Saturday, October 1, 2016

Climate Observations - The Human Element

First off, I should quickly mention a significant temperature record yesterday at Barrow: the high temperature of 48°F was not only a record for the date, but the warmest on record for so late in the season.  Previously the latest date with such warmth was September 24, 1995, when it reached 53°F.  The late September 1995 heat wave was actually more unusual, though, as it reached 62°F in Barrow on the 22nd (and 78°F in Fairbanks the previous day).

And now for a different topic.  I'm sure many readers are aware that daily maximum and minimum temperatures are automatically recorded by the ASOS observing platform at airports across Alaska, and these measurements generally go into the books unaltered as the "official" climate observations.  This isn't always true, however; one reason is that sometimes the ASOS report is in error.  There happened to be an example of this in Fairbanks this week, when the ASOS reported a 24-hour (midnight-to-midnight) high temperature of 59°F on Thursday, but in reality the high was only 51°F.  Oddly the 6-hourly maximum temperatures reported by the ASOS (at 0, 6, 12, and 18 UTC) were correct for Thursday, but somehow the 24-hour maximum wasn't.  The error was caught by NWS personnel and the climate record already shows the correct number.

Looking back at the history of ASOS measurements in Fairbanks, it seems that manual adjustments to the ASOS data are not common, but neither are they rare.  The charts below show the daily differences between the final "official" numbers and the ASOS reports for June through August.  Curiously it seems that the daily low temperature is most often adjusted down by a single degree Fahrenheit.  This didn't happen in summer 2016, but in August 2015 it occurred 4 times between August 10 and August 19.

It's strange and a bit unsettling to find that the ASOS temperature reports are apparently erroneous with some frequency, but in Fairbanks the errors are not large or frequent enough to cause significant differences in the long-term temperature averages; and in any case presumably the vast majority of the errors are noticed and corrected.  The problem is intriguing, but it doesn't make a big difference in the grand scheme of things.

However, the same is not true of the data situation at Bettles - see the charts below.  The maximum temperature differences show rather more frequent and significant changes, and the minimum chart is frankly shocking, with very frequent and mostly downward adjustments to the ASOS data, especially in recent years.  Remarkably, in June-August of this year, the daily low temperatures were adjusted downward on 32 of 92 days, by an average of 3.3°F.  In summer 2014, adjustments averaging 2.9°F were performed on 70 of 92 days.  The question arises immediately as to whether this is an appropriate fix to a very bad ASOS problem, or whether the adjustments themselves are the problem.

The overall impact of the low temperature adjustments at Bettles is significant, as demonstrated in the chart below.  In summer 2014, when the adjustments were greatest, the seasonal mean of daily minimum temperatures changed from 46.4°F to 44.2°F; this difference is comparable to the typical magnitude of changes from year to year.

So what is going on at Bettles?  Well, I've learned that the FAA contract observer believes that the ASOS minimum temperatures are generally too high (at least in summer) and regularly adjusts them down, purportedly based on other thermometers.  The adjustment is different from day to day and is often zero.  From a scientific standpoint, this is difficult to accept, because a systematic error at the ASOS thermometer would show up all the time, not some of the time, and it would presumably affect daily maximum temperatures as well.  NWS personnel have inspected the ASOS temperatures repeatedly on-site and have not uncovered a problem.  Unfortunately there is nothing that can be done about the issue, because the FAA manages the observing program, and the accuracy of climate measurements are not a high priority for that agency.

Additional evidence indicating that the Bettles low temperature data have been corrupted can be found in a comparison of the official (airport) temperatures to the nearby SnoTel site at Bettles - see below.  The SnoTel site runs much colder for overnight lows in summer, but the difference is consistent and the seasonal means are highly correlated from year to year.  Notice, however, the trend towards smaller differences over time, as the official low temperatures have been adjusted downwards more often in recent years, and especially in 2014 and 2016.  Remarkably, the differences between the two sites were smaller in every year from 2010-2016 than in any year from 2003-2009; there isn't much chance that this could happen at random, and accordingly the trend in the differences is statistically significant.

Here's what the chart would look like if the ASOS temperatures were left alone.  There is still a slight trend towards smaller differences over time, but it's not statistically significant.

For additional context on ASOS errors, the daily temperature adjustments are shown below for McGrath.  The situation there looks a lot more like Fairbanks, which suggests that this kind of frequency and magnitude of errors are typical.  In contrast, the data adjustments at Bettles are highly atypical and (in my view) clearly erroneous; and this is quite unfortunate for the integrity of the long-term climate record in Alaska.


  1. "A man with a watch knows what time it is. A man with two watches is never sure."
    -- Segal's Law

    Maybe that applies to some observers of thermometers as well...especially in this Age of Warmth.


  2. A few thoughts.

    Systematic errors need not be linear. Lower temps could have bigger errors. But I'm not sure the data shows this.

    How do we know that ASOS is calibrated right? How about other thermometer? Two thermometers 50 ft apart can legitimately be 5 degrees off.

    Where are the quality controls? Billions, if not trillions, of dollars is relying on accurate measurements.

    How widespread is corrections like these in the rest of the ASOS network?

    This is all really unsettling.

    1. Eric,

      Good point, systematic errors are not constant errors, but a miscalibrated thermometer will be more or less wrong most or all the time.

      I know the NWS has checked the Bettles ASOS temperatures but don't know if they have the final say on whether recalibration or replacement is required. I have no information on other thermometers used in the Bettles situation.

      Good question about Q/C. The ASOS network was not designed for climate monitoring, but we expect it to perform for that purpose. Hence the USCRN project.

      I'll probably do some follow-up on other ASOS stations and also try to establish what fraction of the adjustments are definitely justified from other data.

  3. NWS ASOS (including Bettles) are on a routine maintain schedule, which includes "spec checks" and periodic hardware replacement. There are a couple of weak links in ASOS temp sensing: 1) there is only one sensor and 2) by far the weakest is the aspiration system. ASOS have what I call the "mushroom cap" type radiation shield, which is very good for minimizing solar radiation exposure, but that system has NO natural ventilation. Aspiration has to be done mechanically: a fan system drawn air in from below the "cap" and across the sensor. The air intake can and does get clogged to variable, but sometimes extreme extent, by insects, webs, birch seeds, willow fluff, hoar frost, etc. The fans themselves also seem (to me) to have a very high failure rate. The results of aspirator fan failure are immediate and obvious: temps jump to 10 to 15 degrees F higher than actual. Sometimes the temp will seem to jump up and down by this amount: that occurs when the fan is intermittently failing (I've been told this is often a bearing problem). None of this describes the changes being made to Bettles ASOS.

    As Eric notes, temperatures can varying significantly over short distance, but as a matter of practice we want the temperature to be from one point. What's happening at Bettles is that we have a random collection of temperatures that, in the best case scenario, are from two distinct points. It would be like ransoming inserting the Fairbanks AP#2 low temps into the ASOS record.

    One question for Richard: what are you using for the Fairbanks ASOS mins? One degree changes seem very unusual, and I don't recall that from 2014.


    1. Rick, I'm using the 24-hour max/min reported at 08:53 UTC. A recent example of the 1 degree difference in Fairbanks is September 15 of this year: ACIS/GHCN shows 40F, but ASOS reported a low of 41F at 0853 on the 16th.

  4. Surely NWS or FAA staff know, or are aware of, local WX observers (past and perhaps present) in Bettles, AK.

    For example:

    The Bettles Homepage ( lists the FAA WX. That may not be current. Various Federal land stewards maintain facilities there, and might know who does what for whom.

    If I were curious I'd attempt to speak with the observer(s) to obtain an explanation for the adjustments noted.


    1. My point is that most folks will offer a reason for their choice...if the exchange is friendly and non-confrontational. An explanation as to why the quality of the data is important helps.

      There's likely a reason as to why the Bettles ASOS temps were adjusted by the observer(s). Learning that is important to the discussion.


  5. The 24hr max/min from the METAR is not necessarily the "official" high and low for the station for that day. The 24hr day for PAFA runs from 9Z-9Z. The ASOS reports a 24hr max/min at 0853Z. There's 7 more minutes of time for the low temperature to fall further. This is why you see many days with a minimum temp 1F lower than the ASOS observation because of a late midnight low temperature.

    1. Thanks, that makes sense. I had assumed 0853-0853 was "good enough" but suspected that might not be true. Now I need to find an archive of the original ASOS daily summary messages...

    2. While 7 minutes of additional cooling could be significant, I don't think that's always the case.

      Since Bettles is above the arctic circle, daily temp variability is lowest during the summer and winter and highest during the spring and fall. In January, the daily temp might move five degrees the whole day - if you are lucky. Such slow movements in temp lean towards temps staying the same for a long time.

      Also, the amount of discrepancies in Bettles compared to Fairbanks would tell me that something is different on how things are treated.