Note to Readers: This is an important post, as Willis demonstrates that NASA GISS has taken a cooling trend and converted it into a warming trend for the one GHCN station in Nepal which covers the Himalayas. I offer NASA GISS, either via Jim Hansen or Gavin Schmidt, rebuttal opportunity to this issue on WUWT anytime. -Anthony
Guest Post by Willis Eschenbach
I read the excellent and interesting guest post by Marc Hendrickx about the IPCC and the Himalayas. My first big surprise was the size of the claimed warming. He cites IPCC Table 10.2 which says:
Nepal: 0.09°C per year in Himalayas and 0.04°C in Terai region, more in winter
Well, my bad number detector started ringing like crazy. A warming of nine degrees C (16°F) per century in the mountains, four degrees C per century in the lowlands? … I don’t think so. Those numbers are far too big. I know of no place on earth that is warming in general at 9°C per century.
Marc also quotes the IPCC source paper as saying:
The Kathmandu record, the longest in Nepal (1921–94), shows features similar to temperature trends in the Northern Hemisphere, suggesting links between regional trends and global scale phenomena.
Being cursed with a nagging, infernal curiosity, I thought I’d take a look at the Kathmandu temperature record. Foolish me …
I started by looking at where Nepal is located. It starts at the northern edge of the Indian plains, at the foothills of the Himalayas, and goes up to the crest:
Figure 1. Nepal (yellow outline). Yellow pins show all GHCN (Global Historical Climate Network) surface temperature stations.
So, that was my second surprise – a whole dang country, and only one single solitary GHCN temperature station. Hmmmm … as Marc shows, the paper cited by the IPCC gives the records of a dozen stations in Nepal. So why does GHCN only use Kathmandu in Nepal? But I digress.
Resolving not to be distracted by that, I went to the GISS dataset. I selected “Raw GHCN data + USHCN corrections” in the dropdown menu. (Kathmandu is outside the US, so in this case there are no USHCN corrections.) Typed in “Kathmandu”, and started the search. Figure 2 shows the result:
Figure 2. Kathmandu Air(port) Metadata.
This shows there are three records for Kathmandu Air (Airport), which looks promising. Also, it looks like there is an overlap between the records, which seems good. (There is no sign, however, of a record that is “the longest in Nepal (1921–94)”. The earliest date is 1951, and the latest is 2010. But again I digress.)
Clicking on the top Kathmandu Air link (on the GISS website, not on the graphic above) brings up the following GISS-generated graph:
Figure 3. Kathmandu Air. Three records are shown, as dotted, dashed, and blue.
Here’s an oddity. We have three records, each for different periods. And there is not a single year of overlap in the bunch. Not one.
Now, people think that I mine or search for these odd stations. Not so, I am simply curious about what I read, and this is not an atypical temperature record. Most are somewhat strange. Gaps and breaks in a given record often render large parts of the record unusable. GISS uses a cutoff of 20 years of consecutive data. As a result, the final GISS record for Kathmandu, rather than going from 1951-2009, goes from 1961 to 1980. Fair enough, these are all debatable choices, including the minimum record length cutoff size. In any case, the real problem with Kathmandu is not the record length. It is the lack of overlap which prevents the creation of a continuous record. This means that the apparent overall trend may not be real. It may simply be an artefact of e.g. different thermometers, or different locations. In this case, GISS has side-stepped the question by selecting only one record (shown in blue) for the final record.
How can we get to the graph of this final GISS record including all of their homogeneity adjustments? Well, we could go back to the same GISS website where we started and select a different dataset. However, here’s a trick to go directly from the raw data you are looking at to the final GISS homogenized dataset. Near the end of the URL of the raw GISS dataset under discussion you find the following:
… &data_set=0& …
GISS has three datasets. The raw data is dataset 0. The data “after combining records at the same location” is dataset 1. The final data “after cleaning/homogeneity adjustment” is dataset 2.
So to get the final adjusted result, all you have to do is to change the “0” in the URL to a “2”, viz:
… &data_set=2& …
Figure 3 shows the outcome of making that change:
Figure 3. Final GISS record for Kathmandu. The scale has been changed in both the X and Y axes. Note that they have discarded all segments of the record which are shorter than twenty years in length.
