Rapid generation of reports on post-seismic events with gmProcess: a case study for a dense accelerometric network in Veneto (NE Italy)
Solid Earth Sciences, Seismology
Research area
In 2022, the National Institute of Oceanography and Experimental Geophysics - OGS established a dense accelerometric network in Veneto (Northeastern Italy) to monitor strong ground shaking from earthquakes. Covering over 50% of the region’s municipalities with 318 sites, it enables rapid damage assessment and efficient rescue coordination. The network employs advanced MEMS accelerometers with a high signal-to-noise ratio, capable of recording earthquakes above magnitude 2.0. It utilizes USGS’s open-source “gmProcess” software to analyze ground motion and generate station reports with parameters like PGA. Given the large waveform volume and need for real-time processing, we leverage the TERABIT HPC infrastructure. This case study explores the benefits of integrating HPC technology into seismic monitoring in Northeastern Italy.
Project goals
The project confirms and extends the operational goals of the RAD (Rete Accelerometrica Diffusa del Veneto), a dense accelerometric network of over 350 sensors deployed across 305 municipalities in the Veneto region (NE Italy), funded under POR FESR 2014–2020. The primary goal – to support Civil Protection with near real-time ground motion measurements and rapid post-seismic station reports – has been fully achieved through the development of an automated workflow based on gmProcess (USGS). Key contributions include: (1) redesign and parallelization of core modules, reducing processing time for 504 channels from 30 to 4 minutes; (2) automated generation of station-level PDF reports with waveforms, PSA spectra, PGA tables, and ShakeMap outputs; (3) integration of NTC 2018 for soil-class design spectra; (4) interoperability with ShakeMap via XML export; (5) integration with TeRABIT HPC infrastructure.
Computational approach
The main technological challenges concern the management of large amounts of real-time seismic data and the need to automate the processing workflow with gmProcess. Currently, the OGS servers do not have sufficient computational resources to quickly process data from dense accelerometric networks, so it is necessary to use high-performance computing infrastructures such as Terabit. One of the main limitations is the time required to generate detailed seismological reports (PNG/PDF). This could be reduced by using HPC nodes and optimizing the code for parallel computing. In addition, data transmission could benefit from integration into the high-speed GARR network, which would significantly reduce transmission times. Finally, the workflow is not yet fully automated: Each phase (data acquisition, processing, and report generation) requires manual intervention. The implementation of an automated sequential execution system would improve efficiency, reliability and responsiveness to seismic events.
Image
The diagram shows the seismic data processing workflow using gmProcess. Data is collected and sent to the CRS Venice Physical Server via the OGS Server Cloud, where gmProcess processes it in four steps to generate reports. Data is also transmitted through the GARR network to the RECAS Terabit infrastructure, where another instance of gmProcess runs the same steps. This setup ensures redundancy and higher computational performance for faster and more efficient seismic event analysis.
Key results
The main achievement is a fully operational automated workflow for near real-time post-seismic reporting, developed through extensive adaptation of gmProcess to meet the operational requirements of the RAD network. The summary_plots.py module was completely redesigned to generate comprehensive station-level PDF reports, combining accelerometric waveforms, PSA response spectra, PGA tables, and network maps. Integration of the Italian seismic code NTC 2018 enables automatic computation of design spectra for soil classes A–E based on Vs30 values from Mori et al. (2020), allowing direct comparison between observed and normative spectra for engineering applications. A new export_pga module provides PGA values in XML format compatible with ShakeMap. New subcommands (generate_maps, mapspga) enable automatic generation of station-based and network-level PGA maps consistent with the ShakeMap colour scheme. All modules were rewritten for true multi-core parallel execution, reducing total processing time for 504 channels (168 stations) from approximately 30 to 4 minutes. Integration with TeRABIT HPC infrastructure ensures scalability for thousands of channels. Each station report averages approximately 1 MB; the complete 168-station report is about 25 MB after automatic compression. The workflow was successfully tested on the October 2023 Rovigo earthquake sequence (Ml 4.4) and is fully operational for rapid Civil Protection response.
Resource usage
TeRABIT resources were used via the CINECA Cloud project "OGS23_Tbit05C", which provided a virtual machine and storage volumes dedicated to developing and testing the gmProcess-based automated workflow. The virtual machine hosted the processing environment for benchmarking and optimization of the parallelized modules, enabling systematic performance testing across increasing channel volumes (from 504 up to 4500 channels). The infrastructure was configured to simulate the operational conditions of the RAD network, allowing validation of processing times and workflow scalability under realistic workloads. Storage volumes were used to manage large datasets of seismic waveforms and generated outputs, including station-level PDF reports and PGA maps.
What's next
Following the TeRABIT-supported development phase, the gmProcess-based workflow will be further refined, with several core modules rewritten to achieve greater performance and efficiency. Scalability will be a key priority, with the aim of extending the system to new seismic and accelerometric networks managed by OGS, which will significantly increase the volume of data to be processed in near real-time. Integration with FDSN Web Services will be consolidated, enabling automatic download of waveforms and metadata from any FDSN-compliant provider, thus extending operational capability well beyond the RAD network. A dedicated web platform for interactive visualization of results is planned, allowing rapid consultation of station reports, PGA maps, and ground motion parameters by operators and Civil Protection authorities. Possible integration with UrgentShake, OGS's early warning and rapid impact assessment system, is also foreseen, further strengthening the operational chain for earthquake response. To ensure continuity and scalability, the use of cloud infrastructure will be evaluated alongside OGS internal computational resources. The automatic reporting system is planned for integration into the Civil Protection operational chain as a standard rapid post-earthquake documentation tool.
Image
Top: Architecture of the gmProcess-based automated workflow integrated with the RAD network. Seismic waveforms are acquired via a cloud server and processed on the CRS Virtual Machine, with triggering to the TeRABIT cloud infrastructure for HPC processing. Bottom left: Processing time of the original serial workflow. Bottom right: Execution time scaling of the optimized parallelized workflow across increasing channel volumes (504–4000 channels).
Giorgio Capotosti
Istituto Nazionale di Oceanografia e di Geofisica Sperimentale
Giorgio Capotosti is a technologist at the Seismological Research Center – OGS. He manages accelerometric networks in Veneto, Emilia-Romagna, Friuli Venezia Giulia and Campi Flegrei and uses the USGS software gmProcess software for rapid post-seismic report generation and the calculation of parameters such as PGA and PGV, using the TeRABIT HPC infrastructure. He uses Zabbix and Nagios to monitor the seismic network and writes Python and Bash scripts for data management. He has installed and calibrated AD.EL ASX2000 accelerometers, managed temporary seismic networks, and participated in research projects. He holds a Master’s degree in Geological Sciences and Technologies and has advanced skills in Python, MATLAB, QGIS, Linux servers and OpenVPN. He has presented papers on gmProcess at international conferences and published studies on seismic networks and environmental vibrations.