The Prediction Paradox

The storms we want to predict, are the BIG ones, but our best statistics are the little storms. Statistical precursors, neural nets, linear filters ALWAYS do better on the little storms, not the killer storms.
We will never achieve reasonable prediction until we understand how/where/why BIG storms occur. Everything else is gambling.
But the statistics have not led to better physics. Why is Vsw the best correlator? Especially since Dst is better correlated to Ey and AE to Bz. Why is Kp the best correlated ground-based index?

The storms we want to predict, are the BIG ones, but our best statistics are the little storms. Statistical precursors, neural nets, linear filters ALWAYS do better on the little storms, not the killer storms.