Roast Turkey Since 1995
This posting is a little bit different for me. First off, I’ve made a batch of graphs. That I’ve done so at all is a bit “new” for me. But beyond that, I’ve made the graphs before even starting the article. It was the data that got me looking, but the graphs that made the case. And what is that case?
That it’s a bad idea to play with the instrumentation.
First, the motivation. In looking at the GISS anomaly maps, Turkey is always roasting. There was something about that that just seemed wrong. It ought to have been a more random event; who was Hot Hot Hot and who was not.
So I took a look at the dT/dt report and and noticed that there was a clear “kink” in the data. It was running along flat, then about 1991 – 1995 it popped up and then continued running fairly flat with some rise. Odd, that ‘step function’. CO2 does not do a step function. Airports don’t do a step function. UHI does not do a step function…
( For folks new to this: dT/yr is the “average of the changes of temperature, month now vs the same month that last had valid data, for each year”. An anomaly process similar to First Differences. Then dT is the running total of those changes, or the total change, the “Delta Temperature” to date.)
Turkey - Cumulative Change of Temperature and Change per Year
You will probably want to click on the graphs to get a bigger version to look at.
Notice that the dT line is the “cumulative” dT/yr. Notice also that right around 1990 the dT/yr goes very small in range. The volatility just leaves it. So we see that dT is running along at a lower value, and dT/yr is dead flat for a couple of years, THEN about 1994 dT/yr pops up and dT starts a run for the moon. So what happened between 1990 and 1994?
The Great Dying Of Thermometers took out a lot of thermometers. Further, a bunch of old “Duplicate Number” flags end (that I’ve been calling “modification flags”, but I’m swapping over to the NCDC name) and some new ones begin. Turkey had a fairly large number of thermometers, but that number plunges. Still, it’s a bit large for really seeing exactly what happens thermometer by thermometer. We’ll see a clearer example down below in Mauritius. But for now, we need to slice a little Turkey…
Looking at the temperature data directly, there were a lot of overlapping adds and drops. Duplicate Number flag 0 or 1 that carried forward and overlapped with Duplicate Number flag 2 or even flag 3. So their is a bit of ‘feathering’ or ‘blending’ of the dropping, adding, and changing. I believe that is why when you look at the dT/yr right about 1989 – 1993 it goes to very small volatility, even though a large number of thermometers leave the record (or perhaps, because of it). Eventually we are left with a lot of new Duplicate Number flag thermometers and very few of the older consistent record.
What is a “Duplicate Number” flag? It is an indication that SOMETHING about that data record is different and that it needs different handling than the prior record. It could be a new thermometer at that location, or it could be a new Time Of Observation (changed procedures) or it might be simply that the thermometers and the TOBS are the same, but some “post processing” is different. A different QA method tossing out “outliers” or a different “TOBS adjustment” or just about anything. Basically, it means “Something that matters changed”. I’d say so…
Now look again at the dT/yr line. Notice anything? Run your eye along the peaks at about the 1.3 C level just above the 1 C line. Notice the consistent peaks? Now run your eye along the -1 C line. Notice the consistent peaks starting before 1970? Now run your eye along at about -0.5 C. Notice how much more area is in the down spikes before 1990 when compared with after? Something started to slightly clip the lower excursions in the ’80s, but that 1990 change substantially eliminates them and strongly dampens or blunts any more than about -0.75 C and reduces the number significantly.
Don’t know what it was, but it happens at the same time the Duplicate Number flags change. And We’ve seen this in Canada (as the Smith Effect at Fort Smith) and I’ve seen it in dozens of other cases around the planet. The “warmth” is from clipping of the cold spikes, and it has onset with the Duplicate Number flag change.
The Ugliest Graph You Will Ever Love
In this graph, instead of the average dT/year we see the individual months dT for that year. So there are 12 lines in this “Spaghetti graph” for dT(per month)/yr. Maybe we ought to call it a ‘hair graph’ ;-) and that makes it look useless at this compressed scale. But click on it and enlarge it. Now the monthly extremes are much more visible. The “spikes” of monthly temperature changes are “spikier”. The “clipping” is fairly clearly seen. The start of the newer Duplicate Number flags dampens the really deep downward spikes (as they average with the older Duplicate Number flags) but in 1990 when the old Duplicate Number flags are dropped and only the new carry forward, the “cold spikes” look to me to be lifted to where what in the past was a -6 is now the few -4 about 2001 and the prior -4 spikes are now running about -2.5 C.
