Regionalization of global subduction ground-motion model for application in Central America and Mexico

Abstract

We extend data resources originally developed for Central America and Mexico in the Next-Generation Attenuation (NGA)-Subduction (NGA-Sub) project database and use the data to customize a global ground-motion model (GMM) for regional application. The GMM applies for peak ground acceleration, peak ground velocity, and 5%-damped pseudo-spectral accelerations for periods from 0.01 to 10 s, for the median orientation-independent horizontal component of ground motion. The database for Mexico is extended by considering small magnitude (M < 6) events and three large events (M7.2–M8.3) in 2017 and 2018. The latter are particularly important, because they are well recorded over a broad distance range and apply for hazard-critical conditions. These changes increase the size of the database from 593 recordings from 66 events (NGA-Sub) to 1882 recordings from 121 events. The expanded data are examined relative to a global NGA-Sub GMM, which reveals a number of interesting features including faster anelastic attenuation in backarc than forearc regions and generally similar levels of site response to those in the global GMM. These features are incorporated into the regionally customized model. Advantages of this GMM relative to previous regional models applied at national scale for engineering applications are that its functional form has a physical basis, and its coefficients are well constrained from global data and are only adjusted for the target region as required by regional data.

Keywords

Subduction zones Mexico ground-motion database ground-motion model database anelastic attenuation

Introduction

The NGA-Sub project developed a relational database and ground-motion models (GMMs) for subduction zone earthquakes worldwide (Bozorgnia et al., 2022). The database provides uniformly processed ground-motion data from earthquakes recorded in different tectonic settings and regions, including time series and intensity measure (IM) values (Ahdi et al., 2022; Contreras et al., 2022; Mazzoni et al., 2022). The well-populated magnitude and rupture distance ranges for interface and intraslab events in the database are M = 4.7–9.1 and R_rup = 25–2000 km (interface) and M = 4.0–8.4 and R_rup = 35–2000 km (slab). The database has been used to develop three global GMMs, each of which has regional adjustment factors (Abrahamson and Gulerce, 2022; Kuehn et al., 2023; Parker et al., 2022), and regional models for Chile (Macedo and Liu, 2022), Japan (Si et al., 2022), and northern South America (Arteta et al., 2021).

The NGA-Sub database development had a cutoff date of 2016, meaning that earthquakes after that date were not considered. As inevitably occurs in any such project, application of the cutoff caused important data to be omitted. This was particularly the case in Central America and Mexico (CAM), because of large-magnitude, well-recorded earthquakes that occurred in 2017 and 2018. The present work aimed to augment the NGA-Sub database with this newer information, which was developed in a consistent manner to NGA-Sub. Our second objective was to evaluate the performance of two CAM-regionalized global NGA-Sub GMMs (Parker et al., 2022: Pea22; Kuehn et al., 2023: Kea23). This article is adapted from a report (Contreras et al., 2023a) that introduced the database and analyses. An important topic not addressed here, but which is the subject of a companion article (Contreras et al., 2025), concerns site response in the Valley of Mexico. Portions of the work presented in this article have also been the subject of a conference presentation (Contreras et al., 2023b).

Previous Mexico GMMs

Networks and data sets

In Mexico, the installation of a strong motion network started in the early 1960s, following the 1957 M_w 7.6 Guerrero earthquake. In 1962, three accelerographs were deployed in Mexico City, two at the city center and one at a hill zone site (Ciudad Universitaria, CU). Since 1964, the CU station has recorded several events and has become a widely applied reference site for studies of earthquake ground motions in the Valley of Mexico (Arroyo et al., 2024; Jaimes et al., 2006; Reinoso and Ordaz, 1999).

Starting in the 1980s, the Universidad Nacional Autónoma de México (UNAM, National Autonomous University of Mexico) in cooperation with the University of California, San Diego, significantly increased the number of instruments with the deployment of the Guerrero Accelerograph Network (Anderson et al., 1994). This network is not in operation at present; however, some recording stations were absorbed into the networks installed by the Institute of Engineering at UNAM. Following the 1985 M8.0 Michoacan earthquake, several institutions installed and operated networks. Currently, the main strong motion data operators in Mexico are the Institute of Engineering at UNAM (IINGEN), the Servicio Sismológico Nacional (SSN, National Seismological Service of Mexico) at the Geophysical Institute of UNAM, the Centro de Registro e Instrumentación Sísmica (CIRES, Instrumentation and Seismic Recording Center) in Mexico City, the Seismic Network of Centro de Investigación Científica y Educación Superior de Ensenada (CICESE, Center of Scientific Research and Higher Education of Ensenada) in Baja California, and the Federal Electricity Commission. There are additional smaller networks maintained by other operators in different states of Mexico.

Table 1 summarizes ground-motion data sets (each of which has an accompanying GMM) for Mexico that were developed utilizing recordings from subduction earthquakes. Earthquake ground-motion recordings and modeling have often focused on the Valley of Mexico (Castro et al., 1988; Ordaz et al., 1994, 2024; Rosenblueth et al., 1989; Singh et al., 1987). Mexico City has unique geotechnical conditions, consisting of lakebed deposits, that produce strong site effects. Less attention has been given to the areas outside the Valley of Mexico.

Table 1.

Ground-motion data sets and GMMs for subduction earthquakes in Mexico. Models specific to Mexico City are excluded

Author	Region	Time frame	#Events	#Stns.	#Recordings	Mag.range	R _rup or R_hyprange (km)
García et al. (2005)	Central Mexico	1994–2004	16 intraslab	51	277	5.2–7.4	4–400
García et al. (2009)	Mexico	1985–2004	46 interface	56	469	5.0–8.0	20–400
Hong et al. (2009)	Mexico	1985–2004	40 interface 16 intraslab	51	418 277	5.0–8.0 5.2–7.4	Unspecified
Arroyo et al. (2010)	Mexico(Pacific coast)	1985–2004	40 interface	56	418	5.0–8.0	20–400
Rodríguez-Pérez (2014) ^a	Central & southern Mexico	1995–2011	8 interface 25 intraslab	36	75 121	5.1–8.0 5.0–7.2	50–580 70–540
García-Soto and Jaimes (2017)	Mexico(Pacific coast)	1985–2004	40 interface	56	418	5.0–8.0	17–400
Jaimes and García-Soto (2020)	Mexico	1994–2017	23 intraslab	69	366	5.2–8.2	22–400
Lermo-Samaniego et al. (2020)	Southern east Mexico	1995–2017	86 subduction & others	9	261	5.0–8.2	52–618
Jaimes and García-Soto (2021)	Mexico	1985–2017	40 interface 23 intraslab	56 69	418 366	5.0–8.0 5.2–8.2	17–400 22–400

Near-trench thrust interface earthquakes only.

GMMs

For forearc regions, GMMs for peak ground acceleration (PGA) and modified Mercalli intensity were developed in the late 1980s and 1990s (Anderson, 1997; Anderson and Lei, 1994; Anderson and Quaas, 1988; Ordaz et al., 1989), which are mainly applicable for the state of Guerrero. Perea and Sordo (1988) produced a local GMM for the urban area of Puebla City, and Reyes (1998) used data from the CU station. These GMMs did not distinguish between event types and are not listed in Table 1.

Current practice for ground-motion prediction in Mexico is different for Mexico City and the remainder of the country. The specific procedures used in Mexico City are summarized in Chapter 5 of Contreras et al. (2023a). The GMMs in Table 1 apply for the overall Mexico region or specific subregions, but are generally not used in Mexico City. Among the models in Table 1, the most applicable models to the entirety of Mexico are those of García-Soto and Jaimes (2017) for interface earthquakes and Jaimes and García-Soto (2020) for intraslab events. As shown in Table 1, the interface data set utilized 418 recordings from 40 interface events and exclusively National Earthquake Hazards Reduction Program (NEHRP) B sites (56 in total). This GMM predicts peak acceleration (PGA), peak velocity (PGV), and pseudo-spectral acceleration (PSA) for oscillator periods (T) between 0.01 and 5 s. In addition, it estimates vertical spectral accelerations and vertical-to-horizontal spectral acceleration ratios. The intraslab data set consists of 366 recordings from 23 events, again utilizing only NEHRP B sites (69 in total). The extended database included the relatively recent large earthquakes that occurred in Mexico in September 2017: the 8 September 2017 M8.3 offshore Chiapas event and the 19 September 2017 M7.2 Puebla event. This model estimates PGA, PGV, and PSA for the horizontal and the vertical components, for T = 0.01 to 5 s.

