Here’s a story about how one missing letter, an M, can wreck a whole month’s worth of climate data. It is one of the longest posts ever made on WUWT, I spent almost my entire Saturday on it. I think it might also be one of the most important because it demonstrates a serious weakness in surface data reporting.
In my last post, we talked about the a curious temperature anomaly that Jean S. found in the March GISS data and posted at Climate Audit:
The anomaly over Finland has an interesting signature to it, and the correction that GISS posted on their website confirms something I’ve been looking at for a few months.
The data shown between 4/13 and 4/15 were based on data downloaded on 4/12 and included some station reports from Finland in which the minus sign may have been dropped.
With some work I started back in late December and through January, and with GISS putting stamp of approval on “missing minus signs” I can now demonstrate that missing minus signs aren’t just an odd event, they happen with regularity, and the effect is quite pronounced when it does happen. This goes to the very heart of data gathering integrity and is rooted in simple human error. The fault lies not with GISS (though now they need a new quality control feature) but mostly with NOAA/NCDC who manages the GHCN and who also needs better quality control. The error originates at the airport, likely with a guy sitting in the control tower. Readers who are pilots will understand this when they see what I’m talking about.
I’ve seen this error happen all over the world. Please read on and be patient, there is a lot of minutiae that must be discussed to properly frame the issue. I have to start at the very bottom of the climate data food-chain and work upwards.
First, a discussion about the root of error and the differences between the surface and satellite dataset. I should mention that in the satellite image from NASA’s Earth Observations (NEO), we don’t see the same error as we see in the GISTEMP map above.
Why? Better sensors, maybe, but mostly it has to do with a different data gathering methodology. In the surface data sets, including land and ocean data, most every datapoint is touched by a human hand, even airport data that gets done by automated airport sensors sometimes gets transcribed manually (often in third world and technologically stunted countries). In the surface data, thousands of sensors are spread across the globe, many different designs, many different exposures, many different people with different standards of measurement and reporting. The precision, accuracy, and calibration of the vast surface network varies, especially when we have broad mix of instrumentation types.For example in the US Historical Climatological Network the equipment varies significantly.
In satellite data, the data is measured at a single point with one sensor type, the Microwave Sounder Unit on the satellite, calibrated to a precision source on-board. On-board redundant precision platinum resistance thermometers (PRTs) carried on the satellite radiometers. The PRT’s are individually calibrated in a laboratory before being installed in the instruments. The satellite data is automatically measured and transmitted. In contrast to the surface temperature record, no human hands touch the data gathering or data reporting process. Satellite data generation is far more homogeneous than the mish-mash of surface data.
I think it would be safe to say that the chances of human error in raw surface data are at least an order of magnitude greater (if not several) than error in raw satellite data. Post measurement processing is another issue, but for the purposes of this essay, I’m focusing only on raw data gathering and transmittal.
As mentioned in the recently updated compendium of issues with the surface temperature data by Joe D’Aleo and myself, there has been a move in the Global Historical Climatological Network (GHCN) to rely more and more on airports for climate data. This, in my opinion, is a huge mistake because in addition to those issues
E.M. Smith aka “Chiefio” reports that in GISS (which uses GHCN) worldwide, there has been a wholesale migration towards airport weather data as a climatic data source. In an email sent to me on Jan 20, 2010 he says that
Look at:
http://chiefio.wordpress.com/2009/08/26/agw-gistemp-measure-jet-age-airport-growth/
which as a fairly good descriptions of the problems in the data, we have a global report for GHCN as of that August data. There is more deail in the link, but I think you care about “now”:
Percentage of sites that are AIRPORTS NOW, by decade of record Year S.P S.C S.T S.W EQ. N.W N.T N.C N.P Total 1909 0.0 42.0 15.1 28.2 29.2 36.7 22.8 33.3 44.4 25.4 1919 0.0 36.4 12.8 23.5 25.1 37.7 20.9 35.0 39.8 24.1 1929 0.0 37.0 11.9 27.4 27.7 32.7 20.4 35.9 56.4 24.1 1939 0.0 43.9 17.6 32.0 33.8 29.1 20.2 36.2 51.0 25.1 1949 0.0 32.3 24.4 37.6 44.4 31.8 23.3 39.3 60.9 29.1 1959 0.0 24.0 35.0 50.0 59.4 39.4 30.9 41.0 62.9 37.3 1969 0.0 18.1 39.3 53.2 63.2 40.2 31.4 41.1 61.5 39.0 1979 0.0 17.9 39.1 52.0 64.2 40.7 28.8 41.1 62.3 37.7 1989 0.0 20.7 41.5 52.5 67.8 41.9 29.1 40.8 64.9 37.7 1999 0.0 21.0 53.5 57.4 68.0 53.0 32.6 49.0 59.0 41.6 2009 0.0 17.9 74.0 64.7 66.5 51.5 30.2 45.4 57.3 41.0
I do break outs by continent and by some countries. For the USA, I further do a specific with / without USHCN (the older version, not the USHCN.v2 put in 15Nov09) and findFor COUNTRY CODE: 425
LATpct: 2006 3.7 18.3 29.5 33.2 14.4 0.0 0.4 0.3 0.1 0.1 100.0 AIRpct: 1.3 4.0 6.3 6.7 3.2 0.0 0.4 0.3 0.1 0.1 22.4 LATpct: 2007 8.2 17.2 28.4 26.9 11.2 0.0 3.7 3.0 0.7 0.7 100.0 AIRpct: 8.2 15.7 27.6 23.1 9.0 0.0 3.7 3.0 0.7 0.7 91.8 LATpct: 2008 8.8 16.9 28.7 26.5 11.0 0.0 3.7 2.9 0.7 0.7 100.0 AIRpct: 8.8 15.4 27.9 22.8 8.8 0.0 3.7 2.9 0.7 0.7 91.9 LATpct: 2009 8.1 17.8 28.1 26.7 11.1 0.0 3.7 3.0 0.7 0.7 100.0 AIRpct: 8.1 16.3 27.4 23.0 8.9 0.0 3.7 3.0 0.7 0.7 91.9 DLaPct: 2009 4.3 18.4 29.5 32.5 13.6 0.0 0.7 0.9 0.2 0.1 100.0 DArPct: 2.1 5.7 8.8 8.9 3.7 0.0 0.6 0.8 0.2 0.1 30.7
That in the YEAR 2009 the USA has almost 92% airports in GHCN.
