Alexandra Carvalho

This study develops an artificial neural network (ANN)-based ground motion model (GMM) for the Azores Plateau (Portugal) using a dataset generated through stochastic finite-fault simulations. The simulations are performed for both onshore and offshore rock-site scenarios, employing a dynamic corner-frequency algorithm. Randomized source and path parameters are incorporated to capture the aleatory variability of regional seismicity. The simulated ground motions are validated through a comprehensive statistical framework, confirming that the implemented randomization reproduces realistic variance and inter-period correlations observed in recorded data. The ANN-based GMM is trained using the simulated database to predict spectral acceleration across a wide range of magnitudes and source-to-site distances. The developed model and accompanying dataset together provide a reliable foundation for seismic hazard and risk assessments in the Azores Plateau region.Keywords: Artificial neural network (ANN); Ground motion model (GMM); Stochastic finite-fault simulation; Onshore and offshore scenarios; Spectral acceleration prediction; Azores Plateau (Portugal).Acknowledgments:This work is financed by national funds through FCT – Foundation for Science and Technology, under grant agreement [2023.08982.CEECIND/CP2841/CT0033] attributed to the first author (https://doi.org/10.54499/2023.08982.CEECIND/CP2841/CT0033). This work was also supported by FCT/ Ministério da Ciência, Tecnologia e Ensino Superior (MCTES) under the R&D Unit Institute for Sustainability and Innovation in Structural Engineering (ISISE), under the references UID/4029/2025 (https://doi.org/10.54499/UID/04029/2025) and UID/PRR/04029/2025 (https://doi.org/10.54499/UID/PRR/04029/2025), and under the Associate Laboratory Advanced Production and Intelligent Systems (ARISE) under reference LA/P/0112/2020. This work is partly financed by national funds through FCT (Foundation for Science and Technology), under grant agreement [UI/BD/153379/2022] attributed to the second author. This work is partly financed by national funds through FCT – Foundation for Science and Technology, under grant agreement [2023.01101.BD] attributed to the third author.

The increased deployment of subsea infrastructures and the exploration of marine resources have heightened the need to assess seismic hazard on the seabed, especially in tectonically active offshore regions. The area between the Gulf of Cádiz and the Alboran Sea, which is rich in coastal and submarine assets, is located within the Ibero-Maghrebian Region (IMR) a seismically active region. While previous research has addressed seismic hazard in adjacent inland areas using deterministic and probabilistic approaches, few studies have focused on offshore zones. Moreover, existing models often overlook the amplification effects introduced by bathymetry and seafloor conditions, and they rely on ground motion prediction equations (GMPEs) derived from inland data. Consequently, seismic hazard in marine environments remains poorly constrained. This study aims to evaluate the feasibility and applicability of both probabilistic (PSHA) and deterministic (DSHA) seismic hazard assessments for submarine areas, using updated seismogenic zonation, a high-resolution bathymetric model, and GMPEs suitable for soft marine soils. The objective is to produce reliable peak ground acceleration (PGA) estimates at the seabed and to examine the convergence of deterministic and probabilistic approaches in a complex tectonic context. The analysis was conducted using the OpenQuake (OQ) engine for PSHA and a custom MATLAB© script for DSHA. The seismogenic sources and parameters were taken from the ESHM20 model, and four GMPEs—IDR91, HZ23, NT24, and DKK24—were selected for their relevance to offshore or soft soil conditions. These GMPEs were validated against six regional offshore earthquakes recorded in the IMR. Hazard estimates were computed over a bathymetrically refined grid and expressed as PGA maps for various return periods (100, 475, 2475, and 5000 years). The results show that significant PGA values occur over key submarine fault systems such as the Gorringe Ridge, Horseshoe Plain, and Arzew Fault. Offshore PGA estimates for return periods of 2475 years using NT24 and DKK24 are close to those derived from DSHA, settling convergence between methodologies. The use of seabed-adapted GMPEs resulted in higher and more realistic PGA values compared to inland models. This study demonstrates the applicability of seismic hazard assessment methods to offshore environments and highlights the importance of incorporating seafloor conditions into hazard modelling. The findings offer a methodological basis for improving the seismic design of subsea infrastructure in tectonically complex marine regions such as the IMR.

Azores Islands are seismically active due to the tectonic structure of the region. Since the 15th century, they have been periodically shaken by approximately 33 moderate to strong earthquakes, with the most recent one in 1998 (M_w = 6.2). Nonetheless, due to insufficient instrumental seismic data, the region lacks a uniform database of past real records. Ground motion simulation techniques provide alternative region‐specific time series of prospective events for locations with limited seismic networks or regions with a seismic gap of catastrophic earthquake events. This research establishes a local ground motion model (GMM) for the Azores plateau (Portugal) by simulating region‐specific records for constructing a homogeneous dataset. Simulations are accomplished in bedrock using the stochastic finite‐fault approach by employing validated input‐model parameters. The simulation results undergo validation against the 1998 Faial event and comparison with empirical models for volcanic and Pan‐European datasets. A probabilistic numerical technique, namely the Monte‐Carlo simulation, is employed to estimate the outcome of uncertainty associated with these parameters. The results of the simulations are post‐processed to predict the peak ground motion parameters in addition to spectral ordinates. This study uses XGBoost to circumvent the difficulties inherent to linear regression‐based models in establishing the form of equations and coefficients. The input parameters for prediction are moment magnitude (M_w), Joyner and Boore distance (R_JB), and focal depth (FD). The quantification of GMM uncertainty is accomplished by analyzing the residuals, providing insight into inter‐ and intra‐event uncertainties. The outcomes demonstrate the effectiveness of the suggested model in simulating physical phenomena.

The safety and resilience of school buildings against natural disasters is of paramount importance since schools represent a reference point for communities. Such significance is not only related to the direct consequences of collapse on a vulnerable part of the population, but also due to the importance of schools in the post-disaster recovery. This work is focused on the risk and resilience assessment of school buildings in Lisbon (Portugal) under seismic events. The results of this study, in which a subset of 32 schools are analyzed, are used to define a prioritization strategy to mitigate the seismic risk of the Lisbon City Council school building portfolio and to assess the overall resilience of the school network. Numerical modeling of the school buildings is performed in order to estimate losses in terms of the built-up area of the schools and recovery times associated with different seismic scenarios, which are probabilistically defined specifically for the sites of the buildings, accounting for the local soil conditions and associated amplification effects. Based on the obtained risk estimates, which are compared to reference values established on international guidelines and specialized literature, the Lisbon City Council and LNEC jointly defined a short- and medium-term risk mitigation plan, starting with a detailed inspection and assessment of the most vulnerable school buildings and continuing to the implementation of retrofitting measures.