The present work differs from these previous studies in several respects. Model development considers all applicable data, without the restriction of NEHRP B sites. This was done because site response models integrated within the GMM can address first-order site effects, and by considering such models within the analyses, it is possible to better constrain source and path effects. A second distinction, as noted in the Introduction, is that instead of developing models solely from Mexican data, data from the CAM region are used to adjust a global model that is based on a much larger database and thus better constrained in key attributes such as magnitude and hypocentral depth scaling of ground motions, among others.

CAM data

The NGA-Sub database contains more than 71,000 three-component time series from 1883 earthquakes worldwide. We have extended and improved the NGA-Sub database for the CAM region using events with dates extending through 2018. There are two reasons that events in the extended database were not used in the original database. The first is that the events occurred after a 2016 cutoff date, which excluded from consideration two large intraslab events in Mexico (M7.2 and M8.3) that occurred in September 2017 and one interface earthquake (M7.2), also in Mexico, that occurred in February 2018. Second, the event magnitudes were below a threshold of 6, which had been applied to CAM events during NGA-Sub to control workload.

The additional data enhance ground-motion characterization in the following respects:

(a) Large-magnitude events provide data within a hazard-critical parametric range. Particularly, the 2017 M8.3 earthquake has a larger magnitude than the magnitude break parameter considered in this region for intraslab events (M7.4; Ji and Archuleta, 2018). As such, it can be considered in estimating the slope of the magnitude-scaling function beyond the break magnitude, which is useful both for CAM and globally.

(b) By adding more recordings, including those from large-magnitude events recorded over a wide distance range, it is possible to improve regional path characterization. Specifically, additional recorded ground motions at both forearc and backarc sites help to identify regional backarc effects.

(c) Adding new ground motions increases the number of recordings at selected sites, which enables the evaluation of non-ergodic (site-specific) site response effects at those sites.

In this section, we identify areas where the CAM component of the database was extended and improved. First, we describe the development of source and path metadata for M < 6 events that were included in the NGA-Sub database but for which critical event type classifications (interface, intraslab, shallow crustal, or outer rise) were not made. In the absence of such classifications, those events were not used in model development. We next describe source metadata and ground-motion data for the aforementioned three large events from 2017 to 2018. Finally, we describe path and site metadata for both newly added stations and stations that had been used previously in NGA-Sub.

NGA-Sub event metadata

There are 181 earthquakes from the CAM region in the NGA-Sub database for which classifications of event type (interface, intraslab, shallow crustal, or outer rise) were not provided. As a result, the data from these events were not used during model development. We selected 59 of the 181 unclassified events, each recorded at five or more stations, to develop source and path metadata following NGA-Sub protocols (Contreras et al., 2022). Figure 1 shows the magnitude–distance distribution in CAM for the updated database differentiated by event type (both the M < 6 and larger events are included). The added recordings from M < 6 events are 704, including 418 recordings from interface events and 286 from intraslab events. More information on the event classification and source metadata development is provided by Contreras et al. (2023a).

Figure 1.

Magnitude–distance distribution of the NGA-Sub database in CAM following updates of events with M < 6 and the addition of the three large-magnitude earthquakes that occurred in 2017 and 2018. Data from the NGA-Sub database are taken from the April 2019 flatfile, which includes data for events through 2016.

Post-NGA-Sub events

In this subsection, we describe source metadata for three large earthquakes that occurred after the 2016 NGA-Sub cutoff date: intraslab events on 8 September 2017 and 19 September 2017 (M8.3 offshore Chiapas earthquake and M7.2 Puebla earthquake, respectively) and an interface event on 16 February 2018 (M7.2 Oaxaca). Figure 2 shows event locations as moment tensors and finite-fault models (described further below) along with the locations of ground-motion stations that recorded at least one of the events.

Figure 2.

Event locations (focus and finite fault) for three events in 2017–2018 considered in the expansion of the CAM portion of the NGA-Sub database. Locations of ground-motion stations that recorded at least one of the events are also shown.

Table 2 presents moment tensor and finite-fault attributes of the three events. The seismic moments are taken from the Global Centroid Moment Tensor (CMT) catalog (Ekström et al., 2012) following NGA-Sub procedures. As described by Contreras et al. (2023a), multiple finite-fault models are available in the literature for each event. The selected models indicated in Table 2 were generally chosen because they considered multiple data sources, including nearby strong motion data and crustal displacements, and were published in peer-reviewed journals. In the case of the Chiapas event, tsunami data were also considered, and in the case of the Oaxaca event, a model that considered local seismic data was not available. Contreras et al. (2023a) provide plots of the as-published and trimmed finite-fault models; the trimming followed NGA-Sub protocols by removing portions of rupture rectangles with <15% of the maximum slip (justification provided by Contreras et al., 2022). The Chiapas finite model is a single rectangle whereas the Puebla and Oaxaca events are modeled with two rectangles with equivalent strike and dip angles but different lengths and widths. Additional metadata for each event is included in the source table (Contreras et al., 2024).

Table 2.

Moment tensor and finite-fault (rupture rectangle length and width) attributes of three large earthquakes since 2016 added to the CAM portion of database

Event	Finite-fault reference	Data types considered in finite-fault inversion	M₀ (N·m)^a	M (σ_M)^b	Z_hyp (km)^c	Strike (°)	Dip (°)	Rake (°)	L (km)	W (km)
2017 offshore Chiapas	Melgar et al.(2018a)	Local seismic data; crustal displacement; tsunami	2.82 × 10²¹	8.27 (0.06)	45.9	310	79	−93	160	80
2017 Puebla	Melgar et al.(2018b)	Local seismic data;crustal displacement	6.51 × 10¹⁹	7.18 (0.18)	57.5	299	44	−83	3968	12.517.5
2018 Oaxaca	Li et al. (2020)	Crustal displacement	6.98 × 10¹⁹	7.20 (0.09)	21.9	288	14	71	83	34
									50	42

Seismic moment from CMT catalog; ^bmoment magnitude and magnitude uncertainty; ^chypocentral depth.

The standard deviations on moment magnitude provided in Table 2 are a combination (square root of sum of squares) of magnitude uncertainty from alternate seismic moments (from different finite-fault models) and seismic moment uncertainty associated with the inversion performed in a given model. Details of these calculations are omitted here for brevity but are provided by Contreras (2022).

Ground motions from post-NGA-Sub events

Earthquake ground motions in Mexico from the three large-magnitude events described in the previous subsection were obtained from the RAII-UNAM network (approximately 100 accelerometer stations across the southern part of Mexico), the RACM network (81 accelerometer stations in Mexico City), and a series of smaller networks for which data are disseminated by SSN (broadband seismic network IG, Pérez et al., 2018; SSN, 2020; Valley of Mexico seismic network VM, Quintanar et al., 2018; and Veracruz seismic network UV; Córdoba-Montiel et al., 2018; each instrumented site within these networks has both an accelerometer and broadband seismometer). Locations of sensors from these networks in the vicinity of the three added events are shown in Figure 2.

Data from the RAII-UNAM network were downloaded in acceleration units with instrument corrections already having been applied. Data from the RACM network, following network approval for download, were similarly in acceleration units and included instrument corrections. Data from the three networks disseminated by SSN were downloaded as raw data (counts) and instrument corrections were applied as part of the present work (details in Contreras et al., 2023a). Additional information on the websites used to access the data is provided in Data and Resources.

