Open Mind

Return to Shelby County

April 19, 2007 · 7 Comments

In a recent post, I investigated the temperature records of stations near Shelby county, Tennessee. Three of the station data series were from the Global Historical Climate Network (GHCN), the other two from the Southeast Regional Climate Center (SRCC).

It turns out that three of the stations, Covington, Memphis, and Holly Springs, showed extremely similar temperature changes. But two of the stations — Hernando and Moscow — showed noticeable deviations from the pattern indicated by the other three. Moscow showed large deviations from 1974 to 1980 which the other stations did not, while Hernando didn’t show the general long-term warming trend from 1980 to the present that the other four stations show. Furthermore, while the three “matching” stations showed very similar rates of warming during the modern global warming era (1975 to the present), the other two stations indicated at least the possibility of very different warming rates.

A reader dug deeper into the station histories, and discovered that both the Moscow and Hernando stations has experienced station moves. Also, both these station reports had gaps, with one third of years in the Moscow data series incomplete, and one fourth of years in the Hernando series. So there is reason to doubt their correctness.

This raises a number of questions. Are the data series for Moscow and Hernando really inaccurate? If so, are far more series inaccurate? Are these series correct, and the warming rates really are different from nearby regions? Are they correct, and warming at about the same rate as nearby stations, but random fluctuations caused an apparently different trend?

The best way I know to investigate this issue is: get more data. My initial investigation was limited to the nearest stations I could find to Shelby county, Tennessee. That limited me to three stations from GHCN, so I looked for more data and found two more stations from SRCC. But even so, with only five stations it’s hard to answer the questions raised. So, let’s look at a larger area.

I’ve widened the search area to a latitude-longitude grid box centered on Memphis, Tennessee, of total size 4 deg. latitude by 4 deg. Longitude. This makes the grid-box 276 miles tall and 226 miles wide, big enough to include far more than just five stations. In fact, within that area there are 27 station data series from GHCN alone. These come not just from Tennessee, but from the surrounding states of Kentucky, Mississippi, Georgia, Arkansas, and others. These data do not include Moscow, TN, because that’s not part of the GHCN.

What’s the warming rate for these stations? Fortunately, I have a program which computes the trend rates for a large number of data files automatically, so it wasn’t at all inconvenient to run these numbers. This time, instead of expressing the rate as deg.C/century, it’s expressed as deg.C/yr. Take these numbers and multiply them by 100, and they’re in the same units used for the original post. And here are the results (I’ve made the graph bigger for clarity, but that means you’ll need to click on the graph to get the full view):


The average warming rate for all 27 stations is 0.039 deg.C/yr (3.9 deg.C/century), and is indicated by a red line. The consistency of the estimated warming rates from station to station is quite impressive, except for three of the stations: Hernando, Pocahontas, and University station.

Hernando is one of the stations in the original post; we already have evidence of data problems from that location. We can now see that its estimated trend rate is definitely out of line with the vast majority of stations in its area. We also see that two other stations in this grid-box are out of whack; their data simply are not consistent with the majority of the results.

More importantly, we see that the results of the other 24 stations are very consistent — so consistent that we can confirm, without doubt, that the density of observing stations in this part of the world is more than sufficient to identify the true trend in average temperature change during the modern global warming era.

Of course, the three out-of-whack stations will alter the average we compute. But they’re only a small part of the average, and since two of them give lower-than-realistic results while one gives higher-than-realistic results, their errors tend to cancel each other at least in part. That’s a general result in statistics, that if the errors show no bias for being above or below the truth, then on average they’ll cancel out.

For this grid-box centered on Memphis, we have 24 stations with very consistent results, 3 stations with strong evidence of data problems, and an extremely confident estimate that the warming rate is around 3.9 deg.C/century. We also have more than enough reporting stations to do the job.


A reader asked some questions, so I’ll answer them.

You say:

The consistency of the estimated warming rates from station to station is quite impressive

And I say, since there is actual data involved, what is there to estimate?

Temperature change at any particular location involves two factors. One is the trend. The other is the fluctuations. We can see this in day-to-day weather; as the seasons progress, there will be day-to-day fluctuations in temperature, so one day can be dramatically hotter or colder than the preceding. But there is also a trend: the average temperature will warm from winter to summer. Here, for example, are daily mean temperatures for Portland, Maine, from March 1 to July 31, 2006 (time is measured in “Julian Days”):


We can see lots of day-to-day fluctuation, but we can also see a steady trend; from March through July in Portland, the temperature gets warmer.

The same is true of annual avereage temperature; it has both trend and fluctuations. The fluctuations are natural variation; they’re caused by weather patterns, el Nino/la Nina, and a lot of other factors. Nobody who is knowledgeable about climate or weather, doubts the existence of natural variation.

If we want to know what the trend is, we have to separate the natural variation from the trend. We do this by statistical tests which can loosely be called “trend analysis.” Because of the fluctuations, the result of trend analysis isn’t an exact measure, it’s only an estimate. If there were no fluctuations, we wouldn’t need trend analysis, and there would be no uncertainty in our measurement of the trend.

