Extreme Events Engineering Research

I completed my three semesters of work on the Extreme Events Engineering Vertically Integrated Project (VIP), a civil engineering research project under Dr. Jorge Macedo.

During my first semester, I was tasked with updating/fixing a legacy Python codebase that had gone untouched for 5 years that performed seismic catalog declustering calculations on a dataset of earthquakes using the Zaliapin and Ben-Zion method. With help from teammates (who were working on implementations of other declustering algorithms), I ended up adding some data visualizations shown below.

Declustering, or classifying independent and dependent events and removing the dependent events from the catalog, may be useful for isolating independent events during seismic hazard assessment, development of clustered seismicity models, earthquake prediction research, or seismicity rate change estimation.

A map of earthquakes in South America before seismic catalog declustering was performed
A visualization from my Python implementation of all recorded earthquakes in the catalog we used.
A map of earthquakes in South America after seismic catalog declustering was performed (now with less earthquakes)
A visualization from my Python implementation of earthquakes after the dependent events were removed.

During my second semester, I created a MATLAB implementation of the Zaliapin and Ben-Zion method with more complex visualizations and also started creating a MATLAB k-means declustering implementation. The clusters outputted by the k-means implementation did not correspond to those that should have been detected, implying that the k-means method cannot detect the correct clusters, and that we should evaluate other classification methods (i.e., Gaussian mixture models).

Data visualizations from the MATLAB implementation of the Zaliapin and Ben-Zion method.
Data visualizations from the MATLAB implementation of the Zaliapin and Ben-Zion method. See only the independent events in the 'after' visualizations. The lefthand column presents the events in time and space (measured with latitude), while the middle column presents the events' nearest-neighbor distances as histograms and the righthand column presents the events on a heatmap in the Log(T) vs. Log(R) space. Note that between the before and after visualizations, in the lefthand column, the density of events generally decreases, in the middle column, the density of events that have a low nearest-neighbor distance to each other decreases, and in the righthand column of visualizations, the 2 clusters representing dependent events have been removed.