Probabilistic seismic hazard assessments (PSHAs) provide the scientific basis for building codes to reduce damage from earthquakes. Despite their substantial impact, little is known about how well PSHA predicts actual shaking. Recent PSHA for California, Japan, Italy, Nepal, and France appear to consistently overpredict historically observed earthquake shaking intensities. Numerical simulations show that observed shaking is equally likely to be above or below predictions. This result from independently developed models and datasets in different countries and tectonic settings indicates possible systematic bias in the hazard models, the observations, or both. Analysis of possible causes shows that much of the discrepancy is due to a subtle and rarely considered issue: the conversion equations used in comparing the models—which forecast shaking as peak ground acceleration or velocity—and observations—parameterizations of qualitative shaking reports. Historical shaking reports fill a crucial data gap, but more research is warranted on how qualitative observations relate to instrumental shaking measures for earthquakes.

ABSTRACT Paleoseismic studies show large variability in earthquake interevent times along a fault, with short intervals often separated by long quiescent periods. Some paleoseismologists have interpreted this variability as a product of an earthquake’s partial strain release with the next earthquake occurring sooner than expected because of the remaining residual strain. However, commonly used probabilistic large earthquake recurrence models attribute this variability purely to chance, not the state of strain on the fault. Here, we present an alternative probabilistic model, built on the long-term fault memory model framework that better reflects the strain accumulation and release process. This generalized long-term fault memory model (GLTFM) predicts that this interevent time variability arises from both chance and the state of strain on the fault. Specifically, it estimates when residual strain is likely present and its impact on the timing of the next earthquake in the sequence. In addition, GLTFM assumes that additional accumulated strain always increases earthquake probability. In contrast, the commonly used lognormal and Brownian passage time models predict that the probability of a large earthquake stays constant or even decreases after it is “overdue” (past the observed average recurrence interval) so additional accumulated strain does not make an earthquake more likely. GLTFM’s simple implementation and versatility should make it a powerful tool in earthquake forecasting.

Compilations of earthquake moment tensors from global and regional catalogs find pervasive non-double-couple (NDC) componentswith a mean deviation from a double-couple (DC) source of around 20%. Their distributions vary only slightly with magnitude, faulting mechanism, or geologic environments. This consistency suggests thatfor most earthquakes, especially smaller ones whose rupture processes are expected to be simpler, the NDC components are largely artifacts of the moment tensor inversion procedure. This possibility is also supported by the fact that NDC components for individual earthquakes with Mw<6.5 are only weakly correlated betweencatalogs. We explore this possibility by generating synthetic seismograms for the double-couple components of earthquakes around theworld using one Earth model and inverting them with a different Earth model. To match the waveforms with a different Earth model, the inversion changes the mechanisms to include a substantial NDC component while largely preserving the fault geometry (DC component). The resulting NDC components have a size and distribution similar to those reported for the earthquakes in the Global Centroid Moment Tensor (GCMT) catalog. The fact that numerical experiments replicate general features of the pervasive NDC components reported in moment tensor catalogs implies that these components are largely artifacts of the inversions not adequately accounting for the effects of laterally varying Earth structure.

The availability of digital seismic waveform data enabled compilation of seismic moment tensor catalogs that provide information about earthquake source processes beyond what could be derived from earlier methods that assume double-couple sources representing slip on planar faults. This additional versatility involves additional complexity. Moment tensors are determined by inversions minimizing the misfit between observed and synthetic waveforms, and depend on the specifics of the data inverted, the inversion algorithm, and the Earth structure assumed. Hence, substantial uncertainties arise in moment tensors and quantities derived from them, which can be assessed by comparing moment tensors from multiple global and regional catalogs using different data and inversion procedures. While the double-couple (DC) components of moment tensors are generally determined with greater certainty, non-double-couple (NDC) components for the same earthquake sometimes differ significantly between catalogs. This observation raises questions about the reliability of their determination and hence their geological significance. Using the correlation between NDC components in different catalogs, we quantify the reliability of NDC components in moment tensor catalogs through the determination of the effects of unmodeled and inaccurately modeled effects contained in them. We determine that the NDC components in the Global Centroid Moment Tensor catalog are, on average, more precise than in other catalogs, and thus studies on NDC components should be based on this catalog. Furthermore, their uncertainties are largely unrelated to uncertainties in the DC components. Therefore, the reliability of fault angles derived from a moment tensor is largely independent from the reliability of its NDC components.