And that, dear friends, was my third big surprise. Take a close look at those two records, the adjusted and unadjusted …
As you no doubt observe, one is trending somewhat downwards, while the second is trending distinctly upwards. Hmmm … so, of course, I downloaded the GISS data (from the bottom of the same web page). Here is what they have done:
Figure 4. GISS Kathmandu Airport Annual Temperatures, Adjusted and Unadjusted, 1961–80. Yellow line shows the amount of the GISS homogeneity adjustment in each year. Photo is of Kathmandu looking towards the mountains.
GISS has made a straight-line adjustment of 1.1°C in twenty years, or 5.5°C per century. They have changed a cooling trend to a strong warming trend … I’m sorry, but I see absolutely no scientific basis for that massive adjustment. I don’t care if it was done by a human using their best judgement, done by a computer algorithm utilizing comparison temperatures in India and China, or done by monkeys with typewriters. I don’t buy that adjustment, it is without scientific foundation or credible physical explanation.
At best that is shoddy quality control of an off-the-rails computer algorithm. At worst, the aforesaid monkeys were having a really bad hair day. Either way I say adjusting the Kathmandu temperature record in that manner has no scientific underpinnings at all. We have one stinking record for the whole country of Nepal, which shows cooling. GISS homogenizes the data and claims it wasn’t really cooling at all, it really was warming, and warming at four degrees per century at that … hmmm, four degrees per century, where have I heard that before …
What conceivable scientific argument supports that, supports adding that linear 5.5°C/century trend to the data? What physical phenomena is it supposed to be correcting for? What error does it claim to be fixing?
Finally, does this “make a difference”? In the global average temperature, no – it is only one GHCN/GISS datapoint among many. But for the average temperature of Nepal, absolutely – it is the only GHCN/GISS datapoint. So it is quite important to the folks in Nepal … and infinitely misleading to them.
And when it is cited as one of the fastest warming places on the planet, it makes a difference there as well. And when the IPCC puts it in their Assessment Report, it makes a difference there.
Once again we see huge adjustments made to individual temperature records without reason or justification. This means simply that until GISS are able to demonstrate a sound scientific foundation for their capricious and arbitrary adjustments, we cannot trust the final GISS dataset. Their algorithm obviously has significant problems that lead to the type of wildly unreasonable results seen above and in other temperature datasets, and they are not catching them. Pending a complete examination, we cannot know what other errors the GISS dataset might contain.
[UPDATE] John Goetz pointed out that the likely source of the spurious trend is temperatures in Tingri (see Fig. 1, way back in the high mountains at the upper right at almost 6,000 metres elevation in the tundra) and GISS step 2. GISS says:
… in step 2, the urban and peri-urban (i.e., other than rural) stations are adjusted so that their long-term trend matches that of the mean of neighboring rural stations.
It seems John is right, Tingri is the likely problem. Or to be more accurate, their method is the problem. GISS uses a different method than GHCN to average stations for step 2.
The method of “first differences” is used by GHCN. GISS instead uses the “reference station” method described in the same citation. In my opinion, the reference station method is inferior to the first difference method.
The reference station here is likely Dumka, it is the longest of the nearby stations. Unlike Tingri, Dumka is at an elevation of 250 metres in the plains of West Bengal … hmmm. This should be interesting.
In the reference station method, Tingri gets adjusted up or down until the average temperatures match during the time of the overlap. Then Tingri and Dumka are averaged together. However, let’s take it a step at a time. First, I like to look at the actual underlying data, shown in Fig. 5.
Figure 5. Temperatures in Dumka (India) and Tingri (China). Left photo is Dumka on the lowland plains of West Bengal. Right photo is Tingri in the Himalayan mountains.
So the brilliant plan is, we’re going to use the average of the temperature anomalies in Dumka and Tingri to adjust the temperature in Katmandu, at 1,300 metres in the foothills?
Makes sense, I suppose. The average of mountains and plains is foothills, isn’t it? … but I digress.
