The issue arose of the correlation between temperature patterns in nearby locations. I posted on this topic, showing the strong correlation between annual average temperature patterns in Oxford, England and Paris, France, separated by a distance of 413 km (257 miles). However, a reader stated that he had witnessed significant weather differences within the confines of his own county.
This raises the question, is this testimony simply a false impression based on anecdotal evidence, or is his region actually subject to such strong local variations that the patterns in nearby areas are not really correlated? It seems appropriate to investigate temperature patterns in and around his location: Shelby county, Tennessee.
Shelby county is home to Memphis TN, rightly proud as the cradle of Elvis Presley. It occupies the southwest corner of the state of Tennessee, bordered by the Mississippi river to the west and the state of Mississippi to the south:
Unfortunately, the Global Historical Climate Network (GHCN) does not contain data from any stations in Shelby county. But there is data from GHCN for three locations not too far from Shelby county: Covington TN, Hernando MS, and Holly Springs MS. I also managed to find data for Memphis itself, as well as not-too-distant Moscow TN, from the Southeast Regional Climate Center (there are many Tennessee stations in this network, but of those in the area, only these two had long enough station records for meaningful analysis). This gives five station records, one in Shelby county, the other four nearby. Their locations are indicated by red boxes in this map:
Each station record consists of monthly average temperature, and all of them include at least the period from 1950 to 2000 (the Memphis data go all the way through 2006). First let’s look at the data smoothed using a wavelet transform:
Memphis is consistently the warmest location. A great many similarities are evident in the increases and decreases, although differences are also visible. All five stations show:
- rise to a peak in 1954
- decline to a trough about 1960
- rise to another peak about 1965
- fall to another trough about 1969
- rise to a peak about 1974
- fall to a trough about 1978
At this point Moscow TN departs from the pattern. The other four stations show:
- rise to peak around 1981
- fall to trough around 1984
- rise to peak around 1987
Moscow, on the other hand, which cooled more than the other locations up to its 1978 trough, warms considerably more than the other stations during this period, to a peak in 1986. It’s also worth noting that Hernando MS and Memphis TN, which were very close until 1968, diverge; they still show ups and downs at the same time, but from 1969 onward Memphis is noticeably warmer than Hernando. From 1987 to the present, all five stations again show the same pattern of rises and falls.
The different locations show different temperatures, Memphis being the warmest. We can bring them on to the same “scale” by computing temperature anomaly. Anomaly is the difference between the temperature, and its average at the same location during a reference period. Warmer-than average temperatures give positive anomaly while cooler-than-average give negative anomaly. I chose as reference period the entire time span of data, computed the temperature anomaly for each month, then computed annual averages. Graphing these data gives a plot which is far less smooth (the wavelet transform smooths the data), but shows clearly the strong similarity of the rises and falls of annual average temperature for the different locations (years which did not include data for all twelve months of the year are omitted from the graph):
From this we see that the temperature changes for Covington, Holly Springs, and Memphis are strikingly similar. Hernando and Moscow also follow quite similar patterns, mimicking the same ups and downs most of the time, but not as strongly as the other three; Moscow shows large deviations from 1974 to 1980 which the other stations do not, while Hernando doesn’t show the general long-term warming trend from 1980 to the present that the other four stations show.
Of special interest is the trend rate during the modern global warming era (1975 to the present). We can use the raw data (monthly temperature anomaly) for each station to compute the trend rate, as well as the uncertainty in our estimate of it:
Four of the five stations show warming from 1975 to the present, but Hernando MS shows a very slight cooling trend. However (as the error bars illustrate) the trend for Hernando is not statistically significant — we can’t really tell whether the trend is negative, positive, or zero.
All in all, the five stations show very similar patterns of temperature change; they’re definitely strongly correlated. But they also show differences, particularly the dip in Moscow temperature from 1974 to 1980 and the lack of a detectable trend in Hernando during the modern global warming era. The similarity in pattern for Covington, Holly Springs, and Memphis is nothing short of striking.
It’s also clear that the density of recording stations for this part of the world is sufficient to establish a pattern: consistent warming during the modern global warming era, albeit without the participation of Hernando. We can say with confidence that Shelby county Tennessee has, over the long term, been warming at a rate of about 3 deg.C/century.
13 responses so far ↓
Steve Bloom // April 15, 2007 at 11:52 pm |
So Hernando would have been a good cherry-pick for our friends at CO2 Science.
[Response: It would certainly be their style. But they go even further: they choose a time frame (1930-2000) which is unrelated to modern global warming, but optimal for finding cooling trends IF you restrict yourself to U.S. stations (and the U.S. is only about 1.5% of the surface of the globe). And even at that, most U.S. stations show warming, so they just pick the ones that make the case go their way. I've done several posts showing that if you take the station reported on by co2science, and extend the time frame before 1930 and/or after 2000, the series shows warming rather than the cooling they tout.