&lt;p&gt;Seismologists have recently begun assessing how well probabilistic earthquake hazard models and corresponding maps forecast shaking that actually occurs. Based on several metrics, we find the 2018 National Seismic Hazard Model for California appears to overpredict the level of high-frequency shaking shown by the California Historical Intensity Mapping Project (CHIMP) dataset of maximum observed shaking from the largest Californian earthquakes between 1857 and 2019. Understanding this overprediction, which is qualitatively similar to that observed in France, Italy, Japan, and Nepal, could lead to improved seismic hazard models for California and, by extension, worldwide.&amp;#160; There are three possible reasons for the discrepancy between observed and predicted shaking: hazard models may be biased high (examined here), historical data may be biased low, or misfit may arise by chance.&lt;/p&gt; &lt;p&gt;We examine several assumptions used in prior studies to determine if they cause the overprediction: 1) not considering site effects from variations in V&lt;sub&gt;S30&lt;/sub&gt; (time-averaged shear-wave velocity in the top 30m of soil), which could amplify or deamplify ground motions relative to the reference hazard model, 2) assuming the CHIMP data set is complete at magnitude 5 used as the minimum magnitude (M&lt;sub&gt;Min&lt;/sub&gt;) in hazard calculations, and 3) the ground-motion to intensity conversion equation (GMICE) used to convert hazard model peak ground acceleration (PGA) measures to Modified Mercalli Intensity (MMI) values, consistent with the historic data, and 4) the use of mean versus median hazard. We show that incorporation of site-specific V&lt;sub&gt;S30&lt;/sub&gt; does not appreciably change the models and hence their consistency with CHIMP data, because at the short periods that control PGA and hence MMI,&amp;#160;nonlinear deamplification due to increased soil damping largely offsets linear amplification due to low V&lt;sub&gt;S30&lt;/sub&gt;.&amp;#160; Increasing M&lt;sub&gt;Min&lt;/sub&gt; reduces but does not eliminate the discrepancy. Biases in current GMICE methodology appear to be major cause in the discrepancy. Mean hazard estimates, a modeling choice, predict higher ground motions than the median, which may reduce the discrepancy. We discuss the effects of each of these assumptions, which we expect to be consistent for other areas.&lt;/p&gt;

Estimating the magnitude of historical earthquakes is crucial for assessing seismic hazard. Magnitudes of early-instrumental earthquakes can be inferred using a combination of instrumental records, field observations, and the observed distribution of shaking intensity determined from macroseismic observations. For earthquakes before 1900, shaking intensity distributions often provide the only information to constrain earthquake magnitude. Considerable effort has been made to develop methods to estimate the magnitude of moderate-to-large historical earthquakes using shaking intensities derived from macroseismic data. In this study, we consider earthquakes in California with known instrumental magnitudes to explore uncertainties in estimating the magnitude of historical earthquakes from intensity information alone. We use three California-specific intensity prediction equations (IPEs) and an IPE based on a global ground-motion model (GMM) to determine optimum intensity-based magnitudes for 33 moderate-to-large California earthquakes between 1979 and 2021. Intensity-based magnitudes are close to instrumental magnitudes on average. However, intensity-based magnitudes for individual events differ by as much as 2.2 magnitude units from instrumental magnitudes. This result reflects the weak dependence of ground motions and shaking intensities on moment magnitude and their strong dependence on stress drop. Considering the intensity distributions of the 1906 San Francisco and 1989 Loma Prieta earthquakes, we show that information that could constrain rupture length is discarded when considering only the 2D decay of intensity with distance. We also show that ground-motion intensity conversion equations used in a GMM-based approach may cause a systematic overestimation of large historical earthquake magnitudes. This study underscores both the reducible and potentially irreducible uncertainties associated with using intensity data to estimate magnitudes of historical earthquakes using IPEs and highlights the value of using additional information to constrain rupture dimensions. Using intensity observations alone, moment magnitude uncertainties are typically on the order of a full unit.

