations. Those that started as Urban or mixed settlement in 1900 and Those that were not urban. not urban includes all the anthromes
      seen above

      “Red could be marginally urban getting more urban (marginal to substantial UHI) and black could be clearly urban getting more urban (substantial UHI getting bigger).”

      If that were the case then the difference in trend would be distinguishable.

      “I doubt anyone is going to understand the influence of UHI on the historical climate record by arbitrarily dividing stations into rural and urban.”

      It’s not an arbitary division. Its a division based on the features that cause UHI.

      “Suppose we actually studied UHIs carefully: We could place properly-sited stations at various distances from the urban center of relatively isolated cities of various sizes ranging from 1,000 to 1,000,000. We’d want unambiguously rural stations north, south, east and west of each city, stations near and in the edges of the city and some closer to the center. We’d also want to put several stations around any airport (since that are such an important part of the record). Perhaps 15 stations altogether. We’d monitor the temperature and weather (especially hours sunshine and wind) hourly for a year. Then we might have enough information to estimate how UHI varies with population density and proximity to high population density and some idea of how much different other cities might be due to differences in climate (wind and cloudiness).”

      Much of that work has been done. If you read it you would understand that population plays less of role than you think.

      “From the limited information available, we might find the average station in a city of 1,000,000 has an average UHI of 1.5 degC; 100,000, 1.0 degC; 10,000, 0.5 degC; and 1,000, negligible UHI. (0.5 degC per 10X increase in population.) World population has grown about 4X since 1900 and the percentage of urban population has increased from 13% to 50% according to the UN (Wikipedia), a 15X increase in urban population. That would mean that the average station near a town of at least 1,000 people in 1900 nearby has grown 15X and its UHI increased about 0.6 degC. Perhaps UHI is half this big and drops off faster with population.
      Any scheme for dividing stations into rural and urban should be tested against possible models of how UHI varies with population and how population has changed with time. It’s easy to see how division into two groups could miss most of the impact of UHI.”

      uhi relationship to population is not straightforward. the 100 meters around the site matter most. population is only a proxy for UHI. more later..

      • Frank
        March 18, 2012 at 10:13 AM | #8

        Steve: Thanks for the careful reply. What papers provide the easiest way to learn what is known about UHI as it currently exists?

        isn’t what happens in the closest 100 meters called poor station siting, rather than UHI?

      • Steven Mosher
        March 18, 2012 at 12:15 PM | #9

        Frank:

        It could be poor siting or good siting. The point is that if you really want to explain every detail of the record you’d need that level of information. However, the area surrounding the site is important as well. horizontal advection can evenbe more important than the 100 meters around the site.

        Thing about 100 meters of concrete surrounded by grassland versus 100 meters of grassland in the middle of skyscrapers.
        the 100 meters matters in both, and you cannot neglect the surrounding area.

        For papers I would start with a literature review and then read all the cited work
        REVIEW
        TWO DECADES OF URBAN CLIMATE RESEARCH: A REVIEW OF
        TURBULENCE, EXCHANGES OF ENERGY AND WATER,
        AND THE URBAN HEAT ISLAND
        A. JOHN ARNFIELD*
        Department of Geography & Atmospheric Sciences Program, The Ohio State University, Columbus, OH, USA
        Received 12 July 2001
        Revised 28 August 2002
        Accepted 28 August 2002
        ABSTRACT
        Progress in urban climatology over the two decades since the first publication of the International Journal of Climatology
        is reviewed. It is emphasized that urban climatology during this period has benefited from conceptual advances made in
        microclimatology and boundary-layer climatology in general. The role of scale, heterogeneity, dynamic source areas for
        turbulent fluxes and the complexity introduced by the roughness sublayer over the tall, rigid roughness elements of cities
        is described. The diversity of urban heat islands, depending on the medium sensed and the sensing technique, is explained.
        The review focuses on two areas within urban climatology. First, it assesses advances in the study of selected urban
        climatic processes relating to urban atmospheric turbulence (including surface roughness) and exchange processes for
        energy and water, at scales of consideration ranging from individual facets of the urban environment, through streets and
        city blocks to neighbourhoods. Second, it explores the literature on the urban temperature field. The state of knowledge
        about urban heat islands around 1980 is described and work since then is assessed in terms of similarities to and contrasts
        with that situation. Finally, the main advances are summarized and recommendations for urban climate work in the future
        are made.

        start with the energetic basis of the urban heat island (oke 82)

        SIMULATION OF SURFACE URBAN HEAT ISLANDS
        UNDER ‘IDEAL’ CONDITIONS AT NIGHT
        PART 1: THEORY AND TESTS AGAINST FIELD DATA
        G.