The problem arises from the big jump in the Tingri data around 1970. Using the reference station method, that big jump gets wrapped into the average used to adjust the Kathmandu data. And over the period of Tingri/Kathmandu overlap (1963-1980), because of the big jump the “trend” of the Tingri data is a jaw-dropping 15°C per century. Once that is in the mix, all bets are off.
Obviously, there is some kind of problem with the Tingri data. The first difference method takes care of that kind of problem, by ignoring the gaps and dealing only with the actual data. You could do the same with the reference station method, but only if you treat the sections of the Tingri data as separate stations. However, it appears that the GISS implementation of the algorithm has not done that …
Nor is this helped by the distance-weighting algorithm. That weights the temperatures based on how far away the station is. The problem is that Tingri is much nearer to Kathmandu (197 km) than Dumka (425 km). So any weighting algorithm will only make the situation worse.
Finally, does anyone else think that averaging high mountain tundra temperature anomalies with lowland plains anomalies, in order to adjust foothills anomalies, is a method that might work but that it definitely would take careful watching and strict quality control?
[UPDATE 2] Lars Kamel pointed below to the CRU data. If we take all of the available CRU (originally GHCN) data, we get the following trend for Kathmandu.
Figure 6. CRU monthly and annual temperature data for Kathmandu. Red circles show those years with 12 months of data.
You can see the lack of a trend in the 1950-2000 data. Went down slightly 1950-1975, went up slightly 1975-2000. Gosh.
[UPDATE 3] Steve McIntyre reminded me below of his fascinating 2008 analysis of the numbers and locations of GISS adjustments that go up, down, and sideways. His post is here, and is well worth reading.
[UPDATE 4] Zeke Hausfather has an interesting post on Kathmandu here, with lots of good information.
Sponsored IT training links:
Pass your 000-977 guaranteed using E20-001 study guide and 642-973 dumps questions.
Turboblocke says:
August 12, 2010 at 3:20 am
Thanks, Turboblocke. It is quite true that they claim that they are using records from 119 temperature stations in Nepal, and from those records they say they have calculated a trend for all of Nepal of 0.6°C/decade for the country.
Now, perhaps you believe that, and you are free to do so. I have been burned far, far too many times to be that foolish. To date we don’t have so much as a list of names for these stations, much less the data and metadata for the stations. Weather Underground lists a total of 12 stations in Nepal. And so far I’ve seen data for exactly one station in Nepal. Do you have the data for the other 118 stations? Or for any other stations in Nepal?
I’m sorry, but if you want me to believe in the tooth fairy, I’ll have to see some teeth first. The 119 stations may exist … but we have no idea of the state of their records, how they have been averaged, how much data they are missing, whether they have been “adjusted”, and a host of other questions that need to be answered before believing the claim of a warming of 0.6°/decade. Their claim may well be true, but so far we have no evidence at all.
Finally, at least on my planet, I tend hold off on the saying that claims are “robust” until I actually see the data and algorithms used to produce the claims …
Robert says:
August 12, 2010 at 4:36 am (Edit)
Disproven? As Zeke says in his post,
Seems like my bad number detector isn’t so bad after all, since I claimed the 9°C was an exaggeration …
Also, until you or someone else can provide the data from the “119 stations” that they say they are using, claiming that the studies have been “validated” is a AGW joke.
Zeke has done an interesting analysis of the GSOD data. To see what the GSOD data looks like I took a look at a random Nepal GSOD station (444160) for a random year (1995). (GSOD data is available here.)
Here’s the extent of the 1995 data for that station:
Month, Days of Data
Jan, 4
Feb, 3
Mar, 14
Apr, 15
May, 10
Jun, 14
Jul, 6
Aug, 8
Sep, 9
Oct, 14
Nov, 17
Dec, 15
Only one month out of the year has more than half the days covered … BZZZT. Next contestant please.
I’m sorry, but there is a reason that GHCN doesn’t use those GSOD stations, and I suspect it is because their data is not adequate. Certainly for that month at that station it is not. Zeke, what were your restrictions on number and placement of days per month required to get a month’s average? (By “placement” I mean e.g. that if we have half the days of April covered, we’ll get a very different result if they are all in the first half of April rather than evenly spread throughout the month.)
