The Relationship between Prediction Accuracy and Correlation Coefficient

Abstract

The correlation coefficient (r) between the maximum amplitude (R _m) of a sunspot cycle and the preceding minimum aa geomagnetic index (aa _min), in terms of geomagnetic cycle, can be fitted by a sinusoidal function with a four-cycle periodicity superimposed on a declining trend. The prediction index (χ) of the prediction error relative to its estimated uncertainty based on a geomagnetic precursor method can be fitted by a sinusoidal function with a four-and-half-cycle periodicity. A revised prediction relationship is found between the two quantities: χ<1.2 if r varies in a rising trend, and χ>1.2 if r varies in a declining trend. The prediction accuracy of R _m depends on the long-term variation in the correlation. These results indicate that the prediction for the next cycle inferred from this method, R _m(24)=87±23 regarding the 75% level of confidence (1.2-σ), is likely to fail. When using another predictor of sunspot area instead of the geomagnetic index, similar results can be also obtained. Dynamo models will have better predictive powers when having considered the long-term periodicities.