        Complete Urban Surface Temperatures
        J. A. VOOGT* AND T. R. OKE

        SIMULATION OF SURFACE URBAN HEAT ISLANDS
        UNDER ‘IDEAL’ CONDITIONS AT NIGHT
        PART 2: DIAGNOSIS OF CAUSATION
        T. R. OKE,’ G. T. JOHNSON,’ D. G. STEYN’ and I. D. WATSON

  5. March 14, 2012 at 1:30 AM | #10

    I get around 1950 stations that have data in 1900 and data in 2011. You get 8000?

    • Steven Mosher
      March 14, 2012 at 1:51 AM | #11

      then you dont know what you are doing. its a least squares solution. you solve for all the data you have. not just stations that have complete records.

    • Steven Mosher
      March 14, 2012 at 2:47 AM | #12

      I’ve got a bit more time to explain. There are 8000 stations ( roughly) that are ‘urban’ or
      mixed settlements. Some of those stations will have data from the begining of the period to then end. Some will have data for only portions of record. The least squares approach works
      by making an estimate at every month of the data available for that month. The data is weight based on an area weighted inverse density. So you have two series:

      1. A selection of stations that started as urban and ended as urban. These stations are weighted and averaged.

      2. A selection of stations that started as non urban and transitioned to urban. These stations
      are weighted and averaged.

      The difference between those is our measure of interest.

      Going forward, of course, I can look at different time periods, as jeez suggests above.

      I can also look at only LONG series which is what you are looking at. Long series have the problem of not being spatially representative so that spatial bias can overwhelm the signal.

      That’s something I can look at BUT, an answer that is not spatially complete really doesnt answer the question of GLOBAL bias.

  6. phi
    March 14, 2012 at 1:52 AM | #13

    Changes in perturbations are independent from initial state of urbanization. This is an interesting result.

    • Steven Mosher
      March 14, 2012 at 2:50 AM | #14

      Phi,

      yes. I think the next step is to isolate and look at long series. Also, to come up
      with Anthromes for 1950 or 1980. That’s a bit harder, but doable.

  7. March 14, 2012 at 2:31 AM | #15

    So you are comparing stations that don’t have data in 1900 to Anthrome data for 1900?

  8. March 14, 2012 at 2:48 AM | #16

    I think it would be more interesting to compare nearby station pairs that have a different end point.

    Like Port Townsend vs Olga

    https://sunshinehours.files.wordpress.com/2012/03/qc-avg-port-townsend42613-1900-2011-red-12.png

    https://sunshinehours.files.wordpress.com/2012/03/qc-avg-olga-2-se42788-1900-2011-red-10.png

    • Steven Mosher
      March 14, 2012 at 2:54 AM | #17

      data snooping is not a good way to do your analysis. You will always find what you want.
      In dendro they are called bristlecone pines. I can, for example, find urban sites that have
      cooled. I can find rural sites that have warmed. I can pick cherries all day long with 36K stations, but none of that picking says anything about GLOBAL bias. The question is
      “have urban station ON AVERAGE warmed more than rural stations. You will find, you almost
      HAVE to find stations that present answers that go opposite of what you expect. Partly because UHI itself can be negative and partly because the temperal feild is not completely homogenous

  9. March 14, 2012 at 3:39 AM | #18

    Its not cherry picking to note some stations have cooled, or some months have cooled while others haven’t or some stations warmed for a while and are now cooling.

    I think for a theory to be valid it should work on a micro level too. You disagree. Fine.

    You’ve said in the past that it doesn’t matter how few stations you use, you end up with the same answer.

    Does it matter if you just use the 1900 or so stations that have both 1900 and 2011 data (it would probably be better to stop at 2000 but I don’t have that number)?

    Does it matter if you graph continents separately? 5×5 grid squares?

    I find that it does matter which regions you look at. Some regions stopped warming in the 1980s. Some never seemed to warm much at all. Some stopped in 1998. Some kept warming all the way to 2006.

    • Steven Mosher
      March 14, 2012 at 10:12 AM | #19

      Does it matter if you just use the stations that have data in both 1900 and 2000?
      That’s phase II of the analysis. You see rather than cherry pick through the data to
      find an answer I like, I apply a series of analytical methods without snooping around
      in the data. Start with global. Proceed to long series. Then look at regressions. Then
      look at exemplars.

      And I would NOT just look at stations that had data in 1900 and data in 2000. Why? because of possible lacuna in the middle. I’d start by requiring that 75% of the data
      being prsent between 1900 and 2000. And then, I’d push that to 100%.

      You will see in the next post that with long series the answer is the same. Go figure.

      If you want an answer to the global question “how is the GLOBAL average biased” you
      have to look at more than a single station. Using your method, I could just start and end the discussion by focusing on those cities which are documented to have negative UHI.

      See the next post and then we press deeper. remember we are not looking for single cases of UHI. we are looking to see if the bias present at SOME infects the ENTIRE record.

      Or what happens if we only use rural stations. answer? not much. question? why…

  10. BarryW
    March 14, 2012 at 4:10 AM | #20

    The fly in the ointment with this is that urban in 1900 and in 2000 have different meanings from a standpoint of heat. 1900 -few cars, 2000 – gobs of them. 1900- no air-conditioning, 2000 it’s all over the place. 1900 – no airports, 2000- everyones got one. 1900 – skyscrapers minimal, 2000- depends on which city and whether your in the urban core. 1900 – urban sprawl? Only on the train/streetcar lines. 2000- well we know what that’s like.

    So my point is that you’ve shown that you can’t see it by a change the definition of an area from rural to urban and it may well be to the fact that cities were more rural in their heat signatures than they are now. And using a 10k grid size? Kinda coarse what? Now if you had a temperature transect for a city from 1900 and now….

    • Steven Mosher
      March 14, 2012 at 10:23 AM | #21

      One of the little understood points about UHI is that it varies with a whole host of parameters. Cars in 2000? It depends upon the city. air conditioning ? again depends on the city. Sky scrapers? depends on the city. In the far east? yes. In india, not so much. basically the temperature at a given location is a function of LST within 100 meters AND heat advected horizontally from adjoining areas.

      A city transect will not help you. What the transect will show is that UHI varies across the urban landscape. But the temperature station is located at one place. That place may be a cool part of the city or a hot part of the city.

      Now its funny that people complain about a 5 minute area. The only study that shows any uhi on a global scale used a 5 degree grid ( McKittrick). In any case, At some point I’ll go below this 5 minute scale to the 300 meter scale.

      • BarryW
        March 16, 2012 at 5:48 AM | #22

        The problem is that the transect is only good for the time it was taken. If the station was located outside of the UHI bubble early in it’s history and the bubble grew as the city expanded it might be inside of the bubble when the modern transect was taken. Consider Washington National Airport. When it was constructed (early ’40s) Washington was a relatively small city and the location of the airport was in Northern Va which was relatively rural at the time. It’s now in the heart of a concrete and asphalt jungle. I know from experience that snow totals for the airport almost always somewhat lower and temperatures higher than the surrounding rural areas when I lived there. Without being able to account for that, how can you say there is no effect?

  11. peter azlac
    March 14, 2012 at 1:46 PM | #23

    An interesting analysis that goes somewhat against the CET data where they found a large correction was needed for Tmin temperature values but none for Tmax. Berley found that around 30% of the stations showed a cooling trend – Motl found a similar level in the lower Hadcrut series, as did Tony Brown who, with others, also found that these stations are reasonably well distributed globally but they are also seen to be “well mixed” with warming stations in the SE USA. My question is what happens if you do your analysis for these stations separately from those that give the warming trend? Also, if you apply Empirical Decomposition Analysis or Harmonic Analysis to the warming and cooling series can you extract the reason why they differ – I have no skill with R otherwise I would do this!

