Reply to “Statistical tests of VAN earthquake predictions: Comments and reflections,” by Y. Kagan and D.D. Jackson

Abstract

In this Reply we restrict ourselves to those points that are again raised by Kagan and Jackson [1996], although we discussed them in our separate Replies to the papers by Kagan [1996] and Jackson [1996]. For example, we show, with precise quotations, that their claim: “…most of the VAN parameters have been adjusted after the 1987–1989 test period, i.e., retroactively” does not hold. On the other hand, we do agree with their main conclusion that “the diversity of views on these subjects suggests that the seismological community needs a general strategy for evaluating earthquake prediction methods.” Such a strategy, can now be followed for the evaluation of the VAN predictions published to date (which cover the continuous 9 year period 1987–1995), instead of restricting ourselves to the period 1987–1989 discussed in this debate. Kagan and Jackson [1996] also proceeded to a comparison of test results by Aceves et al. [1996] and by Kagan [1996], and concluded that both tests agree that VAN outperforms a Poisson null hypothesis, if aftershocks are left in the catalog or only partially removed. On the other hand, Kagan and Jackson [1996] claim that the VAN method does not outperform the Poisson null hypothesis “if aftershocks are more fully removed.” However, we clarify that the following happens in the “declustering” procedure of the latter case: successful predictions of main shocks are also eliminated, and furthermore, we have conversion of successful predictions to failures; our remarks are in basic agreement with those made by Aceves et al. [1996].