Pseudo-prospective forecasting of large earthquakes full distribution in circum-Pacific belt incorporating non-stationary modeling

The noticeable increase in the occurrence rate of the great (M≥8.0) global earthquakes since 2004, provokes the extensive investigation of non-stationarity in their temporal distribution. The reliable evaluation of their expected occurrence rates is a highly demanding aspect due to the co-existence of scarce long interevent times (extreme events) and localized clustering. In this work, we establish a two-step modeling procedure of the Markovian Arrival Process (MAP) to simultaneously forecast the occurrence of extreme events and short-term seismicity and we implement catalog-based pseudo-prospective forecasting experiments for the full distribution of the occurrence frequency. MAP model is a two-dimensional point process, whose intensity function is driven by a hidden Markov process, Jt. We adopt the existence of an “idle” state that corresponds to periods of relative seismic quiescence and its occurrence rate, λidle, is estimated a-priori, which is then embedded as a constant in the iterative fitting procedure of the MAP. We show the superiority of the MAP over renewal models and the temporal Epidemic Type Aftershock Sequence model, by producing robust probabilistic forecasts for the full temporal distribution of large earthquakes (M≥7.6) in circum-Pacific belt during 1918–2020, after considering a sensitivity analysis over both the testing periods and the magnitude thresholds.