    • Steven Mosher
      March 14, 2012 at 9:50 PM | #24

      You misunderstand the 30% of cooling stations. The chart did not explain very well that those stations could have different start dates and end dates.

      As for tonyb stuff, see my earlier posts here. As for Motl and others they need to learn to post code so that their results can be replicated. otherwise I have no time for them and disregard their claims until they can do a proper job of things

      selecting cooling stations and then data mining for a reason isnt what I would do with historical data. You might select a subset and put them through that process

    • March 16, 2012 at 7:56 AM | #25

      Hundreds of stations in the Eastern US have been cooling since 1900.

      https://sunshinehours.wordpress.com/2012/03/15/cooling-since-1900/

      And while the actual station data is still unavailable, CRUTEM4 shows a tremendous amount of cooling int he same locations (as well as others).

      http://www.metoffice.gov.uk/hadobs/crutem4/index.html

  12. March 14, 2012 at 6:51 PM | #26

    What packages are required to use BerkeleyEarth?

    • Steven Mosher
      March 14, 2012 at 9:51 PM | #27

      requires bigmemory, biganalytics and RghcnV3.

      If you dont have 4Gb of memory…. Its tough going. even with 4 I have to work in snippets

      • March 15, 2012 at 4:07 PM | #28

        Haha, ok. Thanks. I think I will spare my laptop for the beating then.

  13. Frank
    March 24, 2012 at 1:11 PM | #29

    Steve: Thanks for the references above. Unfortunately, knowing something about the physical mechanisms that contribute to UHIs tells us nothing about how UHI has contaminated the historical record – unless one has a record of how the urban landscape has changed with time. We probably don’t have the necessary information.

    I was imagining a more pragmatic approach. If I randomly place stations in and around various urban areas, how big is the urban bias likely to be (compared with surrounding “truly rural” stations)? Can I develop a model that predicts how the bias changes with population or population density or GDP? Then the model would need to be tested on other cities. That might put a maximum on UHI contamination in the historical record.

    The studies of Oke and others give us some hint of how hard the problem might be. How much do cities differ from each other in the parameters Oke has shown to be important? Even if all cities were roughly the same, the rural area surrounding cities are not. How big is the rural/urban difference when the surrounding rural land is forested vs farmed vs semi-arid grassland vs etc? How many fundamentally difference types of urban vs rural situations do we need to understand before we can say something useful about the amount of UHI experienced by the average urban station today? Has UHI (say for a given population density) changed with time (1920 vs 1980)?

    • Steven Mosher
      March 24, 2012 at 1:38 PM | #30

      The approach is pretty simple. select only rural stations. Also, we can use historical land use

      “I was imagining a more pragmatic approach. If I randomly place stations in and around various urban areas, how big is the urban bias likely to be (compared with surrounding “truly rural” stations)? Can I develop a model that predicts how the bias changes with population or population density or GDP? Then the model would need to be tested on other cities. That might put a maximum on UHI contamination in the historical record.”

      Well you would not use population because its not population that causes UHI. Its what the population does. So a 100K people living in 5sq km transform the landscape differently than 100K people
      living in 200 sq km. As far back as 1973 Oke understood that population was only a PROXY for UHI INTENSITY.. not average UHI, not UHI over a long period, but UHI maximum.
      Further, he noted that the effect was different in American than it was in Asia and later it was found to be different in southern america. Population has really been replaced with better measures, but, it can explain SOME of the effect

      “The studies of Oke and others give us some hint of how hard the problem might be. How much do cities differ from each other in the parameters Oke has shown to be important? Even if all cities were roughly the same, the rural area surrounding cities are not. How big is the rural/urban difference when the surrounding rural land is forested vs farmed vs semi-arid grassland vs etc? How many fundamentally difference types of urban vs rural situations do we need to understand before we can say something useful about the amount of UHI experienced by the average urban station today? Has UHI (say for a given population density) changed with time (1920 vs 1980)?”

      Tough questions. The land use difference is covered by imhoff. Forested gives you the biggest differential. semi arid? no real UHI ( actually SUHI )

  1. No trackbacks yet.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Follow

Get every new post delivered to your Inbox.

Join 31 other followers

%d bloggers like this: