RUTI: Coastal temperature stations
Fig1. World wide comparison coastal/Island temperature trends vs. non-coastal trends from RUTI (Rural Unadjusted Temperature Index)
In the following it is shown, that temperature stations from coasts / islands generally have significantly different temperature trends than near by non-coastal temperature stations.
The focus of this writing is coasts world wide where land (from continents or Islands) meets the three large oceans (The Pacific, Atlantic and the Indian ocean) directly. For this analysis, 35 areas where land meets these oceans where used as studycases. They represent nearly all possible such areas where data is available.
Further more it is shown, that often coastal temperature stations have more heat trend before 1950 than the near by non-coastal stations.
Some consequences hereof:
- Coastal temperature stations are unreliable as indicators for non-coastal areas and vice versa. Often, the warmer trended coast/island area is just a fraction of the total land area and resembles the sea surface trend rather than the land based trend.
- Coastal stations ought not be used for long range smoothing over land areas (or vice versa).
- Adjustments where a non-coastal station are changed to look more like a coastal station (or vice versa) should be avoided.
Data used in this writing - like all RUTI writings - are from unadjusted GHCN. When analysing unadjusted GHCN data in general (for the RUTI project), I repeatedly experienced that temperature stations placed on islands or on the coast or very near the coast had very similar trends, but this trend was significantly different from the trends from temperature stations in the very same area, but just a little further from the coast. Sometimes just 10-20 km makes all the difference.
Since temperature trends obtained from Islands or coast stations are generally very similar, these coastal/island stations are likely to be dominated by the Marine air temperature trend.
Raw average of the 35 study cases using 1961-1990 as baseline, dots are one year and lines are 5 year averages:
Fig2: Worldwide raw average of coastal vs. near by non-coastal temperature trends from the 35 study cases representing nearly all places on Earth where we can directly compare large-ocean trends and non-coastal temperature trends (from the very same data source).
! ! The “beauty” of comparing coastal vs. non-coastal stations from the very same source (GHCN-Unadjusted) is, that whatever adjustment you might find necessary for temperature data stations in general there is no adjustment type that is dependant on a coastal or non-coastal location. Therefore differences approximately as shown in fig2. coastal vs. non-coastal both from Unadjusted GHCN tells a story that is more likely to be somewhat true.
Fig2a. Comparison with Raobcore (30S-30N) tlt. Representing half the globe, and around 2 thirds of the area where it was possible to make the 35 study cases. Both land (non-coastal) and ocean (coastal) trends from this writing appears to produce results rather similar to Raobcore data 1958-2008.
Fig3: Comparison between “GISS Ocean temperatures” and the marine air trend from coastal stations, Unadjusted GHCN. (GISS data taken from the graph that disappeared )
Base period both datasets: 1961-1990. The match between “GISS Ocean temperatures” and the Coastal air temperature trend is rather good all the way from 1880. In general, the different versions of ocean temperatures from Hadcrut, NOAA are rather similar, even between newer SST and Marine air temperature datasets we have similarities (For example HADISST vs. MOHMAT4.3), so:
Unadjusted GHCN temperatures – taken from Marine locations like Coasts and Islands - fairly matches normally accepted ocean temperature trends (SST or MAT).
Could we have expected this match? To some degree yes, because 35 ocean-chunks spread out rather randomly over the world is likely to give a fair match to global MAT and SST.
So for the coastal and island Unadjusted GHCN stations, it seems that data to some degree are being verified.
The shape of the non-coastal temperature trends from the 35 large chunks of land spread around the globe, reminds me of several proxies for temperatures, for example:
Fig4. CRU´s total tree ring density dataset (to be used as temperature proxy) and the non-coastal GHCN Unadjusted shows fine agreement, not much “divergence-problem” here.
I think we can conclude that trees grow on land – or is that taking it to far?
(CRU´s other parameter for tree proxies, Ring width, has a fair mach too).
Fig 4a. - A handful more trends related to temperature.