So clearly, airports make up a significant portion of the climate data.
On the issues of airports as climate station, obvious issues with siting, UHI, failing ASOS instrumentation, and conflicting missions (aviation safety -vs-climate) aside, I’m going to focus on one other thing unique to airports: METAR
What is METAR you ask? Well in my opinion, a government invented mess.
When I was a private pilot (which I had to give up due to worsening hearing loss – tower controllers talk like auctioneers on the radio and one day I got the active runway backwards and found myself head-on to traffic. I decided then I was a danger to myself and others.) I learned to read SA reports from airports all over the country. SA reports were manually coded teletype reports sent hourly worldwide so that pilots could know what the weather was in airport destinations. They were also used by the NWS to plot synoptic weather maps. Some readers may remember Alden Weatherfax maps hung up at FAA Flight service stations which were filled with hundreds of plotted airport station SA (surface aviation) reports.
The SA reports were easy to visually decode right off the teletype printout:
Note that in the example above, temperature and dewpoint are clearly delineated by slashes. Also, when a minus temperature occurs, such as -10 degrees Fahrenheit, it was reported as “-10”. Hang on to that, it is important.
The SA method originated with airmen and teletype machines in the 1920’s and lasted well into the 1990’s. But like anything these days, government stepped in and decided it could do it better. You can thank the United Nations, the French, and the World Meteorological Organization (WMO) for this one. SA reports were replaced by METAR in 1996.
From Wikipedia’s section on METAR
METAR reports typically come from airports or permanent weather observation stations. Reports are typically generated once an hour; if conditions change significantly, however, they can be updated in special reports called SPECIs. Some reports are encoded by automated airport weather stations located at airports, military bases, and other sites. Some locations still use augmented observations, which are recorded by digital sensors, encoded via software, and then reviewed by certified weather observers or forecasters prior to being transmitted. Observations may also be taken by trained observers or forecasters who manually observe and encode their observations prior to transmission.
History
The METAR format was introduced 1 January 1968 internationally and has been modified a number of times since. North American countries continued to use a Surface Aviation Observation (SAO) for current weather conditions until 1 June 1996, when this report was replaced with an approved variant of the METAR agreed upon in a 1989 Geneva agreement. The World Meteorological Organization‘s (WMO) publication No. 782 “Aerodrome Reports and Forecasts” contains the base METAR code as adopted by the WMO member countries.[1]
Naming
The name METAR is commonly believed to have its origins in the French phrase message d’observation météorologique pour l’aviation régulière (“Aviation routine weather observation message” or “report”) and would therefore be a contraction of MÉTéorologique Aviation Régulière. The United States Federal Aviation Administration (FAA) lays down the definition in its publication the Aeronautical Information Manual as aviation routine weather report[2] while the international authority for the code form, the WMO, holds the definition to be aerodrome routine meteorological report. The National Oceanic and Atmospheric Administration (part of the United States Department of Commerce) and the United Kingdom‘s Met Office both employ the definition used by the FAA. METAR is also known as Meteorological Terminal Aviation Routine Weather Report or Meteorological Aviation Report.
I’ve always thought METAR coding was a step backwards, for reasons I’ll discuss shortly.
But first, quick! Spot the temperature and dewpoint in this METAR report:
The following is an example METAR from Burgas Airport in Burgas, Bulgaria, and was taken on 4 February 2005 at 16:00 Coordinated Universal Time (UTC).
METAR LBBG 041600Z 12003MPS 310V290 1400 R04/P1500N R22/P1500U +SN BKN022 OVC050 M04/M07 Q1020 NOSIG 9949//91=
Could you read this and know what the weather is in Burgas? I can, only becuase I’ve looked at hundreds the past few months, but I still have to pick through the report to find it. The reason is that METAR is a variable field reporting format. Data isn’t always in the same position.
In the report above. The temperature and dewpoint is: M04/M07
M04/M07 indicates the temperature is −4 °C (25 °F) and the dewpoint is −7 °C (19 °F). An M in front of the number indicates that the temperature/dew point is below zero (0) Celsius.
Notice also that the entire METAR report is visually more complex. This is fine if you are having computers code it, but many METAR reports are still hand coded by technicians at airports, and thus begins the introduction of human error into the climate data process. Complexity is not a good thing when manual labor is involved as it increases the likelihood of error.
Here is where METAR coding departs from normal numeric convention. SA reports did not have this problem.
In the METAR report above, instead of using the normal way we treat and write negative numbers, some policy wonk decided that we’ll use the letter “M” to report a negative number. Only a bureaucrat could think like this.
So instead of a below zero Centigrade temperature and dewpoint looking like this:
-04/-07
in the “new and improved” METAR coding, it looks like this:
M04/M07
OK not a problem you say? Well I beg to differ, because it forces technicians who manually code METAR reports for transmission to do something they would not do anywhere else, and that’s write down an “M” instead of a minus sign. Using an M is totally counter-intuitive and against basic math training, and increases the likelihood of error.