Table 3 summarizes the numbers of ground-motion recordings that were collected from the different networks for each of the three considered events. In total, 585 three-component acceleration recordings from 219 ground-motion stations were compiled. In terms of number of recordings, 94% of the data (551 recordings) are from the RAII-UNAM, RACM, and IG networks, whereas the VM and UV networks contribute the remaining 6% (34 recordings). Approximately 40% of the stations are located in Mexico City, mainly from the RACM and VM networks, which produces clusters of the collected data within a narrow range of rupture distances. More than 190 recordings are available for each event.

Table 3.

Numbers of ground-motion recordings by earthquake and network

Earthquake	Date	M	RAII	RACM	IG	VM	UV	Total
Chiapas	9/8/2017	8.27	71	61	49	10	2	193
Puebla	9/19/2017	7.18	76	61	51	9	1	198
Oaxaca	2/16/2018	7.2	68	60	54	11	1	194
Total			215	182	154	30	4	585

For all 585 three-component ground motions, data processing and computation of IMs was performed following the methodology applied in the NGA-Sub project (Kishida et al., 2020). These procedures, which are well described elsewhere, produce a high-pass corner frequency f_cHP for each component that is selected in consideration of maintaining adequate signal-to-noise ratio and avoiding artifacts when ground motions are integrated to displacement. As needed, low-pass filtering was also performed, generally to remove high-frequency features evident in Fourier or response spectra. The computer code that was used to process the ground motions in the NGA-Sub project (Bozorgnia et al., 2022) was not available for public distribution at the time this work was performed. Accordingly, we developed an equivalent code in MATLAB. An example application of the procedure utilizing the MATLAB code is presented by Contreras et al. (2023a), which also describes validation to demonstrate that the NGA code (programmed in R) and the MATLAB code produce equivalent results. Ground-motion IMs computed from the processed ground motions, as well as component-specific corner frequencies, are provided in a ground-motion flatfile (Contreras et al., 2024).

Figure 3 shows median-component (RotD50; Boore, 2010) PGA values versus rupture distance for the offshore Chiapas event (similar plots for the other events are provided in Contreras et al., 2023a). The figure highlights data distributions from different networks. Modest flattening of the PGA trend with distance is evident for R_rup < ∼200 km, which is a commonly observed effect that is thought to be caused by finite-fault effects—namely the closest distance to the fault may not produce the slip that controls the ground-motion amplitude (Atkinson et al., 2016). Starting at about 200 km distance and extending to approximately 400 km, from visual inspection PGA appears to decay nearly linearly in log-log space, as expected from geometric spreading. At distances > 400 km, PGA decays more rapidly with distance due to anelastic attenuation (Boore, 2003). Finally, at relatively great distances beginning at about 1000–1200 km, the PGA-distance trend flattens. We interpret this flattening, which occurs at very low amplitudes ( $< ~ 3 \times 10^{- 4} g$ ), as a consequence of biased sampling of the ground-motion distribution at large distances. To avoid these problems, event-specific cutoff distances (i.e. the maximum distance to be considered in ground-motion analyses) were taken as 1000 km for Chiapas, 630 km for Puebla, and 650 km for Oaxaca (details in Contreras et al., 2023a).

Figure 3.

Recorded PGA values for the 2017 M8.3 offshore Chiapas earthquake from different networks: (a) RAII-UNAM, (b) IG and UV, (c) RACM, and (d) VM. RACM and VM consist of stations in Mexico City only.

An additional feature of the data that is unique to Mexico is the extraordinary impact of site response in Mexico City. This appears in the plots as tall “stripes” of data within a narrow distance range, centered on approximately 600 km for Chiapas. This occurs both because of the highly variable site response within the city and the large numbers of recordings that were made, sampling a wide array of geotechnical conditions.

Path parameters

Path parameters compiled for ground-motion analysis are rupture distance (R_rup) and percentage of path through the forearc and backarc, which is defined by a volcanic-arc boundary. The development of these parameters for the NGA-Sub database is described in Contreras et al. (2022). These same procedures were applied for Mexico, with some revisions as described here.

Rupture distance and additional source-site path parameters were computed using station locations and finite-fault models from literature for the three large 2017–2018 events. When finite-fault models were not available, a simulation procedure was used to estimate the finite-fault rupture surface (Chiou and Youngs, 2008; Contreras et al., 2022). The simulation procedure was used for the M < 6 events and for most of the CAM events included in the NGA-Sub database.

As with other subduction zone plate boundaries, Mexico has a volcanic arc. In NGA-Sub, volcanic-arc locations were delineated by drawing (by hand) lines connecting volcanoes as provided by the Smithsonian Institute’s Global Volcanism Program (2013). These lines were used to categorize the forearc (trench-side) and backarc of each subduction zone region. Figure 4 (red line) shows this line in the CAM region. We reconsidered the location of the volcanic front, recognizing that regional volcanoes are located not on a lineament but across a zone, which is referred to as the trans-Mexican volcanic belt (TMVB) (Ferrari et al., 2012) as shown in Figure 4. The TMVB is about 1000 km long in the west-east direction, spanning from the Gulf of California to the Gulf of Mexico, and ranges in width from 90 to 230 km. The TMVB contains 24 stratovolcanoes and calderas, about half of which are near the TMVB’s southern margin. We relocated the forearc-backarc boundary to the southern edge of the TMVB, as shown in Figure 4, so that the backarc would include the full extent of the TMVB region. As a result, some stations that had been classified as forearc in NGA-Sub (e.g. City of Puebla) are now backarc. With both the current and previous boundaries, Mexico City is located in the backarc region.

Figure 4.

Map showing volcanic boundary applied in NGA-Sub (red line), trans-Mexican volcanic belt (orange area), and relocated volcanic front applied for modeling purposes in this study (purple line).

For each source-to-site path, the percentage of the direct-line path in forearc and backarc was computed using the relocated volcanic front boundary in Figure 4. We find that 55.0% of paths are forearc-only, 42.2% originate in the forearc and end (at the station) in backarc, 2.1% originate in the forearc and end outside the volcanic-arc region, 0.4% originate in backarc and end in forearc, and 0.3% are backarc-only. Path data are provided for each record in the project path table (Contreras et al., 2024).

Site parameters

Site parameters for the locations of ground-motion recording stations have been developed for all stations that produced usable ground motions. These parameters update site information that was included in the NGA-Sub database. The procedures used to assign site parameters are conceptually the same as those used in NGA-Sub (Ahdi et al., 2022), but more information on site conditions has been collected in the present work.

Information on geotechnical conditions is much more available in Mexico City than in other parts of CAM. Accordingly, different approaches for assigning site parameters were applied. Site conditions in Mexico City are presented by Contreras et al. (2025). Here, we address site parameters assignments for the remainder of Mexico.

We searched for V_S profile data for sites outside of the Valley of Mexico from literature, dissertations, and consulting reports (drawing on personal contacts). This effort was unsuccessful. No V_S profiles outside of the Valley of Mexico were found, even from Oaxaca and Puebla, which have sites located on relatively soft sediments. Given the lack of V_S data, we apply proxy models for V_S₃₀ estimation that were developed for California. The procedure involved looking up site surface geology, relating the mapped geology to geologic categories used in California (Wills et al., 2015) and then assigning the California-based median value ( $μ_{lnV}$ ) and natural log standard deviation ( $σ_{lnV}$ ). The Mexico geology maps used in this process were 1:250,000 scale maps by INEGI (1994) and INEGI-SGM (2005), with the exception of a 1:50,000 scale map used for Puebla (INEGI, 1985). For sites that were used in NGA-Sub, a similar procedure had been utilized, but geomorphic terrain (Iwahashi and Pike, 2007) was used instead of surface geology and the proxy model was that of Yong (2016). We updated those assignments for all sites located in Mexico using geology-based estimates. V_S₃₀ assignments remain as terrain-based estimates for CAM sites outside of Mexico (Costa Rica, Nicaragua, El Salvador, Guatemala, and Honduras). This use of a proxy relationship derived for one region and applied in another region without local validation is termed Approach III by Ahdi et al. (2022). Approach III carries with it epistemic uncertainty in the mean estimates, which is taken as $σ_{ep}$ = 0.2. Site data are provided for each station in a site table provided by Contreras et al. (2024).