Fig4b. The entire area of China shows a fair match to the non-coastal trends. The above data for China - from Wang et al. 2001, Twentieth-century climatic warming in China in the context of the Holocene - are based on orignal Chinese temperature data and on top of this Chinese temperature proxy data.
Fig4c. Comparison Non-coastal temperature trends vs. USA raw temperature trend (Source). Again, we see a fair match.
Fig5. Here is then "the disapeared GISS graph" showing a direct comparison between land and ocean trends. GISS land shows a somewhat different story than the 35 chunks of land spread of over the world from unadjusted GHCN.
GISS suggests for example, that land temperatures after 1975 rise significantly faster than ocean temperatures. But the puzzle is: The GHCN data does not confirm such a divergence between coastal and non-coastal stations after 1975.
As shown, the GHCN coastal vs. non-coastal certainly can show differences in more marine stations vs less marine stations. But no difference after 1975 as GISS suggests.
In fact, it is not just after 1975 that the differences coastal vs. non-coastal does not match the GISS temperature products.
Fig6. The “GISS land” has more than 0,8 K more heat trend than the GHCN Unadjusted non-coastal stations after 1925, and it happens to follow a near straight line.
There are more layers of adjustments from the GHCN Unadjusted datasets and to the final GISS product. The straight line shape of the above divergence after 1925 reminds me of the straight-line divergence we see between GHCN Unadjusted and GHCN adjusted shown by Giorgio Gilestro:
Fig 7 GHCN adjustments: A mostly straight line in adjustments after 1925 as Gilestro found may give the impression, that the changes in data where centrally decided rather than a randomly scientifically occurring from all kinds of error types and locations.
Should the Non-coastal unadjusted GHCN in fact have more similarities to a dataset like GISS-land at all? Is that a false expectation on my part?
For one thing, the non-coastal datasets represents just roughly 20-25% of the Earths Land area, but how likely is it, that 35 land chunks spread randomly out on the Earth just happens to have a trend very different from all other areas?
(None the less, a close to full Unadjusted GHCN land trend will be ready later this year, and we shall se if this changes anything. Hardly: Some areas not included are central USA, West China, Russia north of 60 N East Europe, Greenland, which are areas known to have small heat trend in the 20th century and from unadjusted GHCN they actually look pretty much like the non-coastal graph…)
Another protest would be: The GISS land includes the coast stations and therefore should have a warmer trend than pure non coastal data. – This is true, but the coast-line is small, and coastal area is likely much less than 10% of the Earths land area. So a “coastal pollution” of temperature data in any normal land temperature set is real, but it is likely to be just a fraction of the divergence between GHCN Unadjusted non-coastal and GISS land due to the small coastal area.
So it seems that there should be some degree of similarity between the 35 chunks of non-coastal land spread out on the globe compared with the combined GISS land ? (or Hadcruts CRUTEM etc.)
!! However, the point of this writing is primarily to simply show that coastal trends are often different from non-coastal trends, and show the main trend differences. The compares with GISS data, proxies and more. is not done to make final conclusions, just to show some relevant comparisons to other temperature trends. I feel its to early to conclude much on the GISS compares, but its relevant to make aware the nature of the differences.
In a natural perspective..
Fig 8. Sketch: Some natural effects on temperatures of the Earth. Graph of sunspot numbers is shown in grey, and may appear to show some compliance with the ocean (coastal) temperature trends over longer periods, that is: Large sunspot cycles seem to make ocean (coastal) temperature trends warm fast, although also the PDO appear important for ocean temperatures. Especially the PDO/AMO seems to show fair compliance with the land (non-coastal) trends.
(Pacific Decadal Oscillation = PDO and Atlantic Multidacadial Oscillation = AMO)
The only period where both PDO is positive and also Sunspot numbers are large is 1978-2005. Here, both coastal and non-coastal temperatures rise fast.
Fig 9. A: A Nice sequence starting when the Sun makes a giant sunspot circle boosting yearly global temperature a stunning 0,7 K up just in few years (!) but then soon the PDO and AMO goes negative and the new heat on Earth fizzles out. A good illustration of how nature can manipulate temperatures strongly. B: I find it interesting, that temperatures around 1910-20 are rather similar to temperatures in the 1970s. Both periods represents low PDO/AMO and sunspot numbers. Not much change over these 50-60 years in temperatures.