Trend analysis of the month of April alone, indicates that during that month the trend led to a change in average temperature of 2.8 deg.C. But from April 24th to April 25th, the temperature got colder by 7.3 deg.C. Would you conclude that we can’t reliably deduce the April trend, because there are natural variations larger than the trend?

You have also ignored my original point, which is that temperature varies significantly within the area you consider one data point to be adequate representation, and which your previous post on the subject proved.

There are indeed significant variations in temperature over regions much smaller than a county. But as this post shows, there are no significant differences between the temperature trend over very large regions, in this case, an area 276 by 226 miles — much larger than a single county.

I also fail to see how you can call the data you have presented as having consistent warming. The data points on your chart vary by at least as much as your estimated warming. Considering that those points, I assume, are the result of more smoothing and averaging, it’s really kind of meaningless.

Because of the fluctuations, the result of trend analysis isn’t an exact figure, it’s only an estimate. So there is some uncertainty in our estimate of the trend, but trend analysis also enables us to compute how much uncertainty there is.

Knowing the most likely estimated trend, together with the uncertainty, enables us to compute a range of values which almost certainly includes the true trend value. This range is indicated by the error bars; the true trend value is very likely (95% probability is the standard in scientific analysis) to be within the range indicated by the error bars.

And, because of the fluctuations, it’s just about impossible for the estimated trend to be identical for nearby recording stations, even if the actual trend is exactly the same. It’s like flipping a coin 1,000 times and counting the number of heads, in order to estimate the chance of heads versus tails. Then flip the same coin another 1,000 times to get another estimate of the probability of heads versus tails. It’s extremely unlikely that you’ll get exactly the same estimate from both tests, in spite of the fact that we know the probability of heads is the same — it’s the same coin!

The natural variations can be very different from place to nearby place, but the trend is very nearly identical. In fact, for the 24 stations in the grid box which don’t show signs of data problems, the value 3.9 deg.C/century is well within the probable range for every one of them. Every station is completely consistent with the estimate 3.9 deg.C/century, and the differences between the estimates are exactly what we would expect if the warming rates were the same. Statistically, it’s pretty much impossible that they will all give precisely the same estimate.

One final note: no smoothing was applied to this data. Yes there is averaging, but only because we’re analyzing the annual average temperature; data from different years were not averaged together.

Categories: Global Warming · climate change

7 responses so far ↓

  • the Grit // April 19, 2007 at 9:48 pm

    Hi tamino,

    You say:

    The consistency of the estimated warming rates from station to station is quite impressive

    And I say, since there is actual data involved, what is there to estimate?

    You have also ignored my original point, which is that temperature varies significantly within the area you consider one data point to be adequate representation, and which your previous post on the subject proved.

    I also fail to see how you can call the data you have presented as having consistent warming. The data points on your chart vary by at least as much as your estimated warming. Considering that those points, I assume, are the result of more smoothing and averaging, it’s really kind of meaningless.

    the Grit

    [Response: See the UPDATE to this post.]

  • nanny_govt_sucks // April 20, 2007 at 3:31 am

    There may be third variables that you still need to account for. I don’t think you can conclude much as of yet.

    Is there a correlation between population changes (or amount of nearby asphalt coverage) and warming rates for the time period in question for these different stations?

    [Response: There may indeed be other variables at work, but the correlation between both the fluctuations (as shown in the previous post) and the trends (as shown in this post) are statistically strong enough that we can draw meaningful conclusions with very high statistical confidence. Shelby county Tennessee *is* getting warmer.

    If you want to study the relationship between population changes and trend rates for this area, the data are easy to find and download. Do the work and report the results.]

  • fergusbrown // April 20, 2007 at 11:48 am

    It occurs to me that this connects fairly directly with Roger Pielke Sr.’s challenge to the reliability of the surface temperature record. Is this deliberate?
    To what extent do you see this as a refutation of Roger’s challenge, at least in part?
    Finally, Roger has frequently pointed out examples of station inconsistencies such as those you mention here. Is the size of this (perhaps with others) inconsistency sufficient to effect the overall trend, either on a US or Global level?
    Neat post, Tamino, keep it up!

    [Response: I wasn't aware Pielke had challenged the reliability of the surface temperature record. The posts on Shelby county, and "A Tale of Two Cities," are meant to address the issue of correlation between nearby stations, not the overall reliability of surface records. So there's no *deliberate* relationship to Pielke's challenge.

    All inconsistencies affect the estimate of the overall trend, so we know the estimated global average temperature trend is not perfect (of course we already knew that, because the data are the sum of both trend and fluctuations). But as this post shows, some of the erroneous data make the trend seem to high, some make it seem too low. Consider, for example, station moves: if we move the station to a warmer location it will introduce a false warming trend into the data; if we move to a cooler location it will introduce a false cooling trend.

    Most delusionists seem to think that erroneous records must necessarily introduce a false trend into the overall average, but the fact is that as long as there's no bias in the errors which favors one direction over another, the "expected value" (in the statistical sense) of the net contribution from the errors is zero. In this case, the biggest effect of erroneous data on the average from multiple stations, is to increase the uncertainty of that average, so the error bars get bigger but the range still includes the true value.]