Adaptation of global GMM for CAM region

The NGA-Sub global GMMs were developed with regional adjustment factors applicable to constant terms, anelastic attenuation coefficients, magnitude-scaling breakpoints (separating strong scaling below the break from weaker scaling above the break), and site response (V_S₃₀ scaling). Levels of empirical constraint for these regional adjustment factors depend on the amount and quality of regional data. Where the data are limited in quantity and quality, such that regional factors are poorly resolved, regional coefficients are set using the global average.

In the case of CAM, data quantity in NGA-Sub was limited due to the relatively small number of recordings per event (average of 9), although not in the number of events (82 events). Data quality was also problematic, with highly uncertain site parameters and most events lacking finite-fault models. For these reasons, regionalization of NGA-Sub GMMs for CAM used a combination of globally and locally constrained components (e.g. Pea22 applied global coefficients for site response and constant terms (slab events), whereas regional coefficients were derived for anelastic attenuation and constant terms (interface events)). The data developed in this study, particularly the well-recorded 2017–2018 events, expand the size and quality of the CAM data. Here, we use the updated data to examine the regional performance of the Pea22 GMM. Specific objectives are to evaluate whether regional coefficients require adjustment to fit the data, to evaluate whether differences between backarc and forearc anelastic attenuation are evident from the data, and to investigate regional site response characteristics. Contreras et al. (2023a) present similar analyses for the Kuehn et al. (2020) model, which is omitted here for brevity. The Abrahamson and Gülerce (2022) model was not considered in our residual analyses because it was not available at the time the analyses were performed.

Approach

Data selection criteria

The combined database for CAM has three components: (1) events in the NGA-Sub database, which are all M ≥ 6, occurred prior to 2016, and have assigned event types; (2) M < 6 events with ground-motion data in the NGA-Sub database, but for which event types were only assigned in this work; and (3) three large and well-recorded events since 2016. This combined database contains 2071 three-component time series from 144 earthquakes in the CAM region. The magnitude–distance distribution of the combined data set is shown in Figure 1. The combined database was assembled into a project flatfile, available from Contreras et al. (2024).

Records from the project flatfile were selected for use in ground-motion analyses in a manner consistent with that of Pea22, as follows:

Metadata necessary for model development is available, including M, rupture distance (R_rup), hypocentral depth (Z_hyp), and V_S₃₀;

Earthquake classified with high confidence as being interface or intraslab;

Earthquake is a mainshock (Class 1; C1) rather than an aftershock (Class 2; C2) according to the Wooddell (2018) method 2 using an 80 km cutoff distance;

R _rup ≤ R_max, where R_max is a maximum distance for a given seismic network and event.

Sensor depth ≤ 2 m;

Interface events with hypocentral depths (Z_hyp) ≤ 55 km and intraslab events with Z_hyp ≤ 200 km;

PSA at $T \leq T_{LU}$ , where $T_{LU} = 1 / (1.25 f_{cHP})$ ;

Earthquakes without multiple event flags; these are events for which the recordings do not indicate that more than one seismic source affected the ground motions;

Earthquakes with source review flags = 0, 1, 2, or 4, which indicate earthquakes that underwent quality control checks and meet metadata quality standards;

Records that capture the start of the P-wave (i.e. those without a late P-wave trigger flag);

After applying criteria 1–10, we only used records from events having at least three recordings.

A criterion from Pea22 that is not applied here is to require that the earthquake epicenter and stations are both located in the forearc region. This criterion was not applied to allow for analyses of potential differences in anelastic attenuation in forearc and backarc regions.

The numbers of events and recordings used for model development vary as a function of period due to Criterion 7, with a range of 329–758 records for combined data from both event types. Figure 5 shows 369 usable interface records in the screened database at rupture distances of 14–1270 km from events with M5.3–M8.0. Of the 369 records, 204 were in the NGA-Sub database presented by Mazzoni et al. (2022). For intraslab events, there are 460 records following screening from events with M5.3–M8.3 and rupture distances of 47–1430 km; 109 of these records were in the Mazzoni et al. (2022) data set.

Figure 5.

Magnitude–distance distribution of the screened NGA-Sub database in CAM.

Residual analysis procedures

Using the screened data, residual analyses were performed to examine various attributes of CAM ground motions, including:

Constant term: Allows for an assessment of the degree to which the data are systematically higher or lower than those for a global NGA-Sub model. The constant term is a regionalized attribute in NGA-Sub models.

Source parameter scaling: Check of whether the CAM data are consistent with magnitude- and source depth-scaling relations in an NGA-Sub model. These source scaling relations are not regionalized in NGA-Sub models.

Distance attenuation: GMMs have terms for geometric spreading and anelastic attenuation, the latter of which are regionalized in NGA-Sub models. We investigate CAM-specific regional attributes of anelastic attenuation, both in the forearc and backarc, and for interface and intraslab events.

Site response: The scaling of site response with V_S₃₀ is a regionalized feature in NGA-Sub models. For CAM, V_S₃₀-scaling terms were regionalized by Kuehn et al. (2023) and Abrahamson and Gülerce (2022) but not by Parker and Stewart (2022). We examine the need for regionalization using the larger data set exclusive of Mexico City, which is a special case that is discussed by Contreras et al. (2025).

Consider an earthquake event $i$ that produces ground motion at site $j$ . Ground-motion IMs (e.g. PGA, PGV, and PSA for a range of T) can be computed for each site, which are denoted $Y_{ij}$ in arithmetic units. We compute the total residual, $δ^{v}$ , as the difference between the natural log of $Y_{ij}$ and a model prediction:

δ_{ij}^{v} = \ln (Y_{ij}) - [μ_{\ln, ij}^{r} (M_{i}, F_{S}, {(R_{rup})}_{ij}) + F_{V, j}]

(1)

where $μ_{\ln, ij}^{r}$ is the mean ground-motion prediction for reference-rock site conditions in natural log units from a GMM. We use the Pea22 GMM with the arguments of moment magnitude (M), event type parameter (F_S), and rupture distance R_rup. F_V is a site amplification model conditioned on V_S₃₀:

F_{V} = F_{lin} + F_{nl}

(2)

where F_lin and F_nl are linear and nonlinear site amplification models, taken as the global model of Parker and Stewart (2022). The use of superscript v on the residual in Equation 1 is to indicate that a V_S30-based site amplification model is considered in their derivation.

To quantify systematic event and site misfits (referred to as event terms and site terms, respectively), we partition total residuals using mixed-effects analyses (Gelman et al., 2014):

δ_{ij}^{v} = c_{0} + η_{E, i} + δ W_{ij}^{v}

(3)

where c₀ is an overall model bias, $η_{E, i}$ is the event term, and $δ W_{ij}^{v}$ is the within-event residual. $δ W_{ij}^{v}$ contains information on misfit of path and site models, which can be further partitioned as,

δ W_{ij}^{v} = c_{1} + η_{S, j}^{v} + ε_{ij}

(4)

where $η_{S}^{v}$ is the site term and $ε_{ij}$ is the remaining residual. The mixed-effects analyses are performed in MATLAB using the fitlme command (documentation available at https://www.mathworks.com/help/stats/linear-mixed-effects-models.html).

If non-zero constant terms or adjusted path coefficients were needed to fit the data, these modifications were made and the GMMs updated. This required re-computation of residuals, event terms, and site terms, to confirm that misfits were removed. In this way, the residual analyses were iterative.