Fig 10. 10 year averages. Is the slower development in ocean temperatures partly due to a delay? The here-and-now heating effect of Earth might show faster for land temperatures than ocean temperatures, and thus, the land graph mostly indicates that the here-and-now heating effect of the Earth today is similar to the effect 1930-45? And perhaps the later warm trend of the ocean is partly because it takes time to warm large oceans?
Maybe so, but these thoughts are perhaps not fully supported by fig 10, since the heating after 1978 is just as rapid for ocean (coastal) as for land. In this case 1978-2005 of high Solar activity, the oceans shows no lag – as if the Solar effect should be more rapidly occurring for oceans (?)
Principles used when evaluating Coastal vs. non-coastal temperature trends.
(Below, the USA coastal areas are used as examples.)
For now, the following areas are used for non-coastal land (blue) and Coastal/Marine temperature trends:
Fig11. Overall Non-coastal (land) and Coastal (marine) temperature trends.
The marine air meets land in different altitudes:
Fig12. I have mostly used the lower coastline stations – red area - for coastal trends in this writing, but areas where higher altitude land meets the Marine air –brown areas - could be used too and normally matches the coastal trends very well. For non-coastal I have used mostly the blue areas but sometimes also green areas (Higher altitude, but behind other higher altitude areas.
Fig13: An example of the “Brown area”, where high altitude meets Marine air: The Himalayas. Here we see a mountain range with a “coastline” even though ocean-water is far away. A “mountain-coastline” can have temperature variation that differs from its surroundings as we see in this example.
Thus, estimating for example temperature development for Himalaya should only for a smaller fraction be based on temperature stations in this small “coast”-zone, see RUTI Himalaya.(Check the Pakistan-Indian studycases for South Himalaya data, see the end of the present writing)
Note: There are a few types of temperature stations, (example: River station in valleys down stream of icy mountain areas. These often show opposite trends compared to the surroundings, or stations in very icy areas where huge peaks and dives can occur) but for now I will focus on the above 4.
Some areas do not show much more heat trend in coastal/island stations than non-coastal:
This has mostly been observed not far from the equator or on coasts in bays oriented East-West.
Around equator winds are generally weak. The trends around equator are also fast changing (like land) due to the ENSO pattern that has a good match with land trends. An then coast running east-west may be less exposed to winds?
I have used 35 study cases, and I will start by illustrating examples from USA where many long temperature series are available and the mechanisms very easily detectable, that is: (The usefulness of my approach can best be confirmed or shown false using high quality data as available for USA)
Fig14. NE USA. In several cases the red coastal area also includes islands. In general for all categories it is best to avoid temperature stations with too much risk of UHI. For practically all stations near towns I have checked position on Google Maps to verify a rural or near rural location. (Yes, that was a lot of work..). If for example a coastal station not that far from a city shows a significantly different trend than the island rural station it will not be used. This I did to make sure that the warmer trend in for many coastal areas is not a UHI phenomenon.
Since the goal is to identify coastal vs. non-coastal trends, there is no need to use stations that appear partly coastal partly non-coastal, especially for areas like USA where so much quality data has been made publicly available.
Fig 15. Coastal stations, NE USA. ”Red stars” on fig 5, I used the superb site Appinsys.com for this, (fantastic site!!!)
Fig 17. Non-coastal stations, NE USA – “Blue stars”
Fig18. Here we see a not so surprising match between the 2 datasets that are not directly affected by marine air, the blue from lower altitude, and the green from higher altitude but sheltered by other high altitude areas closer to the coast.
Fig19. And here the match between islands/Coasts, and then the “frontline” of higher altitude areas, the “High altitude coastline”.
Fig20.And then the comparison of coastal – non-coastal, this is where the mismatch in data occurs, the core of my analysis on coastal - non-coastal stations.