It gets worse. Let’s say the technician makes a boo-boo and puts a minus sign instead of an “M” in front of the numbers for temperature/dewpoint. You’d think this would be alright, and the system would correctly interpret it, right?
Let’s put the METAR report from Burgas Airport into an online METAR decoder.
http://www.wx-now.com/Weather/MetarDecode.aspx
Here’s the report with the easy to make mistake, using minus sign instead of M for the temperature.
METAR LBBG 041600Z 12003MPS 310V290 1400 R04/P1500N R22/P1500U +SN BKN022 OVC050 -04/M07 Q1020 NOSIG 9949//91=
The output from the online METAR decoder reads:
Hey look at that, the temperature is 39°F (3.8°C). Minus signs are discarded from METAR decoding. Note that decoded METAR temperature also comes out the same if the “M” is missing in front of the 04/-07 or 04/M07
If it had been decoded correctly we would have gotten:
(-4) degrees Celsius = 24.8 degrees Fahrenheit
A whole 14.2 degrees F difference!
Reference for METAR decoding:
http://www.met.tamu.edu/class/METAR/quick-metar.html
Also note that METAR data has no year stamp component to the data, so the METAR decoder has no way of knowing this was a report from 2005, not 2010. Since each METAR report is essentially disposable within 24 hours, this presents no problem for pilots, they don’t care. But if you are tracking climate over years using METAR data, not having a year time stamp increases the likelihood of error.
Also the temperature error itself in this case has no bearing on a pilot’s decision to takeoff or land. Unless they are worried about density altitude on hot humid days, the temperature is a throwaway datum. They are mostly concerned about winds, sky conditions, and barometer (altimeter setting). In fact cool/cold days are far better for aviators; see AOPA’s Why Airplanes Like Cool Days Better.
My point here is this:
If a pilot or tower controller sees an erroneous METAR report like this:
METAR LBBG 041600Z 12003MPS 310V290 1400 R04/P1500N R22/P1500U +SN BKN022 OVC050 -04/M07 Q1020 NOSIG 9949//91=
Or this:
METAR LBBG 041600Z 12003MPS 310V290 1400 R04/P1500N R22/P1500U +SN BKN022 OVC050 04/M07 Q1020 NOSIG 9949//91=
Pilots/controllers/dispatchers aren’t likely to care, since current temperature and dewpoint are not important to them at these cooler temperatures. They also aren’t likely to call up the tower and holler at the technician to say “Hey, the temperature is wrong!”. Further, since the METAR report may be reissued sometime within the hour if somebody DOES spot the error, problem solved.
Point is that updates/corrections to METAR data may not be logged for climate purposes, since they are likely to be seen as duplicate reports because of the hourly timestamp.
So, in the case of M’s and minus signs, the propensity exists for erroneous METAR reports to not get corrected and to stay logged in the system, eventually finding their way into the climate database if that airport happens to be part of GISS, CRU, or GHCN datasets.
Maddeningly, even when egregious errors in aviation weather data are pointed out and even acknowledged by the reporting agency, NOAA keep them in the climate record as was demonstrated last year in Honolulu, HI International Airport when a string of new high temperature records were set by a faulty ASOS reporting station. NOAA declined to fix the issue in the records:
NOAA: FUBAR high temp/climate records from faulty sensor to remain in place at Honolulu
The key sentence from that story from KITV-TV:
The National Weather Service said that is not significant enough to throw out the data and recent records.
Hmmm, look at another nearby station and compare the differences. You be the judge.
Does NOAA consider this a climate reporting station? Yes according to NCDC MMS database, it is part of the “A” network, designated for climate:
Clearly, NOAA simply doesn’t seem to care that erroneous records finds their way into the climatic database.
OK back to the METAR issue.
The problem with METAR reporting errors is worldwide. I’ve found many examples easily in my spare time. Let’s take for example, a station in Mirnvy, Russia. It is in Siberia at 62.5° N 113.9° E and has an airport, is part of GHCN, and reports in METAR format.
Weather Underground logs and plots METAR reports worldwide, and these METAR reports are from their database on November 11th, 2009.
It shows a clear error in the 12:30PM (330Z) and 1 PM (400Z) METAR report for that day:
UERR 010330Z 22005G08MPS 9999 -SN 21/M23 Q1026 NOSIG RMK QFE738 24450245 UERR 010400Z 22005G08MPS 9999 -SN SCT100 OVC200 20/M22 Q1025 NOSIG RMK QFE737 24450245 UERR 010430Z 21005G08MPS 4000 -SN SCT100 OVC200 M20/M22 Q1024 NOSIG RMK QFE737 24450245 UERR 010430Z 21005G08MPS 4000 -SN SCT100 OVC200 M20/M22 Q1024 NOSIG RMK QFE737 24450245 UERR 010500Z 21005G08MPS 4000 -SN SCT100 OVC200 20/M22 Q1023 NOSIG RMK QFE736 24450245
Note the missing ” M” on the 12:30PM (330Z) and 1 PM (400Z). It happens again at 2PM (500Z). Of course it isn’t very noticeable looking at the METAR reports, but like the GISS plot of Finland, stands out like a sore thumb when plotted visually thanks to Weather Underground:
Mirnvy, Russia
The effect of the missing “M” is plotted above, which coincidentally looks like an “M”.
Put those METAR reports in this online METAR decoder: http://www.wx-now.com/Weather/MetarDecode.aspx and you get 70F for 12:30PM and 68F for 1PM
What do you think 70 degree F spike this will do to monthly averaged climate data in a place where the temperature stays mostly below freezing the entire month?
http://www.wunderground.com/history/airport/UERR/2009/11/1/MonthlyHistory.html?MR=1
Does NOAA log METAR data from Mirnvy Russia (ICAO code UERR)?
Yes, they do. Plus many other METAR reporting stations discussed here.