Once the iterations related to path coefficients were completed, site response was examined by computing site terms relative to the reference condition in the adjusted GMM:

η_{S, j}^{r} = c_{1} + η_{S, j}^{v} + {(F_{V})}_{ij}

(5)

Superscript r indicates the term is relative to the reference-rock condition of 760 m/s. Reference-rock site terms $η_{S}^{r}$ are not expected to be mean zero, because they represent the difference between data for site $j$ and model predictions for a reference-rock condition. In aggregate, these residuals estimate site response using a non-reference site approach (Field and Jacob, 1995). Plots of $η_{S}^{r}$ with V_S₃₀ illustrate V_S₃₀ scaling, which can be checked against models. Once site response models are developed, the linear portion of the F_V model in Equation 1 can be updated, and all subsequent analyses repeated to ensure that the modified model components are compatible.

Regional path effects

Initial residual analyses using the CAM data with the Pea22 GMM revealed bias terms $c_{0}$ that were generally not distinct from zero (their 95% confidence intervals include zero), no clear trend of event terms with source parameters, and the presence of different path effects in forearc and backarc regions that are not captured in the Pea22 GMM. Accordingly, we focus on path issues initially, including the development of adjustments to the Pea22 model, and then evaluate other model components using the adjusted GMM.

The Pea22 path model uses the following equations:

F_{P} = c_{1} lnR + b_{4} M \ln (R / R_{ref}) + a_{0} R

(6)

R = \sqrt{R_{rup}^{2} + h^{2}}

(7)

The $F_{P}$ term is in natural log units and has negative values. It is additive with source and site response terms in the GMM. In Equation 6, the $lnR$ terms model geometric spreading and are region-independent. The $a_{0} R$ term models anelastic attenuation, which strengthens as $a_{0}$ decreases (becomes more negative). The anelastic term ( $a_{0}$ ) was regionalized in Pea22 (including for CAM) using data from forearc regions (values are not provided for backarc regions). The use of $R$ (Equation 7) in place of $R_{rup}$ is to handle near-source saturation effects, which are not addressed here.

We investigate whether the Pea22 path models are able to capture data trends with distance for interface and intraslab events. We also apply the models to backarc data to investigate potentially faster anelastic attenuation relative to the forearc. Preliminary analyses of Pea22 path model performance examine the trends of $δ W_{ij}^{v}$ (Equation 3) against $R_{rup}$ (these analyses are preliminary because VM sites are excluded and Pea22 is applied without modification to backarc sites). Beginning with forearc sites, Figures 6 and 7 show forearc within-event residuals and their binned means for interface and intraslab earthquakes, respectively. In both cases, no appreciable slope is evident, indicating that the Pea22 model is capturing the data trends. Note that VM data are in the backarc and, therefore, do not affect these plots.

Figure 6.

Trend of Pea22 within-event residuals for PGA, PGV, 0.3 s PSA, and 3.0 s PSA with $R_{rup}$ for interface events and forearc sites.

Figure 7.

Trend of Pea22 within-event residuals for PGA, PGV, 0.3 s PSA, and 3.0 s PSA with $R_{rup}$ for intraslab events and forearc sites.

We anticipate backarc regions to have different rates of anelastic attenuation based on findings elsewhere, mainly Japan (Ghofrani and Atkinson, 2011). Beginning with the null hypothesis that forearc attenuation rates apply in the backarc, we plot in Figures 8 and 9 $δ W_{ij}^{v}$ versus the backarc portion of the rupture distance ( $R_{rup, b}$ ), which is defined as,

R_{rup, b} = F_{b} R_{rup}

(8)

where $F_{b}$ is the fraction of the rupture distance in the backarc. Because the anelastic attenuation effect operates on $R$ and not $lnR$ (Equation 6), in the figures a linear distance scale is used for $R_{rup, b}$ . The results indicate for high-frequency IMs an approximately linear decay of within-event residuals with backarc distance. This suggests that a lower (more negative) anelastic coefficient is needed for the backarc.

Figure 8.

Trend of Pea22 within-event residuals for PGA, PGV, 0.3 s PSA, and 3.0 s PSA with $R_{rup, b}$ for interface events and backarc sites along with fit of backarc model (Equation 9) and its confidence intervals.

Figure 9.

Trend of Pea22 within-event residuals for PGA, PGV, 0.3 s PSA, and 3.0 s PSA with $R_{rup, b}$ for intraslab events and backarc sites along with fit of backarc model (Equation 9) and its confidence intervals.

To remove the faster backarc anelastic attenuation trend, we fit a linear trend line as follows:

δ W_{ij}^{v} = a + a_{0 b} R_{rup, b}

(9)

where $a_{0 b}$ is additional anelastic attenuation required in the backarc. When used with the predictive model, would appear as an additive term,

F_{P} = c_{1} lnR + b_{4} M \ln (R / R_{ref}) + a_{0} R + a_{0 b} R_{rup, b}

(10)

Note that by using $R_{rup, b}$ instead of rupture distance combined with finite-fault parameter h as in Equation 7, an assumption is being made that $R_{rup} >> h$ , which is generally valid for backarc sites when the earthquake is located in the forearc. The intercept parameter a in Equation 9 is not used for forward modeling and is generally small.

Example fits of Equation 9 to the data are shown by the blue lines in Figures 8 and 9. Figure 10 plots $a_{0 b}$ for different spectral periods (along with PGA and PGV) for interface and intraslab events. The results show faster attenuation for interface than intraslab events. For both event types, the stronger backarc scaling reduces at long periods. The faster scaling disappears for T > 3 s for intraslab events but is present for all periods for interface events. Table A1 provides coefficients $a_{0 b}$ for all of the periods considered in the Pea22 model.

Figure 10.

Backarc additional anelastic attenuation coefficient $a_{0 b}$ for interface and intraslab events. Range indicates ±1 standard error of the mean estimate.

Regional site response

Residuals were re-computed using the Pea22 GMM modified to include backarc path effects (Equation 10 with coefficients in Figure 10). We assume site response is independent of event type (justification provided by Parker and Stewart, 2022), which allows the combined use of $δ W_{ij}^{v}$ values from interface and intraslab events. For example, sites in Oaxaca have 2 and 1 recordings from intraslab and interface events, respectively. After excluding Valley of Mexico (VM) sites, the analyses consider 265 stations as shown in Figure 4. About half (133) of the stations are in Mexico, with the remainder located elsewhere in CAM. The database contains 592 three-component ground motions, 380 of which (∼ 64%) are from the 133 Mexico sites.

Using the combined data set exclusive of VM sites, Equation 4 is used to partition $δ W_{ij}^{v}$ into site terms $η_{S, j}^{v}$ and remaining residuals $ε_{ij}$ . Constant terms ( $c_{1}$ ) were found to be nearly zero, which is expected because the mean of $δ W_{ij}^{v}$ is necessarily zero. Site terms $η_{S, j}^{v}$ represent the departure of observed site responses from the global V_S₃₀-scaling model of Parker and Stewart (2022). As such, model performance is evaluated based on trends of $η_{S}^{v}$ with V_S₃₀ as shown in Figure 11 for PGA, PGV, 0.3 s PSA, and 3.0 s PSA. We find the binned means to have no appreciable trend with V_S₃₀. As a result, we conclude that the global V_S30-scaling model is suitable for modeling ergodic site response for sites outside of VM.

Figure 11.

Trend of site terms for PGA, PGV, 0.3 s PSA, and 3.0 s PSA with V_S₃₀ for non-VM CAM sites. Puebla and Oaxaca site terms are highlighted due to soft soils in those regions.

As described in the Site Parameters subsection, Puebla and Oaxaca have soft soil deposits where strong site response effects may be anticipated. These sites occupy V_S₃₀ ranges of 230–700 m/s, as derived from California proxy relationships. As shown in Figure 11, the soft sites in Puebla and Oaxaca (V_S₃₀ < 300 m/s) have site responses that depart substantially from the global ergodic model, including larger low-period amplification (PGA and PSA(0.3 s)) and smaller long-period amplification (PSA(3.0 s)). These patterns of amplification are expected for soft soil deposits of modest depth, although sediment depths in these two cities are currently unknown. Site response for VM sites is evaluated by Contreras et al. (2025).