I didn’t have time to read all the comments, so the following may have been covered. Perhaps it is true that you don’t need to cherry pick, but wouldn’t it be efficient to look at the stations they may find most lucrative to fudge?
Another issue I worry about is if they are willing to fudge these numbers, what happens when they start programming satellites to put enhanced numbers out as raw data? Who watches the watchmen?
There seems to be some misunderstanding of what I am discussing. I apologize for the lack of clarity in my writing.
I am discussing the curious adjustment made by GISS to the Kathmandu record. This is TOTALLY SEPARATE from the question of whether Nepal is warming, since none of the stations used to adjust the GISS Kathmandu record were in Nepal.
“”” Steven mosher says:
August 11, 2010 at 5:08 pm
George E. Smith says:
August 11, 2010 at 2:18 pm
Well since Temperature measurements are good out to 1200 km away from the thermometer; you don’t really need any more than one thermometer for Nepal;
no they are not. The estimation of the GLOBAL TREND OVER TIME is relatively INSENSITIVE to adding additional local stations or subtracting local stations. “””
Well Steven; you perhaps have not noticed that I have constantly complained that “Climate Scientists” don’t seem to have any knowledge of the Nyquist Sampling Theorem; or Sampled Data Theory in general. They seem to act as if statistical manipulations can correct for incorrect sampling. No statistical process can correct the aliassing noise consequences of violation of the Nyquist Sampling Theorem Criterion.
The current methodology of climate data recording; well I suppose it is really weather data; doesn’t even comply with the Nyquist Criterion for even the daily average Temperature calculation at a single station. The result is aliassing noise that makes the true daily average unrecoverable. And as for the spatial sampling which I tongue in cheek mentioned as the 1200 km sufficiency; that fails by orders of magnitude to satisfy the requirements.
So the whole process of recording “global Temperatures” is a complete farce. Now the fans of GISS etc insist that “The Anomalies” which are the simple differences between one unknown number and another unknown number, are “coherent” over distances up to 1200 km. They never explain just what “Coherent” actually means in that context.
In any case that is still irrelevent since the true average global mean temperature over whatever baseline 30 year or whatever time frame they choose, is also a completely unknown number for the very same sampling failure reasons.
And even if they corrected their methods; which they can’t do, because it would take all the money on the planet to buy enough thermometers; it is all for naught, since there is no physical cause and effect connection between a local surface or near surface Temperature measurement, and the energy flows that are occurring at that location at that time; so mean global temperature tells us nothing about whether the earth is gaining or losing total energy.
Willis,
My post focused on record 444540 (Kathmandu Airport), which has a pretty complete record from 1976-present. Many of the other Nepalese GSOD stations are much more fragmentary.
I agree that we were discussing somewhat different subjects (e.g. you focused much more on the GISTemp adjustments). As I mentioned, I see little point in divining manipulations out of the entrails of urban station adjustments. Contrary to popular conception, the -only- adjustments GISTemp makes to individual land station records (with two small exceptions) are to correct for UHI. They effectively replace the temperature record of all urban stations with the distance-weighted average of nearby rural stations. This will result in the station in question being a less accurate representation of the temperature at that specific location (where, for example, UHI is a real increase in temperature), but more characteristic of the region that it is located in. The net effect globally is to reduce the trend over the past century, akin to discarding all urban stations.
I’m trying to track down a full set of station anomalies pre- and post- GISTemp STEP 2 to quantify the distribution of adjustments, and see how their adjustment compares to a temperature record constructed by simply discarding all urban stations.
Eschenbach: I am discussing the curious adjustment made by GISS to the Kathmandu record. This is TOTALLY SEPARATE from the question of whether Nepal is warming, since none of the stations used to adjust the GISS Kathmandu record were in Nepal.
Utter nonsense. Since the GISS urban adjustments are made with reference to rural stations surrounding around Kathmandu, the adjustment is HIGHLY DEPENDENT on whether regional rural stations are warming. Evidence from both GSOD and Shrestha’s paper support the URBAN adjustments made to Kathmandu. You might think that regional climate trends stop at national borders, but no one else does.
Eschenbach: “I’m sorry, but there is a reason that GHCN doesn’t use those GSOD stations, and I suspect it is because their data is not adequate. Certainly for that month at that station it is not. Zeke, what were your restrictions on number and placement of days per month required to get a month’s average?”