There is what looks like a little muting of the “top spikes” but not nearly so much.
Turkey - Monthly Temperature Changes and Cumulative Change
Again, why is completely unknown. One would need to look up what those Duplicate Number flags indicate changed at that location to figure out ‘what changed’. But that it changed and the fact that it changed in a step function is pretty clear…
So, Are The Trends of the Two Segments Different?
You Bet!
Turkey - Segments by Duplicate Number have divergent Trend Lines
A Cleaner Case – Mauritius
Is there a simpler place? Somewhere with fewer thermometers and a clearer point in time when things ‘cut over’? Yes, several. This is just one of a half dozen I found in looking at about a dozen randomly chosen places. The scale here is much shallower than for Turkey, since these are Islands in a warm ocean. But even here we can see a “muting’ of the down spikes of dT/yr such that they don’t reach the -0.5 C line and rarely get through the -0.25 C line after the change of Duplicate Number flags.
Mauritius - Cumulative Change of Temperatures and Change per Year
If we look at the two segments, before and after the 1990 era Duplicate Number flag change, do we find a difference? Oh yeah. Again we do see the top peaks slightly muted, but the bottom going peaks are much more muted and the effect of that “peak clipping” of the dT/yr, and muting the depth and quantity of the bottoms; is a dramatic change in the dT line.
Mauritius - Duplicate Number flag 3 Has A Warming Trend
I’ll spare you the “Hair Graph” of the monthly data… Unless requested…
And it is all done with ‘anomalies’ so it must be right ;-)
(For folks unaware of it, perhaps a long time after this is written, there has been a bit of a “tiff” about the fact that I don’t use anomalies for all my investigations. It’s been asserted that if you use anomalies things come out right and if you don’t they come out wrong. Well, I’m from a school of thought that says you use ALL the tools you have and see if they all agree. It’s more of a “forensics” mind set. I’ve done computer forensics, and one of the things you learn is that it is much much harder to ‘cook the books’ so they come out right under ALL available tools… So always “Come to your opponent out of the Sun.” and “Take nothing that he offers to you.” Sun Tzu. And the best way to assure you have found ‘where the sun is’ in a forensics exam is to come at the data from ALL sides. And never do things in the usual way or the way the other side suggests they ought to be done; if possible, approach from exactly the other direction…)
Conclusion
Now, I have no idea what Duplicate Number flag 3 means. It could be the change from Liquid In Glass to Electronic, or it could be a different QA procedure, or a move from “Stevenson Screen by the Palm Grove” to “near the hangar”, or who knows what. But what I can say is that:
1) It’s a Giant “Dig Here”.
2) It is not CO2.
3) It is directly correlated with the bulk of the change of trend to warming in Mauritius.
4) That warming comes, as have most of the cases I’ve seen, as a Step Function or knee, though there can be some ‘blending’ if different records overlap.
5) Dropping thermometers matters (directly, or as here, indirectly via unmasking Duplicate Number flag 3).
6) The specifics of the Instruments, Modifications, and the Siting Matters. A Lot.
7) It’s not the “kept” vs “tossed” that matters it’s the “kept” vs “replacement” at the mod flag level.
What the Input Data Looks Like – Sample Duplicate Number Flags
OK, I can’t make a new posting full of graphs without having at least some tables… don’t want to have withdrawal shock ;-) so here are some small tables showing some of the “Duplicate Number flag” changes in Mauritius. These are mostly for documentation of what to look for in the data for folks who want to look at the data themselves; those not interested in the input data will want to skip this part.
The first 3 digits are the Country Code “129” for Mauritius. Then 5 + 3 for station and sub-station ID “61986” + “000”. Then the Duplicate Number flag. “0” for this first thermometer record at that place. Then the year (1986 in this first line). Then 12 monthly temps in 1/10 C so that first one “273” is really 27.3 C.
