Fort Smith, Baker Lake, Canada Redux

OK, on the “Alaska Regime Change” posting here:

we ran into some “odd things” in some Canadian data that lead to this posting:

In comments, Boballab has raised objections to the data in Wolfram Alpha for Baker Lake, Canada. On two major counts. First, that the high record of 91 F or 92F was pretty darned hot for a place near the arctic circle. (Though Wunderground shows the same hot day in their data). Then that the “shape” of the graphs were quite different from W.A. but matched for Environment Canada vs GISS. There was also a minor point that the data in W.A. has a “dropout” in the ’70s that the GISS and Environment Canada data do not have.

OK, I’ve been off on my pursuit of geese….

But first, a point about GIStemp:

Remember that it will fill in missing data, sometimes from great distances and over reasonably sized spans of time, from other stations via “The Reference Station Method”. What you see in a GIStemp graph is NOT the original data. It has been highly processed and “filled in”. It has also had a “Q.A.” step done that suppresses extreme low temperatures via replacing them with an average of “nearby” airports that can be quite far away.

I do not know the E.C. methods, but if they are following GISS (and Canada seems to have a history of that, and of being more “advocates” for the Global Warming agenda); then they may well be doing similar “post processing” in a Post-Normal Science kind of way…. A speculative “Dig Here!”

Fort Smith

First off, I went to GISS and clicked on the map for about where Baker Lake is located to get the list of stations

Then I picked another one “nearby” that had a long temperature history. There are many ‘fragmentary’ stations, so the closest one that had a very long record looked like Fort Smith. ( I tried “Cambridge Bay” in Wolfram Alpha but could not get it to stop sending me off to other Cambridges… probably THE most frustrating thing at the site… I’ve not figured out the incantation to force it to a specific station.)

Then I explored the Fort Smith data at other locations as well.

At Wunderground, the record for Fort Smith is given as 95 F (not out of line with the Baker Lake record). The data at Wunderground has a similar “dropout dip” in about 1966 (while GISS shows no such drop out). Checking for data at Wunderground for Fort Smith Canada and 1970 gave a ‘null record’, so they, too, are missing data then.

For comparison, here are the GISS graphs for Fort Smith and from Wolfram Alpha along with the metadata block for both the Wolfram Alpha Fort Smith and Baker Lake pages. The purpose? To show that there is some sort of “generic issue” and not just a data entry faux pas on Baker Lake. (And to cast suspicion that the “generic issue” is the filling in, adjusting, homogenizing, and who knows what else that is in the GIStemp product, so if E.C. matches it, they are likely doing something similar…)

Canada Metadata WA Fort Smith

Canada Metadata WA Fort Smith

Canada Metadata WA Baker Lake Airport

Canada Metadata WA Baker Lake Airport

Canada Fort Smith all data Wolfram Alpha

Canada Fort Smith all data Wolfram Alpha

Canada Fort Smith GISS

Canada Fort Smith GISS

Baker Lake WA May 2011

Baker Lake WA May 2011

Baker Lake NASA GISS May 2011

Baker Lake NASA GISS May 2011

So to me, it’s looking like a generic class of error, and not a “one off” bad data entry. Further, it’s looking like a “post processing” fill in the blanks and homogenize result, not a ‘wrong identity’ (for the simple reason that we KNOW GIStemp has done that, it’s in the code; and two matching random errors of the same type? That would be very unlikely…). I’ll leave it for others to try other sites and see if they can surface any more of these (where the base data looks to be “not there” but where E.C. and GISS both color it in, and find odd ‘trends’ to it). I leave it to others for the simple reason that I’m a bit too swamped at the moment to follow it up properly.

In Conclusion

My “working hypothesis” is that Environment Canada has some sort of “fill in the missing data and homogenize” process similar to GIStemp. That Wunderground and Wolfram Alpha have less “improved” and more “undercooked” data (not quite raw…) That there is a significant ‘data drop out’ in the ’70s for that region of Canada for some stations, and that dropout is ‘colored in’ by the processing at E.C. and GIStemp. ( I know that GIStemp will do so as it’s in the code.)

That, then, likely results in the overall change of the nature of the graphs between the two sets. E.C. and GISS on one side; Wolfram Alpha on the other. Wunderground with the data, but declining to make the long term graph, that would match Wolfram Alpha.

It is possible that there is some other reasonable hypothesis (such as both Wunderground and W.A. getting a load of bogus data from the same place?) that could cause this as a legitimate bit of dirty data. That, too, would stand checking.