Does NCDC classify it as a climate station?
Yep, it is part of the “A” network. Which means it either directly reports climate data and/or is used to adjust data at other stations, such as GHCN stations.
List of GHCN stations:
http://www1.ncdc.noaa.gov/pub/data/ghcn/daily/ghcnd-stations.txt
It is not however, part of GHCN. But there are plenty stations that have this error that are part of GHCN. Yakutsk, Russia, also in Siberia is part of GHCN and has a METAR reporting error. Here’s an example what one off-coded hourly reading will do to the climate database.
The city of Yakutsk, one of the coldest cities on earth, reported a high of 79˚F on November 14th with a low of -23˚F.
Weather Underground seems to have done some quality control to the METAR reports, but the erroneous high temp remains in the daily and monthly report:
http://www.wunderground.com/history/station/24959/2009/11/14/DailyHistory.html
http://www.wunderground.com/history/station/24959/2009/11/14/MonthlyHistory.html
A month later, it happened again reporting a high of 93˚F on December 14th with a low of -34˚F
And the erroneous 93F high temp remains in both the daily and monthly reports, but has been removed from the METAR report, so I can’t show you the missing “M” I observed back in January. I wish I had made a page screen cap.
http://www.wunderground.com/history/station/24959/2009/12/14/DailyHistory.html
http://www.wunderground.com/history/station/24959/2009/12/14/MonthlyHistory.html
When the temperature data was calculated with that error then, this was found:
The average for the day, 30˚F, was some 67˚F above normal, pushing the anomaly for the month of December from 3.6˚F above normal to 5.9˚F above normal… quite a shift!
More examples:
Here’s an example of a properly coded METAR report from Nizhnevartovsk, Russia, for December 11, 2009, but the data is wrong. I’m thinking it was supposed to be M30 but came out M13. The dewpoint value M16 is also erroneous.
Nizhnevartovsk, Russia Dec 7, 2009
METAR USNN 111230Z 00000MPS P6000 SCT026 OVC066 M27/M29 Q1014 NOSIG RMK QFE755 SC062 METAR USNN 111300Z 12005G08MPS P6000 SCT066 OVC200 M13/M16 Q1035 NOSIG RMK QFE772 SC063 METAR USNN 111330Z 12005G08MPS P6000 SCT066 OVC200 M13/M16 Q1035 NOSIG RMK QFE772 SC063 METAR USNN 111400Z 00000MPS P6000 SCT020 OVC066 M28/M29 Q1014 NOSIG RMK QFE755 SC065
And it was not a one time occurrence, happening again on Dec 25th as shown in the Monthly graph:
Nizhnevartovsk, Russia, December 2009
http://www.wunderground.com/history/airport/USNN/2009/12/25/MonthlyHistory.html
The daily graph and METAR reports, notice it happened about the same time (1300Z) and in the same way (M27 then M13) , perhaps pointing to the same technician on duty making the same habitual mistake again. Maybe too much Vodka having to work the Xmas night shift?
Nizhnevartovsk, Russia Dec 25, 2009
METAR USNN 251230Z 11006MPS 2200 -SN SCT014 OVC066 M27/M30 Q1015 NOSIG RMK QFE757 SC055 METAR USNN 251300Z 35002MPS 6000 -SN SCT015 OVC066 M13/M15 Q1010 NOSIG RMK QFE752 SC03 METAR USNN 251330Z 12006MPS 4100 -SN SCT015 OVC066 M27/M29 Q1014 NOSIG RMK QFE756 SC055
http://www.wunderground.com/history/airport/USNN/2009/12/25/DailyHistory.html
It did not appear initially to be in the GHCN list or on the GISS list, but I’ve found that some of the names on Weather Underground are different from the place names in the GHCN and GISS lists. It turns out that if you search in Weather Underground for the station ALEKSANDROVSKOE it will point you to use the data from Nizhnevartovsk. ALEKSANDROVSKOE is a GHCN/GISS station.
I found other instance of METAR errors for that station, this one was quite pronounced on Jan 16th, 2009, lasting for 7 hours before it was corrected.