... 1296198600001986 273 280 274 269 256 240 235 233-9999 250 258 273 1296198600001987 279 278 289 284 267 249 245 234 238 247 257 265 1296198600001988 279 282 287 274 258 241 235 232 235 248 257 273 1296198600001989 277 280 281 270 254 239 229 226 234 243 251 269 1296198600001990 271 275 273 271 257 243 233 231 242 246 255 272 1296198600001991 277 280 283 273 261 246 234-9999 239-9999 258-9999
We see it ends in 1991 with the last 6 months having 3 missing data flags of -9999.
Same Station ID, but we get a “1” Duplicate Number flag (that actually starts in 1954).
... 1296198600011984 273 278 277 266 247 234 222 225 235 241 255 265 1296198600011985 270 272 276 268 255 243 233 230 233-9999 251 271 1296198600011986 273 280 274 269 256 240 235 233 237 250 258 273 1296198600011987 279 278 289 284 267 249 245 234 238 247 257 265 1296198600011988 279 282 287 274 258 241 235 232 235 248 257 273 1296198600011989 277 280 281 270 254 239 229 226 234 243 251 269 1296198600011990 271 275 273 271 257 243-9999 231 242 246 255 272
And ends in 1990. We also have a very short lived Duplicate Number flag 2 that looks like someone ran a 10 year experiment. This is the whole thing:
1296198600021971 275 270 274 267 255 241 232 224 225 237 259 271 1296198600021972 276 269 277 271 260 255 238 239 239 245 265 278 1296198600021973 282 282 284 277 267 247 239 230 237 244 254 263 1296198600021974 270 271 272 265 252 240 230 229 227 239 251 272 1296198600021975 279 280 276 275 266 247 239 229 236 239 261 266 1296198600021976 273 277 277 273 266 252 239 239 235 245 258 273 1296198600021977 279 281 285 278 266 252 242 239 237 246 258 269 1296198600021978 276 280 269 272 261 245 234 234 238 246 257 267 1296198600021979 279 275 276 274 265 247 238 236 238 248 260 279 1296198600021980 276 275 277 277 258 249 240 232 234 247 258 266
Then we get to the “Duplicate Number flag 3” that carries forward as the other die. It starts in 1987 for a bit of blending and “feathering” in with the other records, then, as they die off in 1990+, it is what gives us the temperatures today.
1296198600031987 279 278 289 284 267 249 245 234 238 247 257 265 1296198600031988 279 282 287 274 258 241 235 232 235 248 257 273 1296198600031989 277 280 281 270 254 239 229 226 234 243 251 269 1296198600031990 271 275 273 271 257 243 233 231 242 246 255 272 1296198600031991 277 280 283 273 261 246 234-9999 239-9999 258 284 1296198600031992 284 280-9999 267 254 242 232 229 234 243 255 255 1296198600031993 276 286 280 275 267 248 232 231 232 242 252 266 1296198600031994 277 281-9999 277 262 249 241 235 236 248 252 268 1296198600031995 276 284 279 275 257 248 239 234 239-9999 251-9999 1296198600031996 278 280 283 269 255 245 231 231 232 237 253 277 1296198600031997 276 281 280 274 264 246 236 238 241 247 272 283 1296198600031998 298 291 290 278 265 253 240 236 239 245 254 265 1296198600031999 271 278 280 275 265 250 233 233 236 244 257 267 1296198600032000 275 278 277 270 264 242 235-9999 235 249 262 274 1296198600032001 284 290 285 281 268 243 240 243 247 252 262 277 1296198600032002 284 285 287 276 264 247 244 235 242 252 266 282 1296198600032003 293 284 286 285 273 250 242-9999-9999 254 265 276 1296198600032004 279 290 287 279 260 245 242 240 246 252 265 277 1296198600032005 291 273 288 279 264 251 240 233 237 245 259 273 1296198600032006 282 280 281 277 265 253 241 238 241 248 265 276 1296198600032007 293 282 276 273 265 247 241 239 241 244 260 271 1296198600032008 279 274 273 271 259 243 233 240 246 251 267 277 1296198600032009 287 285 289 284 273 257 241 238 242 255 270-9999
Musings from the Chiefio 
Gimme a Graph with Hair, (Yeah!)