Yes, this could use a lot more detailed documentation (but I’m way behind on a lot of things already so) that will need to fall to others to provide. A very large “Dig Here!!”…

Subscribe to feed

About E.M.Smith

A technical managerial sort interested in things from Stonehenge to computer science. My present "hot buttons' are the mythology of Climate Change and ancient metrology; but things change...
This entry was posted in AGW GIStemp Specific, AGW Science and Background and tagged , , . Bookmark the permalink.

14 Responses to Fort Smith, Baker Lake, Canada Redux

  1. E.M.Smith says:

    Just finished looking directly at the GHCN.v2 data. It does have data for Fort Smith in the ’70s. (It also has TWO “Fort Smith” records for Canada in the “inventory” file, one is at -112 degrees, the other at -111.95 degrees. Oddly, that second one only has data from 1913 to 1929… while the former as 7 different “Modification History” or “Dupicate Number” flags…

    Very strange…

    1970 – 1975 (plus some) is clearly “missing” in the above graph, yet in the GHCN, it has entries for “mod flags”: 0, 1, 2, 3, and 5.

    It almost looks like only some of the “duplicate number” records were used (but the end points are not an exact match) or like the “combining” process had some errors for W.A. and Wunderground.

    Then again, as both have relativly unknown processes, who knows what they’ve got…

    OR if the GHCN has been “improved” compared to the actual raw data available…


    If someone would just come up with a CLEAN and UNADULTERATED temperature data set, that alone would be a great improvement…

  2. Level_Head says:

    The underlying data is a long way from clean and unadulterated, even before the adjustments from dutiful to dubious are applied.

    There’s certainly some confusion going on here, though.

    ===|==============/ Level Head

  3. boballab says:


    Environment Canada when they do infill they tell you they have don so by denoting that the Mean is an estimate.

    Here is an example where EC Estimated the Monthly mean for Baker Lake (In the Year 2010 data):

    Now when you go back to 1969, as an example, where W.A. has the “missing data” you can see that there is no Estimate made, nor an M for missing data. That year they had plenty of data.

    As to the high, that didn’t seem right but when I went through the EC data it had an extreme high of 91.8° F in 1991:

    I also have been looking through the GHCN V2 Raw data file and not seeing any large drop out of data during the time frame of 1965-1977ish

    on a related note I left this in tips a couple of days ago and I don’t know if you saw it but NCDC has changed from GHCN V2 to V3 starting May 2nd:

    Effective May 2, 2011, the Global Historical Climatology Network-Monthly (GHCN-M) version 3 dataset of monthly mean temperature will replace GHCN-M version 2 as the dataset for operational climate monitoring activities. Although GHCN-M version 2 will continue to be updated with recent observations until June 30, 2011, users are encouraged to begin using GHCN-M version 3. Please see: Summary of Recent Changes in the GHCN-M Temperature Dataset and Merged Land-Ocean Surface Temperature Analyses.

    If wonder if we will see a repeat of the USHCN v1 to v2 fiasco with GISS.

  4. What?

    Fort Smith NWT and Baker Lake Nanuvat are several hundred miles apart with very different geography.

    (Minor detail, beware that weather records for some settlements may have come from various sources and slightly different locations (such as for Fort Smith HBC trading post, seaplane landing area, and land airport now there).

  5. Making it even clearer, Fort Smith is in tree and lake country (I’ve been there), Baker Lake may be in tundra.

  6. E.M.Smith says:

    @Keith Sketchley:

    The accepted behaviour in things like GIStemp is based on “The Reference Station Method” that is supported by peer reviewed publications that assert up to 1200 km is “acceptable” for comparison and fill-in. (I’m in the camp that says “that’s just wrong”, but that’s what is ‘accepted’…)

    The point behind a swap to Fort Smith was more of a check for “similar behaviour” to see if there was a generic issue or something mroe subtile.

    What was found, IMHO, is that Wolfram Alpha is a bit poor on how they handle “data dropouts” and just run with what they have, even if it is only, oh, summer data… So the comparison did do the desired job. It pointed at what was the issue.

    And per them being all that different:

    When a Canadian Cold Front runs down this way, you will get the same “plunge” of temperatures all the way from California to Southern Texas to Florida. Yes, the micro-climates matter (both to the day to day particulars and to the volatility / range) but the macro effects are the same over massive areas. It is valid to compare the pattern at Fort Smith to that at Baker Lake to look for clues as to what causes it and why. (as was done… and as showed the way to ‘data dropouts’ as the ‘issue’ most probable.)