Forecasts of the probability of a large earthquake occurring on a fault during a specific time interval assume that a probability distribution describes the interevent times between large earthquakes. However, current models have features that we consider unrealistic. In these models, earthquake probabilities remain constant or even decrease after the expected mean recurrence interval, implying that additional accumulated strain does not make an earthquake more likely. Moreover, these models assume that large earth-quakes release all accumulated strain, despite evidence for partial strain release in earth-quake histories showing clusters and gaps. As an alternative, we derive the necessary equations to calculate earthquake probabilities using the long-term fault memory (LTFM) model. By accounting for partial strain release, LTFM incorporates the specific timing of past earthquakes, which commonly used probability models cannot do, so it can forecast gaps and clusters. We apply LTFM to the southern San Andreas fault as an example and show how LTFM can produce better forecasts when clusters and gaps are present. LTFM better forecasts the exceptionally short interevent time before the 1857 Fort Tejon earthquake. Although LTFM is more complex than existing models, it is more power-ful because (unlike current models) it incorporates fundamental aspects of the strain accu-mulation and release processes causing earthquakes.

There has been considerable discussion as to how to assess when non-double-couple (NDC) components of seismic moment tensors represent real source processes. We explore this question by comparing moment tensors (MTs) of earthquakes in three global catalogs, which use different inversion procedures. Their NDC components are only weakly correlated between catalogs, suggesting that they are largely artifacts of the inversion. A monotonic decrease in the NDC components' standard deviation with magnitude indicates increased reliability of the NDC components for larger earthquakes. The standard deviation begins to decrease for large NDC components exceeding 60%, suggesting that they represent real source processes. Randomly generated NDC components with the same mean and standard deviation as in the MT catalogs only reproduce some of this decrease. Thus NDC components of large earthquakes and NDC components that exceed 60% are likely to represent real source processes.

Broadband seismograms from the EarthScope Transportable Array and Superior Province Rifting EarthScope Experiment (SPREE) deployments are used to map the crust and uppermost mantle structures beneath the failed Midcontinent Rift (MCR) of Minnesota/Wisconsin, USA. The results suggest the existence of a variable zone of mafic underplating that is up to 20 km thick (40-60 deep). We jointly invert receiver functions and Rayleigh wave dispersion curves to quantify the region's crustal and mantle shear-wave velocity structure. Basin sediment thicknesses are mildly asymmetric about the rift axis, with thickest regions immediately beneath the rift. 3-D modeling shows anomalous lower crust and crust-mantle transitions beneath the MCR. Sub-MCR crustal thicknesses are generally >50 km with lower crust Vs of 4.0-4.2 km/s. Away from the MCR, the crust is typically similar to 40 km thick. Strong variations in apparent crustal thickness are found along the MCR, increasing significantly in places. An additional layer of shear velocities intermediate between typical lower crust and upper mantle velocities (4.1-4.6 km/s) exists beneath most of the MCR which is thickest beneath the rift axis and pinches out away from the rift. This structure corroborates previous proposals of the presence of an underplated layer near the Moho. Results cannot distinguish between different mechanisms of emplacement (e.g., mafic interfingering within a subsequently down-dropped lower crust vs. development of a high-density pyroxenitic residuum at the top of the mantle). Also observed are anomalously high (>4.7 km/s) sub-rift shear-wave velocities at similar to 70-90-km depths, suggesting the presence of cold, depleted upper mantle material.