Shoulder Length or Longer, (Yeah!)
Here, Sister, There, Brother,
neh, neh, neh neh neh
(such memorable words…)
BTW, probably there was a XC rendition of Hot Hot Hot among that 16Gb.
Fine work indeed.
Take the rest of the day off…
RR
Pingback: Let's Talk Turkey « Musings from the Chiefio | Turkey Live
@Ruhroh:
Well, you DO recognize the Open Office graphs and know what your contribution to the effort has wrought ;-)
Hadn’t thought of the “Gimme a Graph with Hair” angle, but I really like it…
And right now I’m watching an old B&W movie and sipping on a nice Rosé as I contemplate wether to post the Germany Graph showing no warming (along with a couple of others) or more of The Smith Effect graphs with mod flag changes, or … So I guess that counts as “off” ;-)
FWIW, I’m planning to go country by country and sort them into “Not Warming – You get a Carbon Free Pass” and “Cooling – Should we pay you?” and “Warming with The Smith Effect – BOHICA” and if any: “Warming but don’t know why”.
At a couple of countries per posting that ought to hold me for about 70 postings ;-)
Pingback: Let's Talk Turkey « Musings from the Chiefio | Headlines Today
Wow, my plum tree thought it was spring already, but my car windshield had a different message this morning; 1/8 inch of ice. Not the white frosting kind, more like the thin glass mostly-transparent kind.
Luckily the hose is also full of very cold water so I was able to effect local microclimate change without the deleterious effects of excessive dT/dt on the underlying transparent surface.
My hands are still in major vaso-constriction; like typing with wooden clothespins instead of fingers…No touchtyping today…
Your use of that mochine may be the most valuable thing it ever did.
Well, maybe the second most valuable.
The best thing might be the spouse points I got for actually moving 1 cubic foot of formerly precious electronic functionality out of the house…
Hair, Hair, I say!
RR
The Global Gridded population dataset for Turkey seems to suffer from the same problems as Iraq Syria and Saudi Arabia.
I.E. Population records are kept by administrative districts similar to a US County.
This results in population densities being homogenized across the entire county. Which end up distorting population densities.
If I check the population of Ankara,Turkey it’s 3,700,000.
If use use the Gridded Population Data tool
http://sedac.ciesin.columbia.edu/gpw/wps.jsp
The population is something like 400,000.
The population density is homogenized at we would call ‘county’ level which results in a gross understatement as to the level of urbanization if I depend on the ‘Gridded Population Density’ to determine urbanization.
If I use King County Washington as an example.
If I use the population records kept at the county level I get 2 million people in 2,000 sq miles of land. Giving a population density of roughly 1,000/sq mile.
If I look at the Seattle City records I get
the city of Seattle has 600,000 people crammed into 84 square miles. A population density of 7,000/sq mile.
Where is the thermometer?
In the city of course.
Sorry to be picky EM but your…
“Same Station ID, but we get a “1″ mod flag (that actually starts in 1954).” – shouldn’t that be 1984?
Nevertheless, and excellent article.
REPLY: [ Thanks for the compliment. The data actually start in 1954. I tried to indicate the elision with the … at the top and the comment, but I guess it was unclear. I’d rather have just posted the whole table but folks toss rocks at me for actually posting the whole table where you don’t have to figure out what was going on… So I deleted 30 years (lines) to make this posting ‘less table – more graph’. But then you have what you experienced. What you see is not the whole story. Maybe if I had the 1954 record, then the …, then the 1984 record? Somehow just saying “Here is THE data” and putting it all in is still “calling to me”, rocks or not ;-) -E.M.Smith ]
EM, why do your dT curves for Turkey start at -3C?
