Advancing the ETAS Model to Improve Forecasts of Earthquake Sequences and Doublets

<p><span>Earthquake sequences typically show distinct spatiotemporal patterns, characterized by a power-law decay</span><br><span>of aftershock times and elongated aftershock distributions around the (extended) rupture.</span> <span>A prominent approach to</span><br><span>model seismic clustering in space and time is the</span> <span>Epidemic Type Aftershock Sequence (ETAS)</span> <span>model that differentiates</span><br><span>an independent background seismicity process from a branching tree process for triggered events.</span> <span>The conventional</span><br><span>ETAS</span> <span>approach</span> <span>shows</span> <span>three</span> <span>substantial</span> <span>biases:</span> <span>(1)</span> <span>The</span> <span>assumption</span> <span>of</span> <span>isotropic</span> <span>spatial</span> <span>distributions</span> <span>of</span> <span>aftershock</span><br><span>locations stands in contrast to observations and geophysical models for large mainshocks.</span> <span>(2) The unlimited spatial</span><br><span>distribution</span> <span>allows</span> <span>small</span> <span>events</span> <span>to</span> <span>trigger</span> <span>aftershocks</span> <span>at</span> <span>unrealistically</span> <span>large</span> <span>distances.</span> <span>(3)</span> <span>Short-term</span> <span>incomplete</span><br><span>event records after large mainshock events suggest supposedly smaller aftershock productivity and cluster sizes.</span> <span>The</span><br><span>above</span> <span>biases</span> <span>can</span> <span>lead</span> <span>to</span> <span>an</span> <span>underestimation</span> <span>of</span> <span>the</span> <span>aftershock</span> <span>productivity</span> <span>of</span> <span>strong</span> <span>events,</span> <span>and</span> <span>in</span> <span>consequence</span> <span>to</span><br><span>underpredicted cluster sizes, and to a miss-specification of the spatial aftershock distribution in the case of clearly ex-</span><br><span>tended ruptures. Here, we combine an ETAS-Incomplete model, accounting for short-term aftershock incompleteness,</span><br><span>with an ETAS approach applying anisotropic,</span> <span>spatially restricted distributions of aftershock locations.</span> <span>We evaluate</span><br><span>the benefits of these models by running forecast experiments for the 2019 Ridgecrest sequence and analyzing the oc-</span><br><span>currence frequencies of so-called</span> <span>Earthquake Doublets</span><span>,</span> <span>i.e.,</span> <span>sequences of two or more similarly strong earthquakes</span><br><span>within a small time-space window. The new model provides more realistic sequence forecasts and doublet predictions</span><br><span>and might be of particular interest for (short term) risk assessment units.</span></p>