Wednesday, December 11, 2013

Autumn Forecast Verification

Regular readers may recall that way back in August we discussed some long-range analog forecast indicators for September through November, based (only) on the warm, dry summer that occurred in interior Alaska.  It's always fun to go back and check long-range forecasts to see if there was any semblance to the actual outcome; so here we go.

Listed below are the main (though admittedly tentative) conclusions of the analog investigation, along with the corresponding outcome:

  • Dry September in Fairbanks: incorrect  1.74" vs 1.08" 1981-2010 median precipitation.  Three-quarters of an inch of rain fell on one day, September 11.
  • First 25 °F frost in Fairbanks earlier than normal: correct  Sep 20 vs Sep 26 1930-2012 median
  • First 0.5" or greater snowfall in Fairbanks earlier than normal: correct  Sep 18 vs Oct 6 normal
  • Wet September in Barrow: correct  1.29" vs 0.59" normal
  • Warm October in Barrow, Fairbanks, and Anchorage: all three correct, much warmer than normal
  • Warm November in Anchorage: correct but only barely, 21.9 °F vs 21.3 °F normal

The discussion also showed maps of the 500 mb height and sea-level pressure patterns that have historically been observed in September following warm, dry summers in Fairbanks.  Below is the observed September 500 mb height anomaly: this agreed very nicely with the forecast map, with one of the main features being low heights over the Chukchi Sea.

Finally, here's the observed sea-level pressure pattern from September; this doesn't show the same degree of correspondence with the analog forecast map, but nevertheless below-normal pressure occurred as expected over eastern Alaska and western Canada.
All in all, the forecast exercise was fairly successful and illustrates that simple statistical techniques can - if you're lucky - yield limited long-range predictability.  In the next week or so I'll put together another analog forecast based on the very wet, warm year that Barrow has observed.


  1. Richard, very interesting. My only quibble would be with the first occurrences of a 25 degree day and the first 0.5" snowfall at Fairbanks. Indeed, they were early this year, but the 2nd, 3rd, and 4th events were (I believe) very late. The first occurrences were isolated anomalies in an otherwise different large-scale pattern.

  2. Brian, I agree with you in one sense - the September anomalies were not representative of the autumn season and were quickly erased from memory by the later warmth. But I would argue that the large-scale pattern was quite different for September as a whole; the big ridge that dominated October, and has kept coming back since, was not evident back then. At the time, the cool spell in September was moderately significant, with the week ending Sep 24 averaging 1.35 SD cooler than normal, and the first sub-freezing daily mean (Sep 20) being the 8th earliest since 1930. I think the usefulness of the analogs is that they hinted at this evolution: cool September, warm October and November.