REPLY: [ It actually starts at the other margin and moves backward in time to the early record at -3 C. The original version of this graph had “Now” on the right and I got rocks tossed at me for “having time backwards”, so I swapped it around to match GISS and some other folks graphs with “Now” on the right (though I’ve seen lots of other graphs with “Now” on the left – mostly Ice Age and very long duration graphs). So now you have the data “starting” at -3 C because that’s really the end… I also just noticed that in the swapping the formulas for the slopes of trend lines did not move with the graph… Looks like I’ll need to make a new version with the formula right next to the (now on the other side) trend line… Maybe I’ll just put them in going both ways and folks can choose which way then want time to run. No, then folks will just toss rocks at me for making it too messy and complicated. 8-} I suspect there is no way to avoid taking rocks. I guess you really only get to choose which ones you get ;-)
-E.M.Smith ]
Very good work – Excellent post.
I’m surprised someone from NCDC or GISS hasn’t posted a reply here saying: ” You dumb ****, all you had to do was go to “www…”, and check the meta data. As you can see, the A/C was added in 1990, so the computer model adjusted the data”.
REPLY: [ Well, the trolls and graffiti artists showed up in a herd after I was on the news, but when I basically said “No throwing rocks at your host” they largely went away. There are still a couple some times, though. Per NCDC / GISS folks, I don’t see them participating under their own name outside their own controlled spaces, so I suspect they use pseudonyms. But in any case, one interesting pattern I’ve noticed. it’s the “negative space” pattern. When I’ve hit on something pretty good, I get complete silence. The “hit count” goes way high… but not a peep. A few supporters will post positive comments. Maybe one or two “clarification” questions. Then the hit count ramps up. Kind of like it’s doing right now ;-) -E.M.Smith ]
EM,
Very interesting post and analysis. I too have been looking at Turkey due to the very many stations for the size of the country. I might do a post on this myself incorporating some of what you have found.
!992 seems to have been very cold in Turkey: http://www3.interscience.wiley.com/journal/114078036/abstract. Perhaps if you can look at the 1990-1995 period more closely this would show up. Looking at many of the individual station records 1994 seems to have been very warm. Some stations have a 3 degC jump between these years.
Great graphs BTW, but small nitpick – the dates on your X axes are very hard to decipher.
@vjones: Feel free to use any ideas presented here as you see fit. I’m just turning up ideas for other folks to run with ;-)
BTW, I haven’t found a way (yet) to customize that axis labels as I’d like in Open Office. However, the graphs are ‘way big’ if you click on them and then the dates become much more readable. On the small scale here (especially on my 12 inch screen Mac) the dates are just a blur unless I dig out some reading glasses, and even then it’s just barely readable, maybe. So “puff it up” and it’s better.
To change the x axis graphs, open the edit feature on the graph by right clicking on it and selecting edit. From there right click on the x-axis and select format axis if you have upgraded to ver 3.2 (I forget what they called it in 3.1 but it’s the top option).
Now click on the Label tag and this brings up formatting screen for the layout of the axis. My preferred choices is to select “Tile” or set the degree option to 90°, this spaces the years out better.
Tile will leave the numbers horizontal but spaced out, setting to 90° turns them vertical which I like since then you can see which of the div bars it corresponds to easily.
@Bobalab: “Thanks, I needed that!” ;-)
@vjones: That link is rather fascinating. I particularly found this paragraph compelling:
Among the geographical regions, only Eastern Anatolia appears to show similar behaviour to the global warming trends, except in the last 5 years. All the coastal regions, however, are characterized by cooling trends in the last two decades. Considering the results of the statistical tests applied to the 71 individual stations data, it could be concluded that annual mean temperatures are generally dominated by a cooling tendency in Turkey. The coldest years of the temperature records of the majority of the stations were 1933 and 1992, respectively.
So when they looked at ALL the stations in Turkey, they found a cooling trend (that I would speculate is roughly consistent with the cooling trend I identified in the first part of the data (the blue segment) before the “hockey stick” knee and the Mod Flag change segment (the red segment).
