332 Martin Fischer
PREDICTIVE SKILL MEASURED BY NINO-3 sst ANOMALIES
72-95
72-79
80-87
88-95
1.0 _ - - - - . r-----, r - - - - - , r-----,
Z 0.6
I 02 t----'t----f1t--+---+1t--+---+lI---\---:-I
-.2
1.6 . . - - - - . , ,..---.,...., r--...... ~-r., ,..---.....,
,....,
U
OI)
.,
:e
Il:::
O
Il:::
Il:::
~
<1)
~
1.2
0.8
0.4
48121620
LEAD TIME IN MONTHS
PERSIST.
STANDARD
COUPLED
4 8 121620
Fig. 16.16 Forecast skill obtained with the Lamont model using two different initialization
methods. The results of the standard model (uncoupled initialization) are shown as dashed
lines and those for the new model (coupled initialization) as thin solid lines. As a reference
persistence is shown as a thick solid line. From Chen et al. 1997.
coupling shock and random noise. There is potential to improve the forecasts further, since the initial errors are stiH relatively large. This is a very encouraging
study and further investigations also incorporating more complex models should be
done. It should be mentioned, however, that the results with the coupled initialization might suffer from some artificial skill, since the nudging parameters were optimized with respect to the forecast skill.
16.5 Interpretation ofENSO forecast results
16.5.1 Definition of an appropriate skill measure
An objective measure that is commonly used to evaluate the performance of a
forecast system is the correlation skill. In the case of ENSO forecasts most groups