Regional constants and checks of source model

The residuals analyses performed in this subsection use the Pea22 GMM with two adjustments: (1) backarc path model (Equation 10) and (2) application of a local VM site response model as presented by Contreras et al. (2025) for site response in that region. Residuals analyses performed with those modifications produce the constant terms ( $c_{0}$ ) as shown in Figure 12. Interface events have a small short-period underprediction bias and long-period overprediction bias, although the confidence intervals on $c_{0}$ include zero, and hence, the biases are not statistically significant. For intraslab events, bias is small (absolute values <0.2) for T < 1 s but increase at longer periods. Had the Pea22 model been used without modification, the $c_{0}$ terms would be reduced at short periods, causing the bias trend with period to be much stronger.

Figure 12.

Pea22 GMM model bias for peak acceleration, peak velocity, and PSA for a range of periods: (a) interface events and (b) intraslab events. Range indicates 95% confidence interval of the mean bias.

Figures 13 and 14 show the variation of modified Pea22 event terms ( $η_{E}$ ) with magnitude for the IMs of PGA, PGV, 0.3 s PSA, and 3.0 s PSA for interface and intraslab events. For both event types, the event terms do not exhibit any appreciable trend up to M7.5, which is near the break magnitude $m_{c}$ . For both the interface and intraslab results, the lack of bias for M < 6 events is notable because data in that M range from CAM were not considered in the development of the Pea22 model. For larger magnitudes, no trend is evident for interface events, but positive bias is seen for the M8.3 Chiapas event, which is the largest intraslab event in the NGA-Sub database inclusive of the data extensions in this article. This could indicate that a steeper slope is needed for some IMs for $M > m_{c}$ , although we defer such decisions to future work where other large M events that have occurred globally subsequent to 2016 can be considered. Although not shown here for brevity, plots of event terms with hypocentral depth exhibit no trends, suggesting that the depth models (or lack thereof) do not require modification for CAM applications.

Figure 13.

Trends of modified Pea22 event terms for PGA, PGV, 0.3 s PSA, and 3.0 s PSA with magnitude for interface events. Error bars are 95% confidence intervals. Gray area indicates ±1.0 $τ$ from Pea22 model.

Figure 14.

Trends of modified Pea22 event terms for PGA, PGV, 0.3 s PSA, and 3.0 s PSA with magnitude for intraslab events. Error bars are 95% confidence intervals. Gray area indicates ±1.0 $τ$ from Pea22 model.

Epistemic uncertainty

Pea22 estimated the epistemic uncertainty in their GMM using the uncertainty in the constant terms. For regional models, these uncertainties scale with the inverse of the square root of the number of observations (recordings). Since the number of recordings has increased significantly from 103 in the Pea22 model development (after data selection criteria were applied) to 821 from the present work, a reduction in epistemic uncertainty is expected.

Figure 15 shows the epistemic uncertainty ( $σ_{ε}$ ) in the Pea22 constant terms ( $c_{0}$ ) for the CAM region. This epistemic uncertainty can be used with a scaled-backbone approach in probabilistic seismic hazard analysis (Atkinson et al., 2014). The epistemic uncertainty of the CAM model as derived in the present work is the standard error of the constant terms. Those uncertainties are shown in Figure 15, which shows that they have been lowered by approximately 30%–50%. As such, this is an example of uncertainty reduction from database expansion and improvement.

Figure 15.

Epistemic uncertainty $σ_{ε}$ for CAM from Pea22 as derived using NGA-Sub database and as found from the present analysis using the expanded database. The Pea22 epistemic uncertainties are identical for intraslab and interface events.

Summary and discussion

This research examined ground motions from subduction earthquakes in CAM. We significantly extended the amount of earthquake ground-motion information for CAM relative to what was used in NGA-Sub model development. The added information was from a M8.3 instraslab event in 2017, a M7.2 intraslab event in 2017, a M7.2 interface event in 2018, and many M < 6 events that had been considered in NGA-Sub but lacked adequate metadata to be used in model development. Metadata was added for each of these events, and ground motions were processed for the three large events in 2017–2018 following NGA protocols. The three events produced 585 usable recordings.

The expanded database was then used to evaluate regional features of GMMs for CAM. These analyses produced several important findings, most of which are related to the degree to which the current data support or require modifications to an NGA-Sub GMM. We find that the constant terms in the Pea22 GMM do not require adjustment. We find that ground-motion scaling with source parameters (magnitude and source depth) is sufficient in the GMM. However, the 2017 M8.3 intraslab event is now the largest intraslab event in the global database, and event terms for this earthquake are positive. While this apparent bias could be removed with the use of steeper magnitude scaling at large magnitudes, we are not yet confident in making such a recommendation based principally on results from one event.

We found forearc distance scaling in the Pea22 GMMs to be unbiased, but identify faster attenuation in the backarc for both intraslab and interface events. We provide a modification to the path model to account for these effects (Equation 10 and Figure 10). The faster backarc attenuation is significant at short periods (T < 2 s) and is more pronounced for interface than for intraslab events. The changes to the Pea22 model are contained in Equation 10 with the coefficients listed in Table A1. An implementation of Pea22 with these changes to the path model and the VM site response model described in Contreras et al. (2025) is provided on the NGA-Subduction project web portal (https://www.risksciences.ucla.edu/nhr3/nga-subduction).

We find that the average site responses across the CAM region are consistent with the global models of Parker and Stewart (2022). This is useful because it provides a basis for site response estimation across the region, but with “carve outs” for special areas like the VM where region-specific models should be used (Contreras et al., 2025). The soft sediments in Oaxaca and Puebla produce stronger site responses than those provided by the global model, particularly at short periods. Our ability to study these site response features is limited by a lack of site data. The overall database expansion and model refinements presented here were demonstrated to reduce epistemic uncertainties in the Pea22 GMM by about 30%–50%.

The present findings provide an opportunity to update ground-motion hazard analyses in CAM. Advantages of the present approach include the following:

The models are transparent, allowing for future research where individual components can more readily be interrogated and improved.

The CAM-specific models are calibrated for regional features (constant term, site response, anelastic attenuation, magnitude breakpoint) but are constrained by global data for other features that are generally not found to have appreciable regional variations (magnitude and source depth scaling, geometric spreading).

The GMMs include different rates of forearc and backarc anelastic attenuation, which is important for seismic hazard analyses for backarc locations such as Mexico City.

Supplemental Material

sj-csv-1-eqs-10.1177_87552930251322789 – Supplemental material for Regionalization of global subduction ground-motion model for application in Central America and Mexico

Supplemental material, sj-csv-1-eqs-10.1177_87552930251322789 for Regionalization of global subduction ground-motion model for application in Central America and Mexico by Victor Contreras, Jonathan P Stewart, Juan M Mayoral and Xyoli Pérez-Campos in Earthquake Spectra

Footnotes

Data resources

Raw ground-motion recordings from the RAII-UNAM network, managed by the Institute of Engineering at UNAM, were accessed through the RAII-UNAM network website: http://aplicaciones.iingen.unam.mx/AcelerogramasRSM/Inicio.aspx (last accessed April 2022). Raw ground motions from the RACM network managed by CIRES (Instrumentation and Seismic Recording Center) were requested from the CIRES website: http://www.cires.org.mx/racm_historico_es.php (last accessed April 2022). Our requests made in September 2017 and April 2020 were granted. Raw data from the IG, VM, and UV networks, which are distributed by the National Seismological Service of Mexico (SSN), are accessible through a client called SSNstp upon request (https://doi.org/10.21766/SSNMX/SN/MX). Our request made on 7 June 2020 was granted. SSN personnel are thanked for their work on data acquisition and distribution. The source table, path table, site table, and flatfiles used in this research are accessible at . Daniel De La Rosa assisted with data collection for sites in Mexico, particularly geologic maps. The authors are grateful to the NGA-Sub project for providing the database and models from which this project was developed.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Financial support for the first author was provided by the Chilean National Agency for Research and Development (ANID) through the grants ANID BECAS CHILE-DOCTORADO EN EL EXTRANJERO 72180625, ANID FONDECYT Iniciación 11241337, and the Civil and Environmental Engineering Department at UCLA. The views expressed herein are those of the authors and do not necessarily reflect the views of the CTBTO Preparatory Commission.