Nizhnevartovsk, Russia Jan 16, 2009
Here’s the METAR reports
METAR USNN 151800Z 23002MPS P6000 BKN066 OVC200 M22/M24 Q1009 NOSIG RMK QFE751 SC038 METAR USNN 151830Z 23002MPS 2900 -SHSN SCT020CB OVC066 22/M23 Q1009 NOSIG RMK QFE751 SC038 METAR USNN 151900Z 23002MPS 2100 -SHSN SCT019CB OVC066 21/M23 Q1009 NOSIG RMK QFE751 SC038 METAR USNN 152000Z 24001MPS 5000 -SHSN SCT022CB OVC066 21/M22 Q1009 NOSIG RMK QFE751 SC038 METAR USNN 152030Z 24002MPS 4300 -SHSN SCT020CB OVC066 21/M22 Q1009 NOSIG RMK QFE751 SC038 METAR USNN 152100Z 24002MPS 6000 -SHSN SCT018CB OVC066 20/M22 Q1009 NOSIG RMK QFE751 SC038 METAR USNN 152130Z 25002MPS P6000 SCT020CB OVC066 20/M22 Q1009 NOSIG RMK QFE751 SC038 METAR USNN 152200Z 28002MPS P6000 SCT022CB OVC066 20/M22 Q1009 NOSIG RMK QFE752 SC038 METAR USNN 152300Z 27003MPS P6000 -SHSN SCT016CB OVC066 M19/M21 Q1010 NOSIG RMK QFE752 SC038
http://www.wunderground.com/history/airport/USNN/2009/1/16/DailyHistory.html
The monthly report shows the event:
Nizhnevartovsk, Russia, January 2009
http://www.wunderground.com/history/airport/USNN/2009/1/16/MonthlyHistory.html
It happened twice on Feb 2nd, 2007, and with a space added between the M and 09 on the 0300Z report, it is a clear case of human error:
METAR USNN 020100Z 11010G15MPS 0500 R03/1200 +SN +BLSN VV002 M09/M11 Q1003 TEMPO 0400 +SN +BLSN VV002 RMK QFE748 QWW060 MOD ICE MOD TURB S METAR USNN 020200Z 12009G14MPS 0500 R03/1200 +SN +BLSN VV002 M09/M10 Q1001 TEMPO 0400 +SN +BLSN VV002 RMK QFE747 QWW060 MOD ICE MOD TURB S METAR USNN 020300Z 11008G13MPS 1100 R03/1200 SN +BLSN BKN004 OVC066 M 09/M10 Q1000 NOSIG RMK QFE745 QRD120 MOD ICE MOD TURB SC045 ... METAR USNN 021200Z 18009MPS P6000 -SHSN DRSN SCT017CB OVC066 M07/M09 Q0989 TEMPO 2000 SHSN RMK QFE736 MOD ICE MOD TURB SC042 METAR USNN 021300Z 16009MPS P6000 DRSN SCT016CB OVC066 08/M11 Q0989 NOSIG RMK QFE736 MOD ICE MOD TURB SC042 METAR USNN 021400Z 16008MPS P6000 DRSN SCT016CB OVC066 M08/M11 Q0989 NOSIG RMK QFE736 MOD ICE MOD TURB SC042
http://www.wunderground.com/history/airport/USNN/2007/2/2/DailyHistory.html
The monthly data shows the double peak:
http://www.wunderground.com/history/airport/USNN/2007/2/2/MonthlyHistory.html
I’m sure many more can be found, I invite readers to have a look for themselves by looking for such events at Weather Underground
It is not just Russia that has METAR reporting errors
Lest you think this a fault of Russia exclusively, it also happens in other northern hemisphere Arctic site and also in Antarctica.
Svalbard, Oct 2, 2008
METAR ENSB 020550Z 13012KT 6000 -SN FEW010 SCT015 BKN030 M04/M06 Q1013 TEMPO 4000 SN BKN012 METAR ENSB 020650Z 14013KT 9000 -SN FEW010 SCT018 BKN040 03/M06 Q1013 TEMPO 4000 SN BKN012 METAR ENSB 020750Z 15011KT 9999 -SN FEW015 SCT025 BKN040 M03/M07 Q1013 TEMPO 4000 SN BKN012
http://www.wunderground.com/history/airport/ENSB/2008/10/2/DailyHistory.html
Eureka, Northwest Territory, Canada March 3 2007
It hit 109.4 F (43C) there on March 3rd 2007 according to this METAR report. Eureka is the northernmost GHCN station remaining for Canada. It’s temperature gets interpolated into adjacent grid cells.
CWEU 031600Z 14004KT 15SM FEW065 BKN120 M43/M45 A2999 RMK ST1AS2 VIA YQD SLP150 CWEU 031700Z 15005KT 10SM FEW065 BKN012 43/46 A3000 RMK SF1AS1 VIA YQD SLP163 Decoded: 11:00 AM 109.4 °F 114.8 °F 100% 30.01 in 10.0 miles SSE 5.8 mph - Mostly Cloudy CWEU 031800Z 11003KT 15SM FEW050 FEW065 OVC130 M43/M46 A3001 RMK SF2SC1AS1 VIA YQD SLP164
http://www.wunderground.com/history/airport/CWEU/2007/3/3/DailyHistory.html
In these cases below for Antarctic stations Dome C and Nico, the METAR reports seem to have all sorts of format issues and I’m not even sure how where the error occurs, except that Weather Underground reports a spike just like we see in Russia.
Dome C station Dec 9, 2009
AAXX 0900/ 89828 46/// ///// 11255 36514 4//// 5//// 90010 AAXX 0901/ 89828 46/// ///// 10091 36514 4//// 5//// AAXX 09014 89828 46/// /1604 11225 36480 4//// 5//// 9014
http://www.wunderground.com/history/station/89828/2009/12/9/DailyHistory.html
Nico Station, University of Wisconsin Dec 9, 2009
AAXX 0920/ 89799 46/// ///// 11261 4//// 5//// 92030 AAXX 0920/ 89799 46/// ///// 11103 4//// 5//// 92040 AAXX 0921/ 89799 46/// ///// 11270 4//// 5////
http://www.wunderground.com/history/station/89799/2009/12/9/DailyHistory.html
Admusen Scott Station Jan 14th, 2003
Here’s generally properly formatted METAR data, but note where the technician coded the extra space, oops!
NZSP 131350Z GRID36007KT 9999 IC SCT020 BKN060 M31/ A2918 RMK SDG/HDG NZSP 131450Z GRID36007KT 9999 IC FEW010 FEW020 SCT035 SCT050 M3 1/ A2918 RM K SDG/HDG NZSP 131550Z GRID10008KT 9999 IC BCFG FEW010 SCT020 BKN050 M31/ A2919 RMK VIS E 2400 BCFG E SDG/HDG
http://www.wunderground.com/history/station/89009/2003/1/14/DailyHistory.html
And I’m sure there are many more METAR coding errors yet to be discovered. What you see above is just a sampling of a few likely candidates I looked at over a couple of hours.
Missing M’s – Instant Polar Amplification?
It has been said that the global warming signature will show up at the poles first. Polar Amplification is defined as:
“Polar amplification (greater temperature increases in the Arctic compared to the earth as a whole) is a result of the collective effect of these feedbacks and other processes.” It does not apply to the Antarctic, because the Southern Ocean acts as a heat sink. It is common to see it stated that “Climate models generally predict amplified warming in polar regions”, e.g. Doran et al. However, climate models predict amplified warming for the Arctic but only modest warming for Antarctica.