What data QC filters (purely statistical within the data set, not looking for meta-data adjustments) would you like to see Willis? Even though people are often asking for raw data – name your tune on the QC filters as far as minimum days-per-month, months-per-season, months-per-year, and years-in-baseline and I will happily play it.
Ron Broberg says:
August 12, 2010 at 5:15 pm
As Steven Mosher has spoken about elsewhere, these are choices. You pays your money, and you takes your choice. All choices are more or less defensible.
However, as the number of datapoints necessary to get say a monthly average decreases, the less accurate your results will be, and the more likely that they contain a spurious trend. This is because humans tend to do things like not want to take the temperature when it is really cold, or to go on a vacation the same time every year, and the like. Either of these can easily introduce a large spurious trend. For example, there’s lots and lots of holes in the latter (post 1990) Kathmandu record, which in this case happens to give it a spurious uptrend, as the missing data is more in the winter than the summer …
Zeke Hausfather says:
August 12, 2010 at 2:27 pm
Thanks, Zeke. Your definition of “pretty complete” is pretty different than mine, in that case. Take a look at the record in Update 4 of the head post … doesn’t look at all complete to me. There are huge gaps in the 1980s, and in fact, it is quite demonstrable that the lack of data in the last ten years has led to an artificial inflation of the trend.
So no, I would say it is not complete at all.
I’m sorry Willis, somewhere in your reply, I missed the actual filters you would prefer.
What are the days-in-month, months-in-season, months-in-year, and years-in-baseline that *you* would prefer to see in a QC filtered GSOD? You seemed comfortable rejecting one station based on (lack of) data from one year. Would you be willing to formalize your rejection criteria? Or are you going to continue playing it from the gut?
http://www.nerc-wallingford.ac.uk/ih/www/research/SAGARMATHA/volume2.pdf
Source is not IPCC or GISS.
barry says:
August 12, 2010 at 10:14 pm
Thanks, barry. The problem is not the source. It is the lack of data. Without the data, we have no way of knowing whether they have done the math correctly. I’ve seen too much bad math in this game to blindly trust that. I also find the idea that there are “119 temperature gauges” in Nepal to be somewhat extreme, but maybe that’s just me.
I’m not saying that they are wrong. I’m saying without the data and station names and the like, we cannot know if they are right.
” Zeke Hausfather says:
August 12, 2010 at 2:27 pm
Contrary to popular conception, the -only- adjustments GISTemp makes to individual land station records are to correct for UHI. They effectively replace the temperature record of all urban stations with the distance-weighted average of nearby rural stations. This will result in the station in question being a less accurate representation of the temperature at that specific location (where, for example, UHI is a real increase in temperature), but more characteristic of the region that it is located in. The net effect globally is to reduce the trend over the past century, akin to discarding all urban stations.”
So you think that rural stations are warming faster than urban station?
Or do you think that the yellow curve in “Bringing the Heat to Katmandu Air” has a downward slope?
Ron Broberg says:
August 12, 2010 at 8:06 pm
Thanks, Ron. The answer to your question is far from simple. It depends on the purpose of the data analysis, and how hard we want to squeeze the data.
In my Update 2 above, I show the first level of the data selection criteria that I use. This is the restriction that to be counted, a year has to have all 12 months of data. This is the easiest method, because all other criteria (allowing partial months or partial years) results inevitably in a distortion of the trend.
This is because of the variable nature of the temperature. For example, if we leave out a winter month, the year will average higher, and it will appear as a false upward trend. And a missing summer month has the opposite effect.
So if I’m looking at trends, that’s not a good thing. There are mathematical ways to minimize the effects of this, both for missing days and for missing months. But we can never get it back to zero.
For missing months, we have a couple of possibilities. One is to infill the data with our best estimate of what the missing temperatures really were. There are a variety of mathematical methods for doing this. As the number of adjacent missing months increases, the accuracy of this method declines. In addition, missing months that contain maximum/minimum temperatures have more effect on the trend than other months, and are harder to estimate.