Of course the REAL point is that relying on single stations is a very bad way to estimate global trends.]
guthrie // April 16, 2007 at 5:35 pm |
Moscow seems odd. I wonder why it dipped so far between 1970 and 1980? Was there a change in instruments? Has the place been urbanised?
Or is it a chance artefact,w hich shows up so much because of the small number of places used?
[Response: The data from Moscow are odd for a lot of reasons. For one thing, there are a lot of missing values; of the 58 years in the Moscow record, 19 of them are incomplete (i.e., they don't include all 12 months of the year). The same is true for Hernando; 27 of 102 years in the station record are incomplete (2 years are missing entirely). Also, Moscow shows a greater warming trend than any of the other stations while Hernando shows a lesser trend -- enough greater and lesser that they're "out of line" with the trend indicated consistently by Covington, Holly Springs, and Memphis.
It's possible that there is some problem with the data from Moscow and/or Hernando -- changes in instruments, siting, or procedures could have produced such effects. But it's not my job to evaluate the correctness of the data; readers are free to draw their own conclusions.]
Steve Bloom // April 16, 2007 at 8:26 pm |
Yes. You’d think it would have become too enbarrassing by now to not update beyond 2000 (which I assume would knock out many if not most of their exmples since the 2000s have been a rather toastier decade for the US than the relatively cool 1990s). This combined with the slanted unlinked summary business is completely shameless.
the Grit // April 16, 2007 at 10:21 pm |
Hi tamino,
First, allow me to sing your praises as to the speed and impressiveness of your work! Bravo!
I would also thank you for proving my point. While we can toss the data for Memphis out, considering that it is a rapidly growing urban area, and variations in its temperature history are most likely attributed to an increase of area covered by asphalt than anything else, which includes the entire area you found data for to some extent, your findings for the rest of the area in question still shows a fluctuation greater than the 0.6C Global Warming to date claim. Part of this, I suspect, is due to changes in farming practices and urban sprawl.
While I lack the ability to easily capture the NOAH weather reports each day, it would seem that I am not alone. Your inability to find detailed and reliable data for a region such as Shelby County, which is in the top 20 metropolitan areas in the US, only increases further my doubts as to the completeness of our data on a global scale.
As I’ve had occasion to say more than once today, slow and steady wins the race. Thanks once again for the attention and effort!
the Grit
[Response: We certainly have a different interpretation of the data! The results for Memphis are *strikingly* consistent with the results from Covington and Holly Springs. And of course there's the fact that the trend for Memphis during the modern global warming era (1975-present) is actually less than that for Covington and Holly Springs (and Moscow, although there's reason to doubt the Moscow data).
The really questionable data are Moscow and Hernando, which have a large number of missing values and don't conform very precisely to the pattern of the other three (but are still strongly correlated with them). Even including those stations, we still have strong enough correlation between temperature anomaly for all five stations to establish the validity of global averages based on this density of reporting stations.]
Lee // April 17, 2007 at 1:42 am |
Grit, it appears as if you are making a different objection here than the one to which Tamino responded in this thread.
Your first objection was that the coherence of these data sets was so poor as to require ore data stations than exist. Tamino has shown clearly with these analyses, that there is an impressive correlation in the annual temperatures of stations separated by large distances – opposite sides of a county, here. That answers your earlier objection, seemingly way more than adequately.
You objection here seems entirely different – that the year to year variation is larger than the calculated warming trend, so the warming can’t be real. This does not address Tamino’s demonstration of coherence across separated stations – it isn’t even relevant to it. It also reduces to a statement that trends in a noisy time series cant be detected unless the trend exceeds the noise outright, and this is simply not true. If a line fit gives positive slope, and is statistically distinguishable from a slope of zero, then there is a detectable trend, even if it is smaller than normal annual variation.
Adam // April 17, 2007 at 1:23 pm |
You can find out about stations on the NWS Cooperative network (such as Hernando & Moscow) from here (this is Moscow, you can search for Hernando):
http://mi3.ncdc.noaa.gov/mi3qry/map.cfm?fid=18492&stnId=18492&PleaseWait=OK
Which shows that those two stations have both moved (I didn’t check others). The Google map facility shows the Hernando station is probably in someone’s back yard. I don’t know what checks are done on the COOP observation stations but I wouldn’t be surprised if there are some exposure issues with some of them.
Also, whilst looking I did see a reference to homogeneity checking within the NCDC data netwaork, but forgot to note the URL so lost it. But there’s a pdf on their site somewhere.
the Grit // April 17, 2007 at 6:23 pm |
Hi Lee,
I was speaking of the difference between the various stations, which in many cases is more than a decree Celsius. That was my point. The area that is being assumed to be adequately sampled with one data point, as tamino’s presentation shows, is much to large. If you look at the 1975 to present warming chart, this is readily apparent, with a difference between the hottest and coolest stations of over 5 C.
the Grit
[Response: There are two problems with this argument. First, the two most extreme stations definitely have data problems. Both Moscow and Hernando have a large number of incomplete years (one third of the years for Moscow, one fourth for Hernando), and as Adam discovered, both stations experienced a station move. These non-climate factors not only can, they *will* alter the results of trend analysis.
