Comparison of observed ground-motion attenuation for the 2016/04/16 Mw7.8 Ecuador megathrust earthquake and its two largest aftershocks with existing ground-motion prediction equations

A megathrust subduction earthquake (Mw7.8) struck the coast of Ecuador on April 16th, 2016 at 23h58 UTC. This earthquake is one of the best-recorded megathrust events up to date. Besides the mainshock, two large aftershocks have been recorded on May 18th, 2016, at 7h57 (Mw 6.7) and 16h46 (Mw6.9). These data make a significant contribution for understanding the attenuation of ground motions in Ecuador. Peak ground accelerations and spectral accelerations are compared with four ground-motion prediction equations developed for interface earthquakes, the global Abrahamson et al. (2016) model, the Japanese equations Zhao et al. (2006) and Ghofrani and Atkinson (2014), and one Chilean equation Montalva et al. (2016). The four tested GMPEs are providing rather close predictions for the mainshock at distances up to 200km. However, our results show that high-frequency attenuation is greater for backarc sites, thus Zetal2016 and Metal2016, which are not taking into account this difference, are not considered further. Residual analyses show that G&A14 and Aetal2016 are well predicting the attenuation of ground motions for the mainshock. Comparisons of aftershock observations with Aetal2016 predictions indicate that the GMPE provide reasonable fit to the attenuation rates observed. The event terms of the Mw6.7 and Mw6.9 events are positive but within the expected scatter from worldwide similar earthquakes. The intra-event standard deviations are higher than the intraevent variability of the model, which is partly related to the poorly constrained VS30 proxys.

💡 Research Summary

The 2016 Pedernales megathrust earthquake (Mw 7.8) that struck the Ecuadorian coast on 16 April 2016, together with two large aftershocks (Mw 6.7 and Mw 6.9 on 18 May 2016), generated an exceptionally rich strong‑motion dataset. A total of 69 accelerometric stations from the national RENAC network recorded the events at rupture distances ranging from 26 km to 427 km, with roughly equal numbers of stations located in the fore‑arc (west of the volcanic front) and the back‑arc (east of the volcanic front). The authors processed the raw accelerograms using visual inspection, baseline correction, and signal‑to‑noise ratio checks, then computed 5 % damped response spectra and geometric‑mean PGA values for periods up to 2 s.

Because site effects dominate ground‑motion amplitudes, the study estimated the average shear‑wave velocity in the top 30 m (VS30) for each station using two independent approaches: (1) direct geophysical measurements or H/V spectral ratio‑derived natural periods (the “reference” VS30 set) and (2) a topographic‑slope proxy based on the Wald‑Allen relationship (the “topographic” VS30 set). The two VS30 sets differ by up to 600 m s⁻¹ in the Andean Cordillera, but the authors demonstrate that this uncertainty does not materially affect the comparison with prediction models.

Four ground‑motion prediction equations (GMPEs) developed for subduction‑interface earthquakes were selected for testing: (i) the global Abrahamson et al. (2016) model, (ii) the Japanese Ghofrani & Atkinson (2014) model, (iii) the Japanese Zhao et al. (2006) model, and (iv) the Chilean Montalva et al. (2016) model. All four predict median motions as a function of magnitude, rupture distance, and VS30, and provide total, intra‑event, and inter‑event standard deviations. Notably, Zhao et al. and Montalva et al. do not distinguish between fore‑arc and back‑arc attenuation, whereas Abrahamson et al. and Ghofrani & Atkinson explicitly incorporate separate anelastic attenuation terms for the two regions.

Initial attenuation plots showed that all four GMPEs give comparable median predictions for distances ≤ 200 km. However, at higher frequencies (≥ 1 Hz) the observed motions in the back‑arc attenuate more rapidly than the Zhao and Montalva models predict. Consequently, these two models were excluded from further quantitative analysis.

Residual analyses (observed – predicted, normalized by the model sigma) revealed that Abrahamson et al. and Ghofrani & Atkinson reproduce the main‑shock data very well. When the Abrahamson model is run with the “forearc/unknown” option, it matches observations up to ~130 km; beyond that distance, the back‑arc option yields a steeper decay that aligns with the measured amplitudes at back‑arc stations. The aftershocks, recorded by 61 and 64 stations respectively, were also compared to Abrahamson et al. predictions. The model’s median curves fit the observed attenuation, and the event terms (the average residual for each earthquake) are positive but fall within the expected scatter derived from worldwide subduction events.

A notable finding is that the intra‑event standard deviations derived from the Ecuador data are larger than the sigma values prescribed by the GMPEs. The authors attribute this excess variability primarily to the uncertainty in VS30 proxies, the lack of site‑specific non‑linearity information, and the presence of directivity effects that are not captured by any of the tested equations. Indeed, the Pedernales rupture propagated southward, producing higher ground motions to the south of the fault trace—a classic directivity signature.

The paper concludes that, for the Ecuadorian subduction environment, GMPEs that differentiate fore‑arc and back‑arc attenuation (Abrahamson et al. 2016 and Ghofrani & Atkinson 2014) provide the most reliable predictions for both the mainshock and its large aftershocks. Nevertheless, the study highlights several limitations that must be addressed for robust probabilistic seismic hazard assessments (PSHA): (1) the need for a more accurate, station‑specific VS30 database, ideally obtained through MASW surveys, borehole measurements, or detailed geotechnical investigations; (2) incorporation of directivity and possible non‑linear site response into regional GMPEs; and (3) expansion of the strong‑motion interface database for South America to enable likelihood‑based or Bayesian model testing.

Overall, the 2016 Ecuador megathrust sequence offers a valuable benchmark for evaluating and refining subduction‑interface ground‑motion models, and the authors’ systematic comparison underscores the importance of regional site effects and back‑arc attenuation in seismic hazard modeling.