4 Discussion
The recurrence estimates reported here range from ~220 years (Semidi section) to ~1000 years (Sanak section) for Mw ≥ 8.5 events from geologic data (Table 1) and from 50 years (Mw 8.1, Fox Islands section) to 4750 years (Mw 8.3, Sanak section) from geodetic data (Table 2). The two approaches provide different, but complementary, views of rupture behavior. Along the energetic coasts of Alaska, geologic data capture only the largest ruptures, generally with preserved evidence of vertical deformation above detection limits of >0.2 m (Hawkes et al., 2010; Shennan et al., 2016) or tsunami runup > 5 m above the modern tidal range (Nelson et al., 2015; Witter et al., 2016, 2019). The geologic data also represents events that typically rupture multiple fault sections, which is demonstrated by historical events and inferred for prehistoric earthquakes – in this regard, the geologic recurrence rates should be viewed as participation rates in ruptures >Mw 8.5 for each fault section for which geologic data are available. The geodetic data are used to approximate strain accumulation and release by single fault section, and so recurrence rates of these events are necessarily shorter and the inferred earthquake magnitudes are smaller than events recorded by geology.
A primary advantage of the geodetic recurrence model is that it estimates moment accumulation on the model subduction interface sections in a general way. Because coupling magnitude and coupled area generally trade off in geodetic models, the exact location of the coupled patches is not critical. Instead, our goal is to estimate the moment accumulation budget available for interface ruptures, rather than using geodesy to strictly define rupture patches. By calculating the recurrence on sections we have defined a priori , we provide information that can be interpreted in the broader context of section magnitude frequency distributions and multi-section ruptures. The coupling polygons we present here are not meant to strictly correlate with rupture patches. Complex non-unique coupling models are often presented for subduction zone interfaces, including for some portions of the AASZ (Li et al., 2016), and the relation between interseismic coupling and interface ruptures can be modeled in intricate ways (Small & Melgar, 2021). However, the relation between geodetic coupling models (strain accumulation) and eventual rupture (strain release) is not straightforward, and the assumption that complex depictions of interseismic coupling uniquely predict future rupture patterns is not clearly supported (Noda and Lapusta, 2013; Tsang et al., 2015; Witter et al., 2019). For example, persistent asperity models such as most recently presented by Zhao et al. (2022) for the Shumagin and Semidi sections of the AASZ use recent ruptures to create non-unique backslip scenarios that fit sparse GNSS measurements reasonably well. These models offer limited utility for hazard forecasts because they effectively predict characteristic earthquakes (Schwartz, 1999) unless the persistent asperity assumption is relaxed and modified (Avouac, 2015), and a primary objective in modern seismic hazard modeling is to move beyond the assumption that past earthquakes uniquely predict future ruptures (Field et al., 2014).
Our effort to assign recurrence values along the AASZ points to several potential future studies and opportunities. Geologic studies would be beneficial to characterize rupture behavior in the western ~1,250 km of the subduction zone, for which no data are currently available. Geodetic data are fundamentally important for understanding subduction zone hazard, and a denser permanent GNSS network throughout the AASZ would improve hazard estimates – and especially on the seafloor, where recent GNSS-Acoustic studies (Brooks et al., 2023) have demonstrated the importance of seafloor geodesy. Instead of a single model, future coupling and slip-deficit models from geodesy might be presented as a suite of models that encompass the broadest possible range of uncertainty, such as multiple geodetic models presented in Schmalzle et al. (2014) and Mariniere et al. (2021).
The AASZ experienced a series of major ruptures in the 20th century along much of its length. Recurrence intervals for these largest ruptures are many centuries long, and it was previously assumed that apparently unruptured portions of the interface in the historical period, or seismic gaps (Davies et al., 1981), would be most likely to host future ruptures, and conversely, that regions that ruptured in the 20th century were likely no longer hazardous in the 21st century. However, re-rupture of a historical great earthquake rupture patch by subsequent large events has been documented or inferred along the AASZ (Schwartz, 1999; Tanioka & Gonzalez, 1998; Brooks et al., 2023). Complex rupture overlap is also supported by current observations and models of subduction interface frictional behavior, which demonstrate that subduction interfaces are a mosaic of slip patches along strike and down dip (Lay et al., 2012). The regions of interface slip associated with historical events are very poorly known (Nicolsky et al., 2016; Witter et al., 2019) and the extremely generalized historical rupture areas from aftershocks (Tape & Lomax, 2022) and coarse rupture models (Johnson & Satake, 1993) are insufficient to accurately define historical slip patches. Our recurrence results indicate that most sections we define along the AASZ are capable of > Mw 8 ruptures roughly every century, and that the locations of historical ruptures and presumed spatial variations in interseismic coupling are only loose constraints on future rupture locations and magnitudes.