Interestingly, the METAR coding error has its greatest magnitude at the poles, becuase the differences in the missing minus sign become larger as the temperature grows colder. Eureka, NWT is a great example, going from -43°C to +43°C (-45.4°F to 109.4°F) with one missing “M”.
You wouldn’t notice METAR coding errors at the equator, because the temperature never gets below 0°C. Nobody would have to code it. In middle latitudes, you might see it happen, but it is much more seasonal and the difference is not that great.
For example:
M05/M08 to 05/M08 brings the temp from -5°C to +5°C, but in a place like Boston, Chicago, Denver, etc a plus 5C temperature could easily happen in any winter month a -5C temperature occurred. So the error slips into the noise of “weather”, likely never to be noticed. But it does bump up the temperature average a little bit for the month if uncorrected.
But in the Arctic and Antarctic, a missing M on a M20/M25 METAR report makes a 40°C difference when it becomes +20°C. And it doesn’t seem likely that we’d see a winter month in Siberia or Antarctica that would normally hit 20°C, so it does not get lost in the “weather” noise, but becomes a strong signal if uncorrected.
Confirmation bias, expecting to see polar amplification may be one reason why until now, nobody seems to have pointed it out. Plus, the organizations that present surface derived climate data, GISS, CRU, only seem to deal in monthly and yearly averages. Daily or hourly data is not presented that I am aware of, and so if errors occur at those time scales, they would not be noticed. Obviously GISS didn’t notice the recent Finland error, even though it was glaringly obvious once plotted.
With NASA GISS admitting that missing minus signs contributed to the hot anomaly over Finland in March, and with the many METAR coding error events I’ve demonstrated on opposite sides of the globe, it seems reasonable to conclude that our METAR data from cold places might very well be systemically corrupted with instances of coding errors.
The data shown between 4/13 and 4/15 were based on data downloaded on 4/12 and included some station reports from Finland in which the minus sign may have been dropped.
That darned missing M, or an extra space, or even writing “-” when you mean “M” (which is counterintuituve to basic math) all seem to have a factor in the human error contributing to data errors in our global surface temperature database. To determine just how much of a problem this is, a comprehensive bottom up review of all the data, from source to product is needed. This needs to start with NOAA/NCDC as they are ultimately responsible for data quality control.
It has been said that “humans cause global warming”. I think a more accurate statement would be “human error causes global warming”.
Note: In this post I’ve demonstrated the errors. In a later post, I hope to do some data analysis with the numbers provided to see how much effect these errors actually have. Of course anyone who wants to do this is welcome to leave links to graphics and tables. -Anthony
UERR 010330Z 22005G08MPS 9999 -SN 21/M23 Q1026 NOSIG RMK QFE738 24450245
UERR 010400Z 22005G08MPS 9999 -SN SCT100 OVC200 20/M22 Q1025 NOSIG RMK QFE737 24450245
UERR 010430Z 21005G08MPS 4000 -SN SCT100 OVC200 M20/M22 Q1024 NOSIG RMK QFE737 24450245
Sponsored IT training links:
Ultimate VCP-410 practice files formulated by top experts to help you pass 220-702 and 640-822 exam in easy and fast way.
Discover more from Watts Up With That?
Subscribe to get the latest posts sent to your email.
I haven’t read all the comments, so excuse me if this is a duplication.
Why not intensively examine a recent representative sample of the data that hasn’t been discarded, calculate the error rate and error effect, and correct the records for previous years by that rate?
For those who keep asking about the buoy data: From the sailwx website-
“Most of the information presented comes via a single, fallible pipeline with no redundancy. Data stream is processed automatically, without any human oversight, automatic quality control, or ‘sanity checking’.”
Btw- I’m only using this data to fill in the ‘pole hole’. And this data is easily corrected.
Re: Geoff Sherrington (Apr 18 04:26),
Not to be picky but Oz converted to decimal CURRENCY in 1966, then to decimal MEASUREMENTS in 1972.
On a serious note, I don’t think it matters that much what effect these errors have on “global anomolies”. The fact remains that this is yet more errors in an error riddled dataset.
To rely on this dataset for geopolitic decision making is folly.
@ur momisugly baahumbug (22:21:42) :
“Not to be picky but Oz converted to decimal CURRENCY in 1966, then to decimal MEASUREMENTS in 1972.
On a serious note, I don’t think it matters that much what effect these errors have on “global anomolies”. The fact remains that this is yet more errors in an error riddled dataset.
To rely on this dataset for geopolitic decision making is folly.”
Not to be picky, but I seem to remember currency and weights & measures going decimal at the same time. 1972 was when Imperial weights and measures became illegal.
_Jim (18:13:41) :
“aurbo (14:42:51) :
In fact … the machines including the Model-15 and Model-28 page printers
page printers? The surplus Model 28’s we used in the late 70’s all used continuous roll paper … maybe ‘page printer’ refers to 8 1/2″ wide paper versus paper tape …
Model 28 info page:
http://www.baudot.net/teletype”
You are right. The model-15 and model-28 teletype machines used paper rolls. I was descriminating between their 8½” wide 72 characters per line output and the WU type tape printers that were in current use back then for pasting up telegrams and for stock market tickers. In fact, the Western Electric manual describes the model-15 as a page printer see here.
The problem was that the models formated with standard character symbols had the – sign coded on the upper case “A” (Figs + “A”) key producing binary baudot code of 00010 while on those formated for wx symbols the lower case (Figures) 00010 coding was an up arrow used for one of the wind directions. To avoid confusion on networks where machines supporting both formats might have been used, the minus sign was dropped for the temperature group, the wind and cloud-cover symbols disappeared and when the METAR coding replaced the SA codes, the letter “M” replaced the minus sign for temps.
