The Added Value of Analogs

[donsutherland1]

Briefly, and this is key, analogs are not used to make date specific, temperature-specific, or precipitation-specific predictions. They are intended to offer broad guidance e.g., how patterns might unfold. Such guidance can be for relatively short periods of time or longer periods of time. With regard to the mid-January pattern change, a strong cluster of analog seasons suggested that such a change had occurred in the past (70%). Thus, given the strength of analogs, it was not unreasonable to argue that a pattern change would occur around mid-January (give or take a few days). It did.


Analogs have also served very well in terms of offering guidance for 2004 hurricane season (active, above normal landfalling activity, and the likelihood of a major hurricane making U.S. landfall) and the current pattern now in place. The following is a specific example taken from an issue that arose here for illustrative purposes.

The issue concerned whether the January 24-28 or February 21-25 periods would be more likely to see a plowable snow in Harrisburg, Pennsylvania. Excerpts from the January 10, 2005 post that addressed this issue follow:

In order to compensate for slight timing differences between possible analogs (they are used for guidance not specific point forecasts), I examined the January 21-31 and February 18-28 periods.

I used 3 samples for Harrisburgh, PA:

• Test 1: Seasons similar to the current one with regard to the December ENSO regional anomalies: 15 seasons
• Test 2: Seasons similar to the current one with regard to both the December and October-December ENSO regional anomalies: 10 seasons
• Test 3: Seasons similar to the current one with regard to the December and October-December ENSO regional anomalies that also saw a pattern change to colder for the 2nd half of January: 7 seasons

The data revealed the following:

Test 1:
January 21-31: Average snowfall: 4.2"; 23 events (15 seasons); Seasons with 3" or more snowfall: 5/15 (33%)
February 18-28: Average snowfall: 3.4"; 20 events (15 seasons); Seasons with 3" or more snowfall: 7 (47%)

Advantage to January 21-31: 23.5% more snowfall; 15% more events
Advantage to February 18-28: 40% more seasons with 3" or more snowfall

Test 2:
January 21-31: Average snowfall: 5.6"; 16 events (10 seasons); Seasons with 3" or more snowfall: 3/10 (30%)
February 18-28: Average snowfall: 4.2"; 13 events (10 seasons); Seasons with 3" or more snowfall: 6/10 (60%)

Advantage to January 21-31: 33.3% more snowfall; 23.1% more events
Advantage to February 18-28: 100% more seasons with 3" or more snowfall

Test 3:
January 21-31: Average snowfall: 7.3"; 14 events (7 seasons); 3/7 (42%) seasons with 3" or more snowfall
February 18-28: Average snowfall: 4.6"; 9 events (7 seasons); 5/7 (71%) seasons with 3" or more snowfall

Advantage to January 21-31: 58.7% more snowfall; 55.6% more events
Advantage to February 18-28: 67% more seasons with 3" or more snowfall...

Finally, it should be noted that analogs are far from precise predictors. They do offer some guidance and it appears that they reasonably support the February 21-25 period as having the better opportunity for a plowable snow.

Actual Outcomes:
• January 24-28: No plowable snowfalls; 1.0" total snowfall (1/24)
• February 21-25: 8" of snow fell on February 24; 8.8" from February 21-24 (0.8" on 2/21; 8.0" on 2/24).

[donsutherland1]

The 2/24-25 snowstorm brought New York City's monthly snowfall past 10" and Boston's past 15". This was consistent with the following historical information previously noted:

• In 5/6 (83%) of seasons that saw Boston receive 30” or more snow in January, February saw 15” or more fall there. The average was 20.4”. Since regular recordkeeping began in 1891-92, 36/113 (32%) of seasons saw February receive 15” or more snow and the historic average snowfall for February came to 12.7”.

• In 8/15 (53%) of seasons that saw New York City receive 15” or more snow in January, February saw 10” or more fall there. The average was 11.1”. Since regular recordkeeping began in 1869-70, 46/135 (34%) of seasons saw February receive 10” or more snow and the historic average snowfall for February came to 8.5”.