José Augusto Proença Maia Devienne

José Augusto Proença Maia Devienne

Twitter is a microblogging service for sending short, public text messages (tweets) that has recently received more attention in scientific comunity. In the works of Sasaki et al. (2010) and Earle et al., (2011) the authors explored the real-time interaction on Twitter for detecting natural hazards (e.g., earthquakes, typhoons) baed on users' tweets. An inherent challenge for such an application is the natural language processing (NLP), which basically consists in converting the words in number (vectors and tensors) in order to (mathematically/ computationally) make predictions and classifications. Recently advanced computational tools have been made available for dealing with text computationally. In this report we implement a NLP machine learning with TensorFlow, an end-to-end open source plataform for machine learning applications, to process and classify evenct based on files containing only text.

José Augusto Proença Maia Devienne

The recent exploitation of natural resources and associated waste water injection in the subsurface have induced many small and moderate earthquakes in the tectonically quiet Central United States. This increase in seismic activity has produced an exponential growth of seismic data recording, which brings the necessity for efficient algorithms to reliably detect earthquakes among this large amount of noisy data. Most current earthquake detection methods are designed for moderate and large events and, consequently, they tend to miss many of the low-magnitude earthquake that are masked by the seismic noise. Perol et. al (2018) has focused on the problem of earthquake detection by using a deep-learning approach: the authors proposed a convolutional neural network (ConvNetQuake) to detect and locate earthquake events from seismic records. This reports aims at reproducing part of the methodology proposed by the author, which is the implementation of a convolutional neural network for classification of events (i.e., earthquake vs. noise) from seismic records.

José Augusto Proença Maia Devienne

In Geosciences a class of phenomena that is widely studied given its real impact on human life are the tectonic faults slip. These landslides have different ways to manifest, ranging from aseismic events of slow displacement (slow slips) to ordinary earthquakes. An example of continuous slow slip event was identified in Cascadia, near the island of Vancouver (CA). This slow slip event is associated with a tectonic movements, when the overriding North America plate lurches southwesterly over the subducting Juan de Fuca plate. This region is located down-dip the seismogenic rupture zone, which has not been activated since 1700s but has been cyclically loaded by the slow slip movement. This fact requires some attention, since slow slip events have already been reported in literature as possible triggering factors for earthquakes. Nonetheless, the physical models to describe the slow slip events are still incomplete, which restricts the detailed knowledge of the movements and the associated tremor. In the original paper, the strategy adopted by the authors to address the limitation of the current models for the slow slip events was to use Random Forest machine learning algorithm to construct a model capable to predict GPS displacement measurement from the continuous seismic data. This investigation is sustained in the fact that the statistical features of the seismic data are a fingerprint of the fault displacement rate. Therefore, predicting GPS data from seismic data can make GPS measurements a proxy for investigating the fault slip physics and, additionally, correlate this slow slip events with associated tremors that can be studied in laboratory. The purpose of this report is to expose the methodology adopted by the authors and try to reproduce their results as coherent as possible with the original work.