Second, and most important, you're definitely falling prey to a statistical fallacy. Look at the *error bars*. Trend analysis doesn't give us the perfect true value of the trend rate, it gives a *range* of values, such that the true value is very probably in that range. Even if we accept the Moscow and Hernando data at face value, these data are entirely consistent with Hernando warming at a rate of 2.7 deg.C/century (upper end of the range) and Moscow warming at 3.2 deg.C/century (lower end of the range). If we omit the Moscow and Hernando data, all three remaining stations are consistent with warming as little as 1.9 deg.C/century or as much as 4.3 deg.C/century.
In fact, even if all stations were warming at exactly the same rate, statistically it's astoundingly unlikely that they will all give precisely the same *estimated* rate. It's like flipping two coins, both of which have *exactly* the same probability of heads or tails (50-50 for both coins). If we flip each coin 1,000 times, the chance of getting exactly the same number of heads for each coin is very tiny indeed.
The most likely interpretation of these data is that the trend in Hernando is on the high end of its range, that in Moscow on the low end of its range, and the extreme nature of the estimated values for these stations are due to the station moves.]
Adam // April 18, 2007 at 9:02 am |
I realised that the link I gave didn’t go straight to Moscow, but it’s simple to do: click the browse as guest button, do a search on Moscow (or Hernando) and and select the correct station from the list.
The station location is often listed at the same lat/long for the different locations, but the location doesn’t specify seconds until after ‘97 and the height changes. In ‘81 there is a several month change in lat/long (in minutes) before the station returns to what looks like it’s previous position (based on the height). The station moves again in ‘82 and I think remained in the same place since then.
The ‘97 “move” looks to be just an update to a more precise location, this also coincides with a change of observer (or station manager) as well as change in status etc. The previous observer was in place from the mid ’70s to ‘97.
Hernando made a considerable move in ‘95 and again in ‘03. There were other “minor” moves in between. The ‘95 move seems to roughly coincide with the departure in the anomaly graph (by eyeball).
I haven’t checked other stations.
the Grit // April 18, 2007 at 10:56 pm |
Hi tamino,
You still prove my point. If you can’t believe the data you presented is accurate, even in an area as highly sampled as Shelby County, why should anyone believe data from a much larger area with only one point of measurement is reliable?
the Grit
[Response: No, you prove my point. Despite showing the incredibly strong correlation between these stations separated by more than the size of the county, and showing that all the station data are consistent with the best-estimate warming rate of 3 deg.C/century for this region, you still disbelieve that the density of observing stations is sufficient.
I think you're attached to the idea that we can't estimate the warming rate of the planet, and no amount of data or correlations will convince you. So be it.]
nanny_govt_sucks // April 20, 2007 at 7:27 pm |
Why do you start in 1950? Why not go back to the beginning of these temperature records and see if we can see a whole century’s worth of data rather than half of it, estimating the trend for the rest?
[Response: I started in 1950 because the Memphis time series doesn't start until 1940, and the Moscow series doesn't start until 1948. So, 1950 was the nearest "round number" which would include data from all five stations.
Perhaps I'll update the post to show the entire series for all stations.]
nanny_govt_sucks // April 20, 2007 at 7:40 pm |
I think there is some obvious third-variable influence at work in this top smoothed wavelet graphic. Exclue Hernando and Moscow for the moment because of seeming problems with data, and just look at Memphis vs Covington/Holly Springs. I see 2 things – 1.: Memphis warms relative to the other two in the 70’s and 80’s then cools relative to them in the 90’s. Does this match with economic/population/asphalt expansion during the same times? I think it likely. 2. The fluctuations in the Memphis temps during the 70’s 80’s and 90’s appear to be less severe than for Covington/Holly Springs. Is a third variable damping the temperature swings for the big city?
Looking at the temp differences between Memphis and Covington/Holly Springs, it appears to be about 1 degree in 1950, 2 degrees in 1980, and 1 degree again in 2000. So, my estimation is that there is at least 2 – 1 = 1 degree C of third-variable influence in the temp trends. It is always possible that this influence could be greater because we are only looking at Mempis relative to the other two. The third variable could be affecting Covington/Holly Springs as well!
nanny_govt_sucks // April 20, 2007 at 8:06 pm |
With regard to siting, what effect, specifically do you refer to?
[Response: Siting can refer to changes of location, or to changes in micro-location (moving the instrument from ground to rooftop) or local environment. A detailed study of the impact of non-climate factors on temperature records can be found in Peterson 2003, Journal of Climate, vol. 16, pg. 2941.]
nanny_govt_sucks // April 22, 2007 at 3:33 am |
But didn’t Peterson 2003 basically reject any temperature issues related to siting? Yet you brought up siting as a possible problem with the data from Moscow/Hernando. Was this a mistake, or are you referring to something else?