  7. What threw me is your claim that Fort Smith is “nearby” Baker Lake.

    Looking at a diverse set of stations may be a good approach to looking for generic analysis problems, which I understand is what you tried to do, as would looking at stations that are truly close. (Try Baker Lake versus Rankin Inlet for the latter, but beware Rankin Inlet is on the coast so won’t see the temperature swings of inland locations.

    You can ponder how wide an area would be affected by variations over a very large area (which the Canadian Arctic is) or a smaller area (such as that encompassing Baker Lake and Rankin Inlet). But I wouldn’t expect Fort Smith to have as strong an effect from “Arctic” (polar) phenomena – it is well south, on the AB border.)

  8. Additional information about Fort Smith, regarding variation in locations shown in temperature data databases, other than rounding (such as -111.95 up to 112):

    Such settlements often fluctuate in their economic activity over time, depending on administrative (government) activity, mineral extraction activity, and special projects.

    The 1913-1929 data may reflect designation in 1911 as the administrative centre for a very large area, with building of hospitals and schools occurring during the next few years, and the establishment of an airport in 1928. But I don’t know where that airport was – my impression is that the land airport I visited is significantly above the river where bush planes would land in the very old days (at least in summer, though there probably are small lakes in the general area).

    As an additional example of fluctuation, note that during WW II the US Army used Fort Smith as a transfer point to Norman Wells. (To support building the Canoil pipeline into the Yukon from where fuel could be transported into Alaska, as alternative to ocean transport that might be vulnerable to enemy attack.)

    As well, I wouldn’t be surprised if records go missing, including due to fire. Though the Hudson’s Bay Company, who had trading posts in such places as Fort Smith, may have had central records – Tim Ball analyzed HBC weather records early in his career as a geographer (long before there was a “climatologist” sheepskin).

  9. E.M.Smith says:

    @Keith Sketchley:

    I put “nearby” in quotes as it is a reference to the Reference Station Method claim that a 1200 km away station is near enough to be the same pattern. I, too, think it’s a bit bogus (as I’ve frequently said and shown with places like San Francisco vs Lodi where they have an inverse relationship much of the time. HOT in Lodi causes air to rise, sucking in a cooling fog over San Francisco…)

    Yet there is a kernal of truth to it. There are large patterns that hold much of the time. (i.e. Canada Express reaching to Texas).

    So I’m balancing those two truthes and looking for clue about Wolfram Alpha data. (That clue was found, IMHO, as a poor handling of years with many missing months. W.A. just makes an anual average, even if it only has summer to work with… thus some of the ‘way high’ early years above.)

    All the rest of your historical points are very interesting, and useful to know, but I think it is behaviour of making an ‘anual average’ from whatever sparse months you have that causes the W.A. graphs to be divergent.

  10. I see this as yet another example of the difficulty in collecting and analyzing temperature data.

    I am aghast at the extrapolation over wide distances that are in data bases like NASA’s. While people use variation from a baseline value to reduce the effect of different locations, I am very skeptical that is near enough to make data accurate.

    BTW, two extreme examples of causes of local climate changes:
    – extensive clearing of land for farming, how much does that affect temperatures?
    – reservoirs behind hydro-electric dams. Anyone interested in that might check Dawson Creek and Fort St. John which are roughly downwind from the huge reservoir behind the Peace River dam, built circa mid-60s. (Hudson Hope is very close to it but the airport there was not manned last I checked. Moberly Lake is to the south and may have been an HBC post from way earlier.)

  11. boballab says:


    The effect of Land Use change can best be observed in the peer review papers by Dr. Christy of the University of Alabama, Huntsville (UAH) and Dr. Pielke Sr. of the University of Colorado.

    They did research on the Central Valley of California and how the introduction of irrigation and farming to the desert area of the Central Valley has changed the temperature profile of that area.

    Dr. Christy is also one of the first Climatologists to show that the loss of the “snows of Kilimanjaro” was due to land use change at the base of the mountain and not rising temperatures.

  12. Thanks “boballab”.

    The land clearing theory for Kilimanjaro “makes sense”, but beware that:
    > another theory is that the cause is further upwind: evaporation in the Atlantic ocean
    > Kilimanjaro’s ice cap is now growing, some of the scientists who earlier raised alarm are admitting they did not do a good job of measuring and investigating.

    That’s a major problem with the climate subject – incomplete studies and people jumping to conclusions from what may just be “curious coincidences”. ;-)

  13. Pingback: Will The Real Blue Hawaii Please Stand Up! | Musings from the Chiefio

Comments are closed.