Just curious, what does Hadrcrut make out of this? What are then the Hadcrut trends for NE USA?
Fig21.The black graph is the GHCN unadjusted versions of the Hadcrut stations used. The yellow graph is the Hadcrut version of Hadcrut data. (You cant compare the black with the yellow 100% because, not all Hadcrut stations is in GHCN)
I do find it suprising though, that the Hadcrut temperature trends appears so incredibly much warmer than the bulk of the area indicates, the non-coastal stations.
To see how Hadcrut has chosen ONLY coastal stations or Urban stations is sad. How can anyone with an honest attempt to do science, for this NE USA area choose 2 Boston stations, one in the larger city Syracuse, and then one in New York of all places, and then a (warm trended) coastal island station to represent the bulk of this area? Assuming that the professional scientists working with Hadcrut has not noticed the huge difference between coastal and non-coastal, then why are 4 of 5 stations urban???
And similar for SE USA:
Fig22. SE USA
Fig23. Again, we see a significant difference coastal vs. non-coastal, and Hadcrut more in line with the coastal fraction of the land area. In SE USA there are not as many large towns as in the NE area, but Hadcrut chose urban stations anyway: Raleigh, Jacksonville, Atlanta, Charleston. Some “scientist” seems to believe that "UHI is not a problem", but what science tells that one should choose mostly larger cities for temperature stations?
One more example from the USA, NW USA:
Fig24. NW USA. When the mountains reaches the coast with varying altitude, it get a little more complex to figure out which temperature stations are affected by Marine air and which aren’t. For NW USA we now have a large higher altitude area not much affected by Marine air, the green stars.
Fig25. Again, by using high quality data from USA one can to some degree make things add up despite mountains: The low and higher altitude stations seems to have a fair match in trends. No doubt, there are some stations that is hard to group due to the more complicated geographical scenery.
Fig26. And a match between the Coastal and “High-altitude” coastal sites.
Fig27. Colder trend for non-coastal stations as usual.
Now, I want to show you readears a situation where data was really hard to get my hands on.
One might think this was "Zambia" or something, but no, temperature data is hard to get hands on in Central Europe:
Fig28. For central Europe, the Atlantic “coastline” is a little complex, and so, like for Japan – NE China we compare oceanic Islands (The British) with a chunk of central Europe non-coastal.
Fig29. Significant difference coastal-noncoastal central Europe. I could have used Italy for "non-coastal", it would give the same result.
The non-coastal area "Central Beneluz and Germany area" is taken from
In this article it is shown how hard a work it is to get useful data from this area, but also how strong and "robust" the temperature trend finally became.
We might also want to analyse The British Islands themselves for coastal vs. non-coastal trends, but there is a problem doing so:
Fig30. I sketched the areas (blue areas) I find best for finding non-coastal stations. There where a good number of stations, but not one single stations (Except for Birmingham, London areas etc) enables a compare of the early warm years 1930-50 with the recent warm peak 1990-2010. Within the blue areas, not one single station has data before 1948, and we remember that the largest differences between coastal and non-coastal generally is 1920-45.
The numbers (like “51-64”) on the map are the years of data public available….
If the fine people of Ireland or UK decided to demand the full original data from these temperature stations, it would be very interesting: Remember, you have paid for these data with your taxes… The exact same one could say to the people of Spain, France, Belgium, Holland, Germany, Denmark and especially POLAND: Go get those temperatures, this would be perfect.
Here´s a Tip: See i you can arrange for a newspaper to ask the local meteorological institute for the missing data. I can help explain which stations are "missing".
Allegedly, The Irish could not collect temperatures properly before 1950, and British very poorly before 1950 and after 1975…
However, many nations could. For example for Pakistan, Bangladesh, Zambia, Zimbabwe, Paraguay, Mexico, Sudan, Japan etc. we have plenty plenty complete temperature series without interruption back to around 1900 – 1920. But the West Europeans failed to make a fair quality of temperature data?
35 study cases used for RUTI, Coastal temperature stations :
Many more RUTI writings and results, see links at the end of this writing.