As stated in an earlier post, the whole problem began in an attempt to limit the number of characters per line bcecause of the slow transmission rate. This continues to cause problems today even though time is no longer an issue. For example, many errors occur in precipitation climatology because in the need to save transmission time, there is no positive zero. That is, if no precip has occurred in the past 6 or 12 or 24 hrs, no coded precip group is sent. If a precip group of all zeroes is sent, it connotes a trace! Thus you can’t tell the difference between no precip having occurred or the fact that there was precip but the precip group was missing or not observed. This not only occurs in the hourly observation transmissions but in the primary sequence in the synoptic code as well. I would say that precipitation data errors are much more frequent, mostly due to missing data, than you’ll find with temperatures. But, hey…that’s climate science.
Ok lets throw out the entire temperature record and instead just make geopolitical decision making based on the science. Which is that CO2 emissions are likely to significantly increase temperatures. All agreed? excellent.
One more comment of Leif’s observations.
As was indicated, Russians, like many other people in cold climes, usually leave out the “minus” or “plus” simply due to the fact that the difference between +15C and -15C is blindingly obvious to anyone around, and fairly easily deduced from a calendar.
HOWEVER, in the range around +/-5C, where the difference is not necessarily so immediately sensible, my experience was that people — at least in the part of Russia where I lived — would speak of, for example, 2 degrees мороза (of frost) or тепла (of warm).
At the same time, this was only a colloquialism. Anytime people were talking about temperatures AS DATA, they would use минус (minus).
Amazing how just a small oversight can turn -40°C into plus 40°C. You’d think this would be carefully monitored so as not to corrupt the results. Yet, no one seems to care. Now if the error worked the other way, i.e. exaggerated cooling, you can be sure they’d do everything to keep the error from happening.
As someone who has worked in shop floor data collection the metar ergonomics escape me. if it were designed to be morsed or sent in clear spoken using phonetic alphabet; i could understand the need to eliminate symbols, and a leading ‘M’ to denote sign would be reasonable, but other bits of the message use symbols..!!..???.
No-one reading a print out or a guage face marked as minus is unnerringly going to replace it with an ‘M’. Requires the operator to actively think about each reading rather than shunt it from eye to pen.
As an aside, when I did this for a living there were a few rules of thumb we used.
1) any manually entered data contains 1 error per 300 key presses. i.e even when the operator is putting the right bit of info into the correct entry field 1 in 300 characters will be incorrect (standard figure drawn from US DoD Generic Task Analysis studies). If you are lucky these will occur in non-key ,non-value, not-analysed fields.
2) Murphy’s Law and its principal corollary applies to data entry errors…..
3) The overall error rate on manual input is about 5% (Data entered in wrong fields, data not entered, wrong document type used etc etc etc includes errors due to point 1)
4) If a document isn’t entered within 15 minutes of its arrival you are on a not very protracted slope to it’s never being entered at all.
5)Errors not detected or trapped at or near to point of entry are never detected until they interfere with a critical process or are found under embarrassing circumstances by regulators and external auditors (Murphy’s Special theory for corporate record keeping……)
Hence the emphasis on Bar-Coded documents, EDI etc…
R. Gates (19:54:37) :
What difference or effect will this all make in the actual global temperature anomalies?
Inevitably, the Official-Answer ™ is going to be “The errors are insignifigant because of averaging and smoothing.”
Another log on the fire to do away with anomalies.
The use of Degrees Kelvin would have eliminated a whole host of errors, misinterpretations, and willful deceit regarding temperature data.
The use of anomalies is just sloppy practice.
Wren (14:26:11) :
If the recorders at airports are more likely to forget minus signs( whether indicated by “-” or “M”) than recorders not at airports, a move toward airport stations as a data base could introduce an upward bias in temperature records until there was no longer a move toward airport stations.
However, I doubt recorders at airports are any more likely to forget minus signs than other recorders. I think I am demonstrating a classic case of skepticism rather than a classic case of confirmation bias.
Seems you overlooked a detail there, Wren. METAR is the format for records from airports, nowhere else. It’s only airports that have the issue with using the letter M (or not, as the case may be) to signify a negative Celcius reading. For non-METAR formats, a ‘-‘ symbol may be retained to signify a negative reading; for METAR, it will definitely be dropped and show a positive Celcius reading.
Therefore, as there is a continuing trend towards using a higher proportion of airport data, so the issue becomes a growing one, creating an ongoing false trend.
Interesting stuff. There is of course no 100% adequate replacement for an experienced human brain in doing QA, but certainly the situation described above should be amenable to a computer algorithm to help identify the most obvious cases for further human review. Sure, the difference between -1C and 1C would be a toughie. But -10C and 10C, looking at historical records for which is more likely in that location on that date, ought to stick out nicely even to a computer to say “Hey, somebody take a look at this one. . . “.
A comment about methodology after a little further thought…
While the occurances of these errors will tend to increase the anomoly in the short term, the anomoly can’t really keep growing indefinitely due to these sorts of problems unless the rate of errors actually increases (or systemically starts getting applied only to colder and colder temp readings.)
So, while temperature may have risen in the short term as more and more data sets used METAR and introduced the potential problem, eventually it’s going to have to top out and the anomoly will stabilize at (Actual_Anomoly + Avg_Measurement_Error)
It makes one wonder if the recent stabilization and declines that the ClimateChange guys are lamenting is due to this levelling-off.
It also makes the conspiracy-minded wonder if this temperature-error-coding possibility was behind the decision to bias reporting stations largely toward airports… show a spike in temperature and then try to use that to scare the world fast before it levelled out. Someone could easily have run different data sets through the computer to see what the result would have been.