A recent topic of interest is the performance of probabilistic seismic hazard maps relative to historical shaking datasets. Maps developed for different countries appear to overpredict shaking relative to historical shaking datasets. To explore whether this discrepancy arises because of incompleteness in historical datasets, we consider maps and historical data from California. Current probabilistic seismic hazard maps for California appear to predict stronger short period shaking than historical maxima captured by the California Historical Intensity Mapping Project (CHIMP) dataset between 1857 and 2019. We estimate that CHIMP has a magnitude completeness between M 6-6.6, whereas California hazard maps assume a minimum magnitude (MMin) of 5. Disaggregating the maps shows that earthquakes smaller than M 6 and 6.6 respectfully contribute about 25% and 45% of the hazard across California. Increasing the hazard map's MMin to 6 and 6.6 respectively reduces the discrepancy between predicted and observed shaking by approximately 10-20% and 30-35%. These reductions are not enough to bring the maps and data in alignment. Thus, MMin inconsistencies contribute to, but are not the sole cause of, the discrepancy between predicted and historically observed shaking. These results may be generally applicable to maps elsewhere, although MMin will vary for different historical datasets.

The North American craton preserves nearly two billion years of geologic history, including three major rifts that failed rather than evolving to continental breakup and seafloor spreading. The Midcontinent Rift (MCR) and Southern Oklahoma Aulacogen (SOA) show prominent gravity anomalies due to large volumes of igneous rift-filling rock. The Reelfoot Rift (RR), though obscure in gravity data, is of interest due to its seismicity. The ca. 1.1 Ga MCR records aspects of the assembly of Rodinia, whereas the ca. 560 Ma SOA and RR initiated during the later breakup of Rodinia and were inverted during the assembly of Pangea. Comparative study of these rifts using geophysical and geological data shows intriguing similarities and differences. The rifts formed in similar tectonic set-tings and followed similar evolutionary paths of extension, magmatism, subsidence, and inversion by later compression, leading to similar width and architecture. Differences between the rifts reflect the extent to which these processes occurred. Further study of failed rifts would give additional insight into the final stages of continental rifting and early stages of seafloor spreading.

In two classic papers, Burke and Dewey (1973) and Dewey and Burke (1974) proposed that continental rifting begins at hotspots - domal uplifts with associated magmatism - from which three rift arms extend. Rift arms from different hotspots link up to form new plate boundaries along which the continent breaks up, generating a new ocean basin and leaving failed arms termed aulacogens within the continent. In subsequent studies, hotspots became increasingly viewed as manifestations of deeper upwellings or plumes, which were the primary cause of continental rifting. We revisit this conceptual model and find that it remains useful, though some aspects require updates based on subsequent results. Many three-arm systems identified by Burke and Dewey (1973) are now recognized to be or have been boundaries of transient microplates accommodating motion between diverging major plates. Present-day examples include the East African Rift system and the Sinai microplate. Older examples include rifts associated with the opening of the South Atlantic in the Mesozoic and the North Atlantic Ocean over the last 200 Ma, rifts in the southern U.S associated with the breakup of Rodinia, and intracontinental rifts formed within India during the breakup of Gondwanaland. The microplates form as continents break up, and are kinematically distinct from the neighboring plates, in that they move separately. Ultimately, the microplates are incorporated into one of the major plates, leaving identifiable fossil features on land and/or offshore. In many cases the boundaries of microplates during continental breakup are located on preexisting zones of weakness and influenced by pre-existing fabric, including older collisional zones. Hotspots play at most a secondary role in continental breakup, in that most of the associated volcanism reflects plate divergence, so three-arm junction points may not reflect localized upwelling of a deep mantle plume.