Another way to adjust for the problem is through compensation. This means to omit the month directly opposite (six months out) in the same year. So if August is missing, you remove all data for February as well. This avoids having to guess at the temperature in August, and keeps your yearly average near the true answer. I tend to infill rather than omit.
With that as prologue, above I gave the first level of data selection criteria I use (all months in a year must be present). Second level is to allow 1 missing month, provided it is either estimated or compensated for as described above. Third level is to allow 2 missing months, as long as they are in different quarters and are compensated or estimated, and so on. More than three missing months makes me nervous. I will often work slowly down through the levels and graph the results using the different criteria, so I can see the patterns as they evolve.
Regarding daily data, the same rules apply. First level is no missing days, then 1, and so on. Many (in some records most) months have a number of missing days. So rules for days are perforce less strict than for months.
Now, you don’t need to have all the days to get a reasonable monthly average. For example, you could get a decent estimate of the monthly temperature average if you took the temperature every other day. On the other hand, if you had data for half the days and they were all in the first half of the month, your monthly average could be way off. So the distribution of the days with data in the month is as important as the absolute number of days. A criterion that I commonly use is that a month must have at least half the days, and no gaps longer than 4 days.
Again, however, I put all of these criteria in as variables, so I can adjust the criteria and see the results. That way I can determine, what does the complete data say? What does adding in the second-best data out of the bunch say? What happens to the trend as we add in worse and worse data?
When there is less and less daily data, more ingenious methods can be used to leverage what little data there is. For example, when data is scarce I sometimes use cluster averaging rather than straight averaging, to correct for clumped daily data and thus allow the use of less data.
So that’s how I do it. I asked about Zeke’s criteria to see how he is handling that question. There is no absolute “right way”, we’re talking about choices here, and I use different criteria for different situations and conditions. About the only thing you can’t do is use years with missing months without adjusting for that through estimation or compensation.
Yes, that is clear.
Over at the Blackboard the “experts” seem to agree that rural stations are warming faster than urban stations, and that Kathmandu temperatures are not Kathmandu temperatures but those of these fast warming rural stations.
Bill Tuttle
Re Darjeeling’s population:
Data from:
URBAN MANAGEMENT IN DARJEELING HIMALAYA -A CASE STUDY OF DARJEELING MUNICIPALITY
At Hansen’s Gourmet Climate Pizza Shop, we serve only the best pies!
Why are our pies so tasty to the liberal mind? Because we don’t use bulk cooking methods; we cook the data one weather station at a time. Each earth location’s data is handled separately, adding here and subtracting there, to make each scrumptious time series perfectly seasoned to compliment the whole pie. That’s why the media and liberals swallow the whole pie in one big bite.
Hansen’s Gourmet Climate Pizza Shop: We Do It Left (and that’s right)!
Unfair to GISS! Hansen actually publishes his [snip] “corrections”. GHCN and CRU keep theirs close to the chest
I’ve enjoyed reading all of the comments but I have to say, all of you are wrong! Nepal / Katmandhu has definitely increased in temperature by the 5 degrees C on average that GISS says is has in the past century and will continue to do so in every other century. And this my friends is due to the fact that you are all overlooking one important factor: Nepal / Katmandhu is way up there is those there mountains and thus is closer to the sun.
David L. Hagen: August 13, 2010 at 12:53 pm
Bill Tuttle
Re Darjeeling’s population:
ZOMG! Has anyone told GISS? They’re gonna have to double Darjeeling’s UHI adjustment, making it Worse Than They Thought!
PeterK: August 13, 2010 at 9:30 pm
I’ve enjoyed reading all of the comments but I have to say, all of you are wrong! Nepal / Katmandhu has definitely increased in temperature by the 5 degrees C on average that GISS says is has in the past century and will continue to do so in every other century. And this my friends is due to the fact that you are all overlooking one important factor: Nepal / Katmandhu is way up there is those there mountains and thus is closer to the sun.
Good catch, Peter, and with plate tectonics causing an increased uplift in the Himalayas, that 5ºC could very well be on the low side — the sun is ‘way hotter than the inside of the Earth, and it’s millions of degrees down there!
*koff*
Willis you can estimate the importance of missing months by just dropping them
Guess what happens if I drop all the decembers?