How good is a paleoseismic record of megathrust earthquakes for probabilistic forecasting?

Time-dependent earthquake forecast depends on the frequency and number of past events and time since the last event. Unfortunately, only a few past events are historically documented along subduction zones where forecasting relies mostly on paleoseismic catalogs. We address the role of dating uncertainty and completeness of paleoseismic catalogs on probabilistic estimates of forthcoming earthquakes using a 3.6-ka-long catalog including 11 paleoseismic and 1 historic (Mw> or =8.6) earthquakes that preceded the great 1960 Chile earthquake. We set the clock to 1940 and estimate the conditional probability of a future event using five different recurrence models. We find that the Weibull model predicts the highest forecasting probabilities of 44% and 72% in the next 50 and 100 yr, respectively. Uncertainties in earthquake chronologies due to missing events and dating uncertainties may produce changes in forecast probabilities of up to 50%. Our study provides a framework to use paleoseismic records in seismic hazard assessments including epistemic uncertainties.