No More Cold Turkey?

I’ve been intrigued by climate measurements in Turkey since becoming aware of the large number of stations there (more than any other country in Europe, even in relation to its size) while evaluating the ‘Station Dropout Problem‘. E.M. Smith (Chiefio) posted an analysis of the temperature record in Turkey a few days ago and clearly found a step change the record, so I had to go back and do some digging myself. Here’s one of E.M. Smith’s graphs:

Figure 1. Turkey – Cumulative Change of Temperature (dT) and Change per Year (dT/yr) (graph: E.M. Smith)
For anyone who has not been following Chiefio, I should explain that his dT method is a variation of ‘First Differences’ as a means of examining temperature data anomalies independent of actual temperature.

“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.”

The dT graph shows a flat trend up to 1901 then a cooling trend up to 1951.  After that, flat for a few years, then a small step change upwards in 1961, then flat again until a major step up in 1995. So what happened in 1995 to create a sudden jump like that? What information is there about climate in Turkey?

One study [1] examined the 63-year period starting from 1930 to 1993 and used temperature records from 85 climate stations. It concluded that annual mean temperatures were generally dominated by a cooling tendency in Turkey  and that 1933 and 1992 were the coldest years.  The paper also notes that only Eastern Anatolia appeared to show similar behaviour to the global warming trends, except in the last 5 years.  It seems that 1992/93 were particularly cold years in Turkey, attributed to the effect of the eruption of Mt Pinatubo [2].

Climate in Turkey has been closely linked to the NAO:

“During the positive phases of the North Atlantic Oscillation indices (NAOIs), northeasterly circulation increased, and thus spatially coherent and significant cold signals dominate over the majority of Turkey.” [3]

The NAO index is highly negative in 1990 (1.23) and 1992 (1.11), but transiently flips to positive in 1994-1996 (-0.61; -1.01; -0.18). It couldn’t be this simple, besides the record for Turkey stays warm after 1995/6. What else? Chiefio noted changes of modification flags in the station data files – anything to do with lost stations? Well – here’s the station count overlaid on his graph: 

Figure 2. Figure 1 overlaid with interpretation of trends and the station count for Turkey in the GHCN V2.mean datafile. 
Oooh that is interesting.  Could we possibly have a ‘March of the Thermometers’ in Turkey?  I’ve also had another way of looking at it.  I’ve been looking at trend data for individual temperature records: a linear regression through the data for each station over its reporting lifetime.  When I plotted counts of the station data against the year when the stations were either added to or dropped from the record (Figure 3) the dominance of stations with a cooling trend is apparent. 

Figure 3. Counts of stations plotted again the year when the station [a] was added or [b] ceased
providing data in the GHCN v2.mean file.
There are very few stations at the beginning of the record, and the first significant additions are in 1951, then in 1961 50 new stations were added (Fig 3a), with further additions in the following years. In Fig 3b we see a massive loss of both stations with a warming trend (51) and even more of those with a cooling trend (108). A predominance of stations with a warming trend remains. Could this be responsible for the effect seen in the dT graph?

Having done a bit of searching on line, it seems that there was a massive closure of climate stations in Turkey during 1990 – more than 600 stations were closed during the decade [4]. OK that fits with what is in the record.  Here’s what the loss of stations looks like on a map:

Figure 4. Maps showing station temperature trends for (top) all stations active during 1880 to 2010 and (bottom) for stations active after 1990*. ( *Note 25 of the 39 current stations are shown; the remainder have insufficient data to calculate a trend for this period (less than 10 years without missing months). 
Note that the station temperature trends indicated by the colour of the dots in Fig. 4 show a current rapidly warming Turkey. A lot of those dots in the lower map mark trends of more than 5 deg.C per century – this is just the rate of warming over the period 1990-2010. By contrast, over longer periods (upper map) many areas are cooling. I find this a bit puzzling – so what about looking at some individual station data?

Figure 5. Graphs from four current stations in Turkey showing data before and after
adjustment by GIStemp. (data from
Well, these four plots are fairly typical of currently active stations in Turkey. Istanbul/Gozt is the longest record, but 62 years are dropped by homogenisation adjustment because GIStemp requires to match it with a rural station to correct for Urban Heat Island effects and it is assumed the only rural station in range does not begin reporting until 1943. The shame of this method is that temperatures in Istanbul in the 1880s are similar to those in the most recent decade. The other three stations also show little evidence of warming prior to adjustment; Antalya and Diyarbakir even have a cooling trend.  Fourteen of the 39 stations active in 2009/10 have an overall cooling trend; 25 are warming. Only one is ‘Rural’ (it happens to have an overall cooling trend);  22 of the stations are ‘Urban’ (4 cooling; 18 warming);  16 are ‘Peri-Urban’ (9 cooling; 7 warming). Sixteen stations are truncated as those in Fig. 5.; fifteen are warmed during adjustment by GIStemp; eighteen are cooled by adjustment.

In checking database records with the GISS site, I’ve found quite a few “Lazarus Thermometers” that stopped reporting into the GHCN file in 1990, but have recently started again. One of these, Yozgat, is shown below (Fig. 6). Note that gap – Yozgat has still not warmed – could the absence of it and similar records from anomalies since 1990 have resulted in a ‘lift’ of the anomaly value by warmer trending thermometers?  Strangely this and four very similar others current for 2010 are now no longer available from the GISS Station Data page under “after homogeneity adjustment” and are only present as “after combining sources at same location”.  Unless this is in error this would suggest they are also no longer in the “Adjusted data” file and therefore do not contribute to GIStemp anomaly calculation and gridding. Why?

Figure 6. Graph for Yozgat, one of five similar stations recently resurrected.
(data from

E.M. Smith’s dT method, using the unadjusted GHCN input to GIStemp, shows cooling in Turkey, a trend that is confirmed by publications.  However, the dT graph shows step changes that indicate a possible effect of variability in the thermometer record – first the addition and then loss of stations.
Looking at the overall pattern of stations and the data within them, the stations that remain seem badly served by GIStemp adjustments in relation to the raw data available.
These elements together seem to contribute to recent warming of anomaly values in Turkey.


  1. Türkeş, M, Sümer, U.M. and  Kiliç S. (1995) Variations and trends in annual mean air temperatures in Turkey with respect to climatic variability. International Journal of Climatology 15(5): 557 – 569
  2. Tayanç, M., Im, U., Dogruel, M. and Karaca, M. (2009) Climate change in Turkey for the last half century. Climatic Change 94:483–502
  3. Türkeş, M. and Erlat. E. (2009) Winter mean temperature variability in Turkey associated with the North Atlantic Oscillation. Meteorology and Atmospheric Physics 105 (3-4): 211-225
  4. Türkeş, M. (2009) An Assessment on the Precipitation Data Recovery in Turkey.  Presentation at Med-CLIVAR Workshop, Barcelona (link)
This entry was posted in Uncategorized. Bookmark the permalink.

6 Responses to No More Cold Turkey?

  1. Anonymous says:

    Well, If you don’t bake it long and hard enough, it will never turn out the way you want it to…

    Nice work, Sir.


  2. chiefio says:

    Bravo! Very well done. All that digging and detail and citations that E.M.Smith is too lazy to do. Oh, Wait, I am E.M.Smith 😉

    BTW, the dT method was developed to do 2 things. One was to give some kind of baseline measure of reality via an anomaly method, the other was to detect “the thing in the data to which GIStemp was reacting with their anomaly method”. IMHO, it has done very well at the latter, not so well at the former. That is, it shows a dramatic step function when they play with the instruments and only records a valid real trend during the periods of relative instrument stability. Though, curiously, I find that the most “valuable” combination… It lets you see both the dropping real trend in Turkey AND the spike when they diddle the thermometers….

    It does report overall changes of temperature over time (dT from end to end) that are roughly in line with the GISS anomaly maps, so I think that the same thing that causes it to find the instrument change spike also impacts GIStemp anomaly calculations and to about the same degree and direction (though they do a lot of ‘smearing around the data’ via interpolations, in-fill, etc. and they do anomalies “Basket of thermometers A in the baseline” compared to a “Different Basket of thermometers B in the present” that make the comparisons inexact.

    In the end, I think this article does a very good job of showing why “Station Change” and in particular “Station Dropouts Matter”.

  3. VJones says:

    Thanks Guys! It was a bit rushed in the end as I had trouble with the graphics. Also the conclusions are rather weak – but perhaps better that way as I’d rather folks made up their own mind.

    @RR – you’ve give me the idea for a recipe – Global Warming Layer Cake, although I think it only looks hard baked. It is actually a delicately balanced confection that could actually collapse if overcooked!

  4. tonyb says:

    Hey Verity

    I have the climate references for the Byzantine Empire from 380 AD to the sacking of Constatinople in 1452. How about adding those to your Turkish calculations 🙂


  5. Sinan says:

    This is excellent stuff. Please take a look at my new animation of patterns of adjustments in the GHCN-v2 data set while paying close attention to Turkey post-1961. It looks like there are 4 “hot” regions for positive adjustments between 1960. By the mid-1980s, only the thermometers in the U.S. need adjustment. I have no idea what is going on after 1990.

  6. Sinan says:

    See Here is looking at the USA for animation of continental U.S. adjustments between 1970—2009.

Comments are closed.