I disagree with Robert Geller's hard-line stance against earthquake prediction (Nature 472, 407–409; 2011). Although early warning signs are diverse, fleeting and often subtle, they can also be surprisingly strong, even for moderate earthquakes (see, for example, T. Bleier et al. Nat. Hazards Earth Syst. Sci. 9, 585–603; 2009).

More than 100 years of seismology have led to an advanced understanding of the tectonic forces that cause Earth's plates to move, slide past each other and collide. But when it comes to earthquake prediction, the seismological approach has always been to try to understand how past events happened and to develop probability models for 'predicting' when the next ones might occur. This analysis has built-in statistical uncertainties that are of the order of years, decades, even centuries — and there is no way around it.

Any good seismologist will recognize the limitations of earthquake prediction. But the study of earthquakes should include the tracking down and investigation of all the different signals that Earth produces before a catastrophic rupture. If seismologists can't do it alone, can't we do it collectively across disciplines?