ORCID iDs

Victor Contreras

Jonathan P Stewart

Xyoli Pérez-Campos

Supplemental material

Supplemental material for this article is available online.

References

Abrahamson

Gulerce

(2022) Summary of the Abrahamson and Gulerce NGA-SUB ground-motion model for subduction earthquakes. Earthquake Spectra 38(4): 2638–2681.

Ahdi

Kwak

Ancheta

Contreras

Kishida

Kwok

Mazzoni

Ruz

Stewart

(2022) Site parameters applied in NGA-Sub database. Earthquake Spectra 38(1): 494–520.

Anderson

(1997) Nonparametric description of peak acceleration above a subduction thrust. Seismological Research Letters 68(1): 86–93.

Anderson

Lei

(1994) Nonparametric description of peak acceleration as a function of magnitude, distance, and site in Guerrero, Mexico. Bulletin of the Seismological Society of America 84(4): 1003–1017.

Anderson

Quaas

(1988) The Mexico earthquake of September 19, 1985—Effect of magnitude on the character of strong ground motion: An example from the Guerrero, Mexico strong motion network. Earthquake Spectra 4(3): 635–646.

Anderson

Brune

Prince

Quaas

Singh

Almora

Bodin

Oñate

Vásquez

Velasco

(1994) The Guerrero accelerograph network. Geofísica Internacional 33(3): 341–371.

Arroyo

García

Ordaz

Mora

Singh

(2010) Strong ground-motion relations for Mexican interplate earthquakes. Journal of Seismology 14: 769–785.

Arroyo

Ordaz

Singh

(2024) Prediction of Fourier amplitude spectrum of ground motion in Mexico City from subduction thrust earthquakes. Geofísica Internacional 62(2): 783–801.

Arteta

Pajaro

Mercado

Montejo

Arcila

Abrahamson

(2021) Ground-motion model for subduction earthquakes in northern South America. Earthquake Spectra 37(4): 2419–2452.

10.

Atkinson

Bommer

Abrahamson

(2014) Alternative approaches to modeling epistemic uncertainty in ground motions in probabilistic seismic-hazard analysis. Seismological Research Letters 85: 1141–1144.

11.

Atkinson

Yenier

Sharma

Convertito

(2016) Constraints on the near-distance saturation of ground-motion amplitudes for small-to-moderate induced earthquakes. Bulletin of the Seismological Society of America 106(5): 2104–2111.

12.

Boore

(2003) Simulation of ground motion using the stochastic method. Pure and Applied Geophysics 160: 635–675.

13.

Boore

(2010) Orientation-independent, nongeometric-mean measures of seismic intensity from two horizontal components of motion. Bulletin of the Seismological Society of America 100(4): 1830–1835.

14.

Bozorgnia

Abrahamson

Ahdi

Ancheta

Al Atik

Archuleta

Atkinson

Boore

Campbell

Chiou

BSJ

Contreras

Darragh

Derakhshan

Donahue

Gregor

Gulerce

Idriss

Kishida

Kottke

Kuehn

Kwak

Kwok

AOL

Lin

Mazzoni

Midorikawa

Muin

Parker

Rezaeian

Silva

Stewart

Walling

Wooddell

Youngs

(2022) NGA-Subduction research program. Earthquake Spectra 38(2): 783–798.

15.

Castro

Singh

Mena

(1988) The Mexico earthquake of September 19, 1985—An empirical model to predict Fourier amplitude spectra of horizontal ground motion. Earthquake Spectra 4(4): 675–685.

16.

Chiou

BS-J

Youngs

(2008) NGA Model for Average Horizontal Component of Peak Ground Motion and Response Spectra. Report no. 2008/09, November. Berkeley, CA: Pacific Earthquake Engineering Research Center (PEER), November 2008. UC Berkeley.

17.

Contreras

(2022) Characteristics of subduction zone ground motions with an emphasis on Latin America. PhD Dissertation, Civil & Environmental Engineering, University of California, Los Angeles, CA.

18.

Contreras

Stewart

De La Rosa

Mayoral

Pérez-Campos

(2023a) Regional customization of global subduction ground motion models for Mexico. Report GIRS-2023-09202309. Los Angeles, CA: Natural Hazards Risk and Resiliency Research Center, B. John Garrick Institute for the Risk Sciences, University of California.

19.

Contreras

Stewart

De La Rosa

Mayoral

Pérez-Campos

(2023b) Empirical site response of Mexico City developed from regional customization of global subduction ground motion models. In: SSA Annual Meeting, San Juan, Puerto Rico, 17–20 April.

20.

Contreras

Stewart

Kishida

Darragh

Chiou

BSJ

Mazzoni

Youngs

Kuehn

Ahdi

Wooddell

Boroschek

Rojas

Órdenes

(2022) NGA-Sub source and path database. Earthquake Spectra 38(2): 799–840.

21.

Contreras

Stewart

Mayoral

Pérez-Campos

(2024) Regional customization of global subduction ground motion models for Mexico. DesignSafe-CI. Available at: https://doi.org/10.17603/ds2-bj9h-zm65

22.

Contreras

Stewart

Mayoral

Pérez-Campos

(2025) Valley of Mexico site response from non-reference site approach. Earthquake Spectra, https://doi.org/10.1177/87552930251316816.

23.

Córdoba-Montiel

Iglesias

Pérez-Campos

Sieron

; Grupo de Trabajo del Servicio Sismológico Nacional; Valdés-González

Singh

Pacheco

(2018) The broadband seismological network of Veracruz, Mexico: Toward a regional seismotectonic interpretation. Seismological Research Letters 89(2A): 345–355.

24.

Ekström

Nettles

Dziewonski

(2012) The global CMT project 2004-2010: Centroid-moment tensors for 13,017 earthquakes. Physics of the Earth and Planetary Interiors 200–201: 1–9.

25.

Ferrari

Orozco-Esquivel

Manea

(2012) The dynamic history of the Trans-Mexican volcanic belt and the Mexico subduction zone. Tectonophysics 522–523: 122–149.

26.

Field

Jacob

(1995) A comparison and test of various site response estimation techniques, including three that are not reference site dependent. Bulletin of the Seismological Society of America 85(4): 1127–1143.

27.

García

Singh

Herráiz

Ordaz

Pacheco

(2005) Inslab earthquakes of central Mexico: Peak ground-motion parameters and response spectra. Bulletin of the Seismological Society of America 95(6): 2272–2282.

28.

García

Singh

Herraiz

Ordaz

Pacheco

Cruz-Jiménez

(2009) Influence of subduction zone structure on coastal and inland attenuation in Mexico. Geophysical Journal International 179(1): 215–230.

29.

García-Soto

Jaimes

(2017) Ground-motion prediction model for vertical response spectra from Mexican interplate earthquakes. Bulletin of the Seismological Society of America 107(2): 887–900.

30.

Gelman

Carlin

Stern

Dunson

Vehtari

Rubin

(2014) Bayesian Data Analysis. 3rd ed. Boca Raton, FL: CRC Press.

31.

Ghofrani

Atkinson

(2011) Forearc versus backarc attenuation of earthquake ground motion. Bulletin of the Seismological Society of America 101(6): 3032–3045.

32.

Hong

Pozos-Estrada

Gomez

(2009) Orientation effect on ground motion measurement for Mexican subduction earthquakes. Earthquake Engineering and Engineering Vibration 8(1): 1–16.

33.

INEGI (1985) Geologic map of Puebla and Tlaxcala. Puebla E14B43, Scale 1:50,000. Convenio Instituto Nacional De Estadística, Geografía E Informática, México (in Spanish).

34.