Theory of truth vs just outright lying–
http://alethonewsa.wordpress.com/2010/04/18/global-warming-fraud-in-worlds-most-highly-regarded-scientific-journal/
http://freestudents.blogspot.com/2007/05/grapes-of-math-global-warming-fraud.html
http://alethonewsa.wordpress.com/2010/04/18/global-warming-fraud-in-worlds-most-highly-regarded-scientific-journal/
On the issue of dropped zeroes:
some time back I questioned the data showing that January 2010 was the warmest January on record in the northern hemisphere.
I said at the time, could this be due to transcription error ? Because this finding sure goes against all reasonable expectations. A few dropped minus signs would go a long way to explaining this conclusion.
WOW !!
Even when there is an “honest” mistake, it is a warming bias !!!!!!!
Mr. Watts,
from your article I deduce that your “thesis” is : Minus sign errors in temp. data are responsible for the warming effect in the global trend.
For this to be valid you must have a positive answer to following 3 questions:
1. Is the percentage of the total local temp. data stations which have this error, statistically significant?
2. Is this error systematic, i.e, it always appears in the same stations at the same time?
3. This error is not presently taken into account in statistical processing of temp. data?
Unless you are able to answer to this questions, your investigation amounts to the discussion of a real but statistically insignificant error.
Cheers
Since these human errors are detectable in polar regions, it should be possible to use the polar sites to determine the human error rate, then project the bias effect at temperate latitudes.
The bias estimation problem is greatly simplified by the fact that the error should almost exclusively go in one direction. As an earlier commenter noted, people aren’t likely to be entering M’s by mistake:
Since these human errors are detectable in polar regions, it should be possible to use the polar sites to determine the human error rate, then project the bias effect at temperate latitudes.
The bias estimation problem is greatly simplified by the fact that the error should almost exclusively go in one direction. As an earlier commenter noted, people aren’t likely to be entering M’s by mistake:
If there are “superfluous M” errors being made then it should be possible to estimate the error rate by selecting data from temperate-region airports at times when the average temperature is high enough that a superfluous M would stand out. (It wouldn’t do to estimate the superfluous M error rate using equatorial airports, where they would always stand out, because the operators would not be sometimes called upon to enter “M”s, making the mistake less likely at the equator.)
As Anthony points out, it is easy to see how the missing-M mistake could be made. In contrast, the most likely reason a person would enter an M by mistake would be if he actually misinterprets the positive temperature reading as a negative one (possible if relaying a celsius reading perhaps, or reading a digital readout in a hurry). That is very different from omitting an M, or typing a minus sign instead, and should be much less likely.
It seems that the resulting warm bias could be quite large. Is this the explanation for the reported high Siberian temperatures for the winter of 08-09? If the error is substantial, developing an estimate of its size would be important.
ANTHONY or MODERATORS.
Did you see this offer of assistance at Glenn McKee (12:34:07) : as I did not see any response.
Wren,
You need to understand the way that this operates. Missing M’s in the data are not a systematic error, they are a random error, as shown by Anthony in his post.
Let’s say that a station in Siberia has a relatively steady reading of -35C (or M35) for the entire month of February. Now, let’s say that on a particularly cold day (about -43C for the entire day, just for example) for 8 hours the M is missing in the dataset, so for 8 hours the temperature is reported as 43C instead of -43C. This is a +86C error for 8 hours. Instead of a daily average of -43C for that day, the daily average becomes -14.3C for that day, for a difference of nearly 29C for the daily average due to this error. Even when this erroneous data gets averaged in to an entire month and we assume that the other 29 days had a correct average which was stipulated to be -35C, that monthly average suddenly becomes -34.3C, which is a +0.7C error in the monthly average (for this calculation I assumed a 30 day month, 29 days with an average of -35C, and one day with an erroneous average of -14.3C).
I made the math much simpler than I am sure it really is, but just using basic math it is possible to show that if 8 hours per month of data at a really cold station are missing the M, it could easily throw off the monthly average by anywhere from +0.5 to +1.0C.
Since we are talking about a “global temperature anomoly” on the order of +0.5C which is supposed to be a “major cause for alarm” and we “need” to spend trillions of dollars to “do something about it”, it is critically important to note that if cold/polar stations even have 8 corrupted hours of data over an entire month of 720 hours of data, the monthly average magnitude of that error might well be > +0.5C
This is why this finding is so critical.
Alec Rawls (05:55:48) :
Alec Rawls (05:56:38) :
> As Anthony points out, it is easy to see how the missing-M mistake could be made.
And note that Alec has just pointed out how easy it can be to leave the ‘/’ out of </blockquote> 🙂
@Skeptic
The error DOES NOT HAVE TO BE SYSTEMATIC in order to be statistically significant.
See my post above.
It only takes EIGHT HOURS of bad data from a Siberian or Polar site in an ENTIRE MONTH to throw off the monthly average by +0.7 C.
Also, no, this error would NOT be currently taken into account in the current statistical processing of the data. The whole reason for this post was real-life erroneous Finland data that stood out like a sore thumb. Only AFTER it was mentioned to the “powers that be” was it actually admitted to be in error and subsequently corrected.
If outside sources had not brought it to their attention, the erroneous data would have remained in the databases.
It does not matter if the error is random or systematic, what matters is the frequency and location of the error. All it takes is 8 hours of bad data out of 720 hours of monthly data to throw the average off by quite a lot at a cold station.
@MattB
Ok lets throw out the entire temperature record and instead just make geopolitical decision making based on the science. Which is that CO2 emissions are likely to INSIGNIFICANTLY increase temperatures. All agreed? excellent.
FTFY