Research & Projects
Research & Projects
Embedded Files
Transient Electromagnetism (TEM) inversion model collected over the Valles Caldera in New Mexico, where I tested the effects of varying receiver and transmitter heights.
Summer of Applied Geophysics Experience (SAGE)
Summer of Applied Geophysics Experience (SAGE)
As part of the SAGE 2025 cohort, I learned to collect gravity, magnetotelluric (MT), seismic reflection/refraction, and transient electromagnetism (TEM) data. I used Aarhus software to create 2D inversion models of Loupe TEM data to profile the near subsurface of the Valles Caldera and presented my findings at the end of the program. Thank you to everyone who supported me on the way to SAGE, and to all those I met there!
2D Simulation of Tufa Growth
Modeling Tufa Morphogenesis
Modeling Tufa Morphogenesis
Tufas are dendritic carbonate precipitates that form in highly alkaline lakes, such as Mono Lake in California. They are used as paleoclimate archives and evidence of microbial life. One model for their growth is a process known as Diffusion Limited Aggregation (DLA). DLA occurs when there are no advective forces and diffusion is the primary means of particle transport. Branching patterns, such as those you might see in a snowflake, frost on a window, or mineral veins in a rock, are characteristic of DLA. To date, no quantitative comparisons between tufa shape (e.g., branching patterns) and DLA exist. Here, I build a computational model of DLA with the intention of comparing my outputs to real-world three-dimensional (3D) models of tufas. I aim to test whether my models are statistically similar or different to my samples. My initial efforts have successfully recreated branching morphologies with enough detail to enable this comparison. Researchers have also pointed out that fluid flow may modify the shape of tufas. Therefore, as a future step, I intend to modify my models to include an advective component and test the effects of increasing current on tufa shape.
The GitHub repository can be found here.
GRAPES in the Pacific Northwest
Spring 2024GRAPES in the Pacific Northwest
This project focused on learning Julia and evaluating the applicability of a newly developed model, as described in the referenced paper, to Washington state. The model achieved good performance, but it was determined that incorporating additional events and stations would be necessary to improve the results.
This project focused on learning Julia and evaluating the applicability of a newly developed model, as described in the referenced paper, to Washington state. The model achieved good performance, but it was determined that incorporating additional events and stations would be necessary to improve the results.
Objectives
Objectives
This project utilizes the GRAPES graph neural network model, developed by Tim Clements for the U.S. Geological Survey, which can be found here. The model is designed for real-time use with seismic stations to predict peak ground acceleration (PGA) as an earthquake begins to rupture, making it suitable for early warning systems. This repository demonstrates how to retrieve seismic events from online databases in the Pacific Northwest and use GRAPES to predict PGA. My test focused on evaluating how well the model can be applied in areas it was not originally trained on.
Results
Results
I successfully ran the model and demonstrated its application with seismic stations within Washington State. Although the model produced a high mean absolute error, this is likely due to the limited number of events and stations used. Washington experiences many low-magnitude earthquakes, which add additional challenges to accurately predicting ground motion in the area, as the model is primarily trained on larger events.
GitHub repository link is here.
XGBoost Earthquake Magnitude Prediction
Fall 2023XGBoost Earthquake Magnitude Prediction
Inspired by Pablo Lara's model, my project partner Benz and I aimed to create an XGBoost ensemble model to predict the magnitude of earthquakes in the Pacific Northwest.
Inspired by Pablo Lara's model, my project partner Benz and I aimed to create an XGBoost ensemble model to predict the magnitude of earthquakes in the Pacific Northwest.
Objectives
Objectives
This was our first real machine learning project, and we had to learn various key methods, including data cleaning techniques, proper splitting of data into training and testing sets, and how to ensemble the model to run with different K-Folds for each run.
Results
Results
We were able to produce a model with a mean absolute error score of 0.2003. This is very good considering most events in Washington are very low magnitude. The next step would be to see how does the model handle larger magnitude events, or earthquakes outside of Washington.
GitHub repository link is here.
Page updated
Google Sites
Report abuse