INEGI (1994) Geologic map of Oaxaca. Oaxaca E14-9, Scale 1:250,000. Convenio Instituto Nacional De Estadística, Geografía E Informática, México (in Spanish).

35.

INEGI-SGM (2005) Carta Geológica de la República Mexicana escala 1:250,000, años 1995-2005 Convenio Instituto Nacional De Estadística, Geografía E Informática (INEGI)—Servicio Geológico Mexicano (SGM), México (in Spanish).

36.

Iwahashi

Pike

(2007) Automated classifications of topography from DEMs by an unsupervised nested-means algorithm and a three-part geometric signature. Geomorphology 86(3/4): 409–440.

37.

Jaimes

García-Soto

(2020) Updated ground motion prediction model for Mexican intermediate-depth intraslab earthquakes including V/H ratios. Earthquake Spectra 36(3): 1298–1330.

38.

Jaimes

García -Soto

(2021) Ground-motion duration prediction model from recorded Mexican interplate and intermediate-depth intraslab earthquakes. Bulletin of the Seismological Society of America 111(1): 258–273.

39.

Jaimes

Reinoso

Ordaz

(2006) Comparison of methods to predict response spectra at instrumented sites given the magnitude and distance of an earthquake. Journal of Earthquake Engineering 10(6): 887–902.

40.

Archuleta

(2018) Scaling of PGA and PGV deduced from numerical simulations of intraslab earthquakes. Santa Barbara, CA: Department of Earth Science, University of California.

41.

Kishida

Darragh

Chiou

BSJ

Bozorgnia

Mazzoni

Contreras

Boroschek

Rojas

Stewart

(2020) Chapter 3: Ground Motions and Intensity Measures. PEER Report No. 2020/03202003. Berkeley, CA: Pacific Earthquake Engineering Research Center, UC Berkeley.

42.

Kuehn

Bozorgnia

Campbell

Gregor

(2020) Partially non-ergodic ground-motion model for subduction regions using NGA-Subduction database. PEER Report 2020/03202003. Berkeley, CA: Pacific Earthquake Engineering Research Center, UC Berkeley.

43.

Kuehn

Bozorgnia

Campbell

Gregor

(2023) A regionalized partially nonergodic ground-motion model for subduction earthquakes using the NGA-Sub database. Earthquake Spectra 39(3): 1625–1657.

44.

Lermo-Samaniego

Jaimes

Sánchez-Sesma

Campuzano-Sánchez

Cruz-Jiménez

Campos Enriquez

(2020) Ground motion prediction model for southeastern Mexico removing site effects using the earthquake horizontal-to-vertical ratio (EHVSR). Geofísica Internacional 59(4): 257–272.

45.

Shan

Zhu

Qiao

Zhao

(2020) Geodetic model of the 2018 Mw 7.2 Pinotepa, Mexico, earthquake inferred from InSAR and GPS data. Bulletin of the Seismological Society of America 110(3): 1115–1124.

46.

Macedo

Liu

(2022) A nonergodic ground motion model for Chile. Bulletin of the Seismological Society of America 112(5): 2542–2561.

47.

Mazzoni

Kishida

Stewart

Contreras

Darragh

Ancheta

Chiou

BSJ

Silva

Bozorgnia

(2022) NGA-Sub relational database. Earthquake Spectra 38(2): 1529–1548.

48.

Melgar

Pérez -Campos

Ramirez-Guzman

Spica

Espíndola

Hammond

Cabral-Cano

(2018b) Bend faulting at the edge of a flat slab: The 2017 Mw7.1 Puebla-Morelos, Mexico earthquake. Geophysical Research Letters 45: 2633–2641.

49.

Melgar

Ruiz-Angulo

Garcia

Manea

Ramirez-Herrera

Zavala-Hidalgo

Geng

Corona

Pérez -Campos

Cabral-Cano

Ramirez-Guzmán

(2018a) Deep embrittlement and complete rupture of the lithosphere during the Mw 8.2 Tehuantepec earthquake. Nature Geoscience 11: 955–960.

50.

Ordaz

Arroyo

Singh

Salgado-Gálvez

(2024) A PSHA for Mexico City based solely in Fourier-based GMM of the response spectra. Soil Dynamics and Earthquake Engineering 187: 109025.

51.

Ordaz

Jara

Singh

(1989) Riesgo sísmico y espectros de diseño en el estado de Guerrero. Report No 8782/9745. Ciudad de México: UNAM, Instituto De Ingeniería (in Spanish).

52.

Ordaz

Singh

Arciniega

(1994) Bayesian attenuation regressions: An application to Mexico City. Geophysical Journal International 117(2): 335–344.

53.

Parker

Stewart

(2022) Ergodic site response model for subduction zone regions. Earthquake Spectra 38(2): 841–864.

54.

Parker

Stewart

Boore

Atkinson

Hassani

(2022) NGA-subduction global ground motion models with regional adjustment factors. Earthquake Spectra 38(1): 456–493.

55.

Perea

Sordo

(1988) Direct response spectrum prediction including local site effects. In: Proceedings of the Eleventh European Conference on Earthquake Engineering, 6–11 September, Paris.

56.

Pérez-Campos

Espíndola

Pérez

Estrada

Monroy

Bello

González-López

González Ávila

Contreras Ruiz Esparza

Maldonado

Tan

Rodríguez Rasilla

Vela Rosas

Cruz

Cárdenas

Navarro Estrada

Hurtado

Mendoza Carvajal

Montoya-Quintanar

Pérez-Velázquez

(2018) The Mexican national seismological service: An overview. Seismological Research Letters 89(2A): 318–323.

57.

Quintanar

Cárdenas-Ramírez

Bello-Segura

Espíndola

Pérez-Santana

Cárdenas-Monroy

Carmona-Gallegos

Rodríguez-Rasilla

(2018) A seismic network for the Valley of Mexico: Present status and perspectives. Seismological Research Letters 89(2A): 356–362.

58.

Reinoso

Ordaz

(1999) Spectral ratios for Mexico City from free-field recordings. Earthquake Spectra 15(2): 273–295.

59.

Reyes

(1998) El estado límite de servicio en el diseño sísmico de edificios. PhD Dissertation, School of Engineering, National Autonomous University of Mexico (UNAM), Mexico (in Spanish).

60.

Rodríguez-Pérez

(2014) Ground-motion prediction equations for near-trench interplate and normal-faulting inslab subduction zone earthquakes in Mexico. Bulletin of the Seismological Society of America 104(1): 427–438.

61.

Rosenblueth

Ordaz

Sánchez-Sesma

Singh

(1989) The Mexico earthquake of September 19, 1985—Design Spectra for Mexico’s federal district. Earthquake Spectra 5(1): 273–291.

62.

Midorikawa

Kishida

(2022) Development of NGA-Sub ground-motion prediction equation of 5%-damped pseudo-spectral acceleration based on database of subduction earthquakes in Japan. Earthquake Spectra 38(4): 2682–2706.

63.

Singh

Mena

Castro

Carmona

(1987) Empirical prediction of ground motion in Mexico City from coastal earthquakes. Bulletin of the Seismological Society of America 77(5): 1862–1867.

64.

Smithsonian Institute’s Global Volcanism Program (2013) National Museum of Natural History, Smithsonian Institution. Available at: https://volcano.si.edu/

65.

SSN (2020) Servicio Sismológico Nacional, Instituto de Geofísica, Universidad Nacional Autónoma de México, México. Available at: https://doi.org/10.21766/SSNMX/SN/MX

66.

Wills

Gutierrez

Perez

Branum

(2015) A next generation VS30 map for California based on geology and topography. Bulletin of the Seismological Society of America 105(6): 3083–3091.

67.

Wooddell

(2018) New earthquake classification for the NGA-subduction project. In: Proceedings of the 11th National Conference on Earthquake Engineering (11NCEE), Los Angeles, CA, 25–29 June.

68.

Yong

(2016) Comparison of measured and proxy-based VS30 values in California. Earthquake Spectra 32(1): 171–192.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB