Desarrollo y aplicación de una herramienta computacional para la detección y localización automática de terremotos en Costa Rica a través de técnicas de aprendizaje profundo y Big Data

The systematic characterization of the seismic potential, the geodynamics of the fault zones and
their possible effects on the most vulnerable populations as well as on developing economies,
strongly depends on permanent seismological monitoring and the development of modern
methods of computation and management of massive data (Big Data). The exponential growth
(in the number of stations and their azimuthal distribution) of seismographic networks at an
international level is allowing the observation of fault physics processes never seen before in
nature, accelerating the comprehensive understanding of the nucleation and propagation of large
earthquakes. In Costa Rica, the OVSICORI-UNA seismographic network, made up of more than
100 seismological observatories distributed throughout the country, has allowed seismologists to
understand in a broad sense, the detail of the seismotectonic processes that are generated
during the seismic cycle of hundreds of faults in the country. The increasing instrumental
densification generates an increase in the amount of time series that must be analyzed, so their
acquisition and processing are of high computational and infrastructure cost. The associated
methodological and scientific cost lies in 1) the loss of low amplitude tremors that would
otherwise be part of the seismic catalog, if not masked by natural or anthropogenic seismic
noise, 2) the loss of systematization in errors measurement and 3) the limited number of
processes that can be observed in real time through conventional data processing methods.
Through the use of the high-performance computing resource at Centro Nacional de Alta
Tecnología (CeNAT) through deep learning paradigms and massive data management (Big
Data), this project aims at providing a computational tool for detecting earthquakes in Costa Rica.
In addition, using the recording capacity of the OVSICORI-UNA (more than 2 Terabytes per year
of seismological data), it is expected to use modern techniques for the training and massive
processing of data and its adaptation to hierarchical models of layers with non-linear processing
neurons to the extraction and implementation of a deep learning model that has the capacity to
recognize the different seismogenic sources in Costa Rica. Once a conditioned and trained
model is available for signal processing, it will be integrated into a workflow with other relevant
subsystems and its performance will be evaluated. This tool has the potential to facilitate the
earthquake localization process and its tectonic interpretation for scientists, systematically
improving the understanding of the behavior and spatio-temporal evolution of fault zones in
Costa Rica

Desarrollar una herramienta computacional para la detección automática de eventos
sísmicos en Costa Rica mediante el empleo de técnicas de aprendizaje profundo y de manejo masivo de
datos

1) Teoría y Metodologías en Computación
2) Aplicación de la computación en distintos dominios científicos, tecnológicos, organizacionales y sociales