OK, we have a peer reviewed document from someone who looked at a boatload of thermometers all over Turkey:
The study covers a 63-year period starting from 1930 and uses temperature records from 85 climate stations.
and uses a variety of techniques to determine the true trends by region inside Turkey:
First, spatial distributions of the annual mean temperatures and coefficients of variation are studied in order to show normal conditions of the long-term annual mean temperatures. Then variations and trends observed in the annual mean temperatures are investigated using temperature data from 71 climate stations and regional mean series. Various non-parametric tests are used to detect abrupt changes and trends in the long-term mean temperatures of both geographical regions within Turkey and individual stations.
In what sure looks to me like a fairly rigorous and in depth manner.
And they find it’s cooling and not making a Hockey Stick warmer….
And folks wonder why I think dropping stations matter and why I think GHCN “has issues” with station changes introducing bias…
BTW, the present re-imagining by GISS has Turkey 2-4 C Hot Hot Hot…
Here’s a thought I haven’t analyzed very much.
Does the ‘Hair’ graph of a combined data set give a visual indication that non-homogeneous data sets are being combined?
I did remember more of that fine anthem of the 70’s;
Hair baby,
THair mama,
everywHair daddy daddy,
Hair, Hair, Hair, Hair, Hair, Hair, Hair,
Head it,
Tail it, long as you will plot it,
That Haaaaaaaaaaaair…
OK, it’s early here…
RR
@ruhroh
I think we’re thinking the same thing…
It looks to me like 2 thermometers are being averaged where one has ‘higher lows’ than the other one. The result is a dampening of the down excursions… Then you take away the cooler older thermometer and “voila” it’s warmer… But I’ve not had the time to work through the details to test the theory.
The other explanation that might make sense is that you have, say ,a LIG and Electronic thermometer being run side by side for a couple of years to validate the cutover. They end up more closely monitored than otherwise; so the guy reporting the LIG temp sees the electronic one says 20.51 where his glass one says, by his read, 20.4x and just decides that instead of doing the round down that he would have done to 20.4 he wil run with the 20.5 that ‘matches’. With the added risk that during the ‘both there’ they calibrate the elctronic to match the Liquid in Glass so the ‘records match’ then the next year at “calibration time” it’s done to a reference standard and now you take, oh, a 1/2 C jump as you ‘calibrate corrctly’… but the ‘splice’ is now broken…
FWIW, I’ve now done reports on about 3/4 of all countries in the world. Substantialy all of the Amercas, Pacific Basin, Asia, Europe. Africa is about 1/2 done. I’ve seen a Very Clear and a Very Persistent pattern. Staions fall into one of a very few number of patterns:
1) NILL data. Far more than I’d like. The country is in the country file, but has NO data. Libya, for example. These will be filled in from “nearby”
2) Dead Flat of Falling trend, but truncates between 1980 and 1990 or so. These will then be filed in from somewhere else.
3) Some are flat or falling and persist to today.
4) Many have a “hockey stick” (often with onset about 1990 with a Mod Flag change, some in 1980, some about 2006-7, some with both a 1990 and a 2006 ) that is characterized by a substantially FLAT or slightly rising lead in, then a STEP FUNCTION to warming trend on the Mod Flag change. Sometimes with a bit of ‘blending’ or ‘feathering’ as the stations overlap from about 1987 to 1992. These will be used to do most of the inflilling as other stations exit.
And not much else at all.
Not a hint of a peep of the pattern CO2 would make with steady rising or “rising steady with an imposed ‘rolling’ such as from the PDO flip”.
What we have here is a Mod Flag and instrument change driven step function, by country, being masked with all the in-fill, homogenizing, UHI “adjusteing” etc.
E.M.,
I’ve taken your Turkey dT graph and looked at the station volatility:
http://diggingintheclay.blogspot.com/2010/03/no-more-cold-turkey.html
Very interesting. In fact the whole ‘thermometer’ thing in Turkey is a bit of a travesty as far as really following the climate.
REPLY: [ Just finished reading it. Marvelously done. Just marvelous. Folks, read that article. I found the “what?”, but vjones has found the “why” for Turkey, IMHO. At a minimum, part of the why and with specifics for several stations. -E.M.Smith ]
Pingback: CRU, NCDC, and GISS not Scientific « Musings from the Chiefio