The Midcontinent Rift of North America is a ca. 1.1 Ga, 3000-km-long failed rift that nearly split the Precambrian continent of Laurentia. Unlike most continental rifts, which are filled with a mixture of volcanic rocks and sediments, the Midcontinent Rift contains a large volume of flood basalts that were emplaced during both syn- and post-rift stages. Consequently, the Midcontinent Rift, which comprises the Keweenaw large igneous province, is the most significant positive anomaly on gravity maps of central North America. We investigated the mantle conditions required to produce this large volume of flood basalt and the observed two main stages of emplacement. To explore whether these magma volumes required a plume or, instead, could have resulted from the increased ambient mantle temperatures expected for the Neoproterozoic, we used a geodynamic model for a range of ambient mantle and plume temperatures under different scenarios of lithospheric extension. The most favorable scenario for the generation of both syn-rift and post-rift lavas combines a plume with excess temperatures between 175 and 225 °C introduced during the syn-rift phase and ambient mantle potential temperatures between 1393 and 1443 °C, with an initial lithospheric thickness not exceeding 150 km for 3 mm/yr extension rates.

Earthquake scientists have traditionally relied on relatively small data sets recorded on small numbers of instruments. With advances in both instrumentation and computational resources, the big-data era, including an established norm of open data-sharing, allows seismologists to explore important issues using data volumes that would have been unimaginable in earlier decades. Alongside with these developments, the community has moved towards routine production of interpreted data products such as seismic moment tensor catalogs that have provided an additional boon to earthquake science. As these products have become increasingly familiar and useful, it is important to bear in mind that they are not data, but rather interpreted data products. As such, they differ from data in ways that can be important, but not always appreciated. Important - and sometimes surprising - issues can arise if methodology is not fully described, data from multiple sources are included, or data products are not versioned (time-stamped). The line between data and data products is sometimes blurred, leading to an underappreciation of issues that affect data products. This note illustrates examples from two widely used data products: moment tensor catalogs and Did You Feel It? (DYFI) macroseismic intensity values. These examples show that increasing a data product’s documentation, independence, and stability can make it even more useful. To ensure the reproducibility of studies using data products, time-stamped products should be preserved, for example as electronic supplements to published papers, or, ideally, a more permanent repository.

Prior studies have repeatedly shown that probabilistic seismic hazard maps from several different countries predict higher shaking than that observed. Previous map assessments have not, however, considered the influence of site response on hazard. Seismologists have long acknowledged the influence of near-surface geology, in particular low-impedance sediment layers, on earthquake ground-motion at frequencies of engineering concern. Although the overall effects of site response are complex, modern ground-motion models (GMMs) account for site effects using terms based on V S30 , the time-averaged shear-wave velocity in the upper 30 m of the Earth’s surface. In this study, we consider general implications of incorporating site terms from modern GMMs using site-specific V S30 as a proxy in probabilistic seismic hazard maps for California. At the long periods (1–5 s) that affect tall buildings, site terms amplify the mapped hazard by factors of 1–3 at many sites relative to maps calculated for the standard reference soft-rock site condition, V S30 = 760 m/s. However, at the short periods of ground-motion that are the main contributors to peak ground acceleration (PGA) and thus affect smaller structures, only negligible effects occur due to nonlinear deamplification of strong ground-motion at high frequencies. Nonlinear deamplification increases as the shaking level increases. For very strong shaking, deamplification can overcome the linear amplification, yielding net deamplification. We explore the implications of these results for the evaluation of hazard maps. Because site effects do not change the maps appreciably at short periods, we can exclude site response as an explanation for why the maps overpredict historically observed shaking as captured by the California Historical Intensity Mapping Project (CHIMP) dataset. The results are expected to be generalizable to regions that are comparable to California in terms of structure and seismicity rates. In low-to-moderate-seismicity regions where the hazard reflects weaker shaking, nonlinear site response is expected to be less important for the hazard.