  • Adam // April 20, 2007 at 3:02 pm

    It’s worth remembering that certain networks do site-checks and data checks to keep the quality as high as possible. I don’t know, but I’d bet the SYNOP network is the main input to the global temp. record?

    [Response: As far as I know, the GHCN is the data source used for land estimates by NASA GISS. They do a number of quality checks for all stations, and they apply a number of known corrections to U.S. and near-U.S. (southern Canada and northern Mexico) data. I believe HadCRU (Hadley Centre/Climate Research Unit) uses a different data set.

    I downloaded the GHCN data a few years ago, and I'll bet there's a more up-to-date version available. One of these days I'll update.]

  • the Grit // April 22, 2007 at 9:10 pm

    Hi tamino,

    You have presented an informative explanation of how you are calculating your statistics, but I would also like to know why we should care? Really, assuming that there is an upward trend in average temperature that can’t be explained by population growth in this area, and, if I understand you correctly, it’s not an everyday thing, but a few sporadic extra warm days, big deal. If you can show that only our winters will be warmer, I will applaud Global Warming, as that will be less firewood I have to cut and split.

    Saying this another way, the trouble I have with the way the science is presented is mainly in the generalizations. If every day is not going to be warmer everywhere, then come up with a better name.

    the Grit

    [Response: I agree that "global warming" is far from an ideal name for the phenomenon. The phrase conjures images of warm, pleasant sunny days even in winter, and as you say, that doesn't sound like a bad thing. Another poorly descriptive name is "greenhouse effect." Greenhouses keep your garden warm by blocking air circulation, not by blocking infrared radiation (although they can do that too). The warming action of greenhouse gases has nothing to do with blocking air circulation. If it were up to me, I'd adopt different terminology. But I'm not in control of the English language; the names have been adopted over time and aren't likely to change any time soon.

    The real problem with global warming rapid climate change is not that the climate will be too hot or too wet or too dry -- I doubt there is such a thing -- but that it will be *different*. Species adapt to the way things are, and we've got a helluva lot riding on the way things are right now. When circumstances change we adapt to that too, but it takes time; if the change is rapid, we (and other living things) have a hard time keeping up.

    For example, we can handle deserts. We even have people living in deserts in the American southwest. We also have a lot of farmers growing food in the fertile (and reasonably well-watered) midwest. But if rapid climate change causes the midwest to turn into a desert, we'll have to adapt, and that can be tremendously costly, both in terms of money and human suffering. It'll be hard enough for us in the U.S. Suppose your farm dries up and your land becomes a dustbowl? Sure, you can move somewhere else and farm some other land, but imagine the hardship and poverty you'll suffer as a result. From the news reports I see, it's happening right now to farmers in Australia. Now imagine yourself a subsistence farmer in northern Africa.

    Another example: sea level rise. If it goes up 20 feet, our coastal cities will be underwater. Imagine you live in Miami. Sure you can move inland and rebuild -- but again, the cost, the poverty, the disruption will be a tremendous burden. Now imagine your family farm in Bangladesh underwater; you were already on the edge of poverty, now you're on the edge of extinction.

    I expect it'll be very hard on us in the developed world. It'll be far harder on the billions in the undeveloped world. But they're not the ones who have brought this about, we are. To quote from Cain, "Am I my brother's keeper?"]

  • the Grit // April 25, 2007 at 8:48 pm

    Hi tamino,

    On the other hand, imagine that the Alarmists are wrong, and we invest trillions of dollars in trying to prevent something that either isn’t happening, or is beyond our control. The measures suggested to curb greenhouse gas emissions will kill almost as many people in developing countries as your worst case scenario, from starvation, lack of progress in medical practices, lack of mosquito control, and many other things that will be sacrificed in the name of controlling carbon emissions.

    My scenario is a certainty; yours is a guess.

    the Grit

    [Response: Your scenario is absolutely NOT certainty. That is, perhaps, the most pernicious lie in this debate.

    I'm not accusing you of lying, I'm sure you actually believe it -- but this whole idea has been planted by the fossil-fuel-industry CEOs who want to use your fear of economic downturn to keep lining their pockets with record-breaking profits. Reducing carbon emission *can* be done without destroying our economy, the estimates of "trillions" are either on century-long timescales or just plain inflated for propoganda purposes, and the implication that anyone is proposing measures to sabotage life-saving efforts for underdeveloped nations is not just a lie, it's an *insult*. Not just to my beliefs -- to your intelligence.

    Don't let the folks who spend billions upon billions upon billions bribing politicians and paying lobbyists get you caught in their web of deceit. Think long and hard, my friend. You've been had, and if you don't realize it soon, then when you do finally comprehend that that you've been swindled into bringing about your own downfall, it just might be the most bitter moment of your life.]

  • the Grit // April 28, 2007 at 9:07 pm

    Hi tamino,

    I started writing a reply, but it got out of hand for the tiny window given for comments. Thus, I posted it on our blog,

    Oh, and thanks much for keeping the discussion civil. You would not believe how refreshing that is!

    the Grit

Leave a Comment