Thursday, October 4, 2018

Winter '17 / '18 - Arctic Oscillation (AO) Analog Verification

The analog forecasting technique seeks similarities to the AO state in the run-up to the coming winter with AO run-up states of winters past.  Presented here is the verification of AO analogs for the '17 /'18 winter.

The NEWxSFC method ranks analog years by their the sum of square errors (SSE) statistic.
Lower SSE errors ==> stronger analog

Constraining the number of analog winters for analysis to five is arbitrary.

Arctic Oscillation (AO) Index Analog Forecast Verification
In the run-up to Winter '17 / '18 ... '73 / '74 was the leading analog followed by '08 / '09 ... '54 / '55 ... '95 / '96 ... and '75 / '76.  Winter '17 / '18 AO started weakly negative and remained weakly negative until meteorological winter's end.  MAR-18 AO (not shown) crashed to -0.941.
 
A qualitative assessment of the forecast's accuracy would rate all analogs except '73 / '74 as 'poor' ... IOW ... useless.
 
OTOH ... analog #1 mimicked the observed behavior of the AO associated with a weak La Nina; whereas ... .  strong La Nina conditions prevailed during the winters of '73 / '74.
 
CONCLUSION:  The analog forecasting technique provided useful guidance for Winter '17 / '18.
 
---
An alternative forecasting technique looks at the AO/s 'sign' (i.e., positive or negative) for any calendar-year's month preceding the pending winter as a potential leading indicator of the AO's sign for upcoming D-J-F period.

Results from a chi-square 'test for independence' infers a statistically significant relationship ... at the 95% confidence level and a p-value < 0.05 ... between NOV's AO sign and AO's sign of the upcoming D-J-F period.

IOW ... if NOV's AO is negative (positive) ... then the average AO state during the upcoming winter will also be negative (positive); although the classification model is stronger ... i.e., lower false alarm rate ... for the predictor's month with negative signs than positive.  This is opposite of the relationship found for the NAO.


BOTTOM LINE:  If NOV's AO is negative ... chances are good the AO state will average negative during the D-J-F period.
 
NOV '17 AO/s sign was negative (-0.078).  The 2x2 contingency technique predicted correctly AO state for Winter '17 / '18 would average less than zero.  AO for the D-J-F period averaged (0-.076).

No comments: