I am currently a Research Scientist working in the Data Assimilation group at the Nansen Environmental and Remote Sensing Center in Bergen, Norway. Broadly speaking, I am interested in developing Data Assimilation and Machine Learning techniques that can constrain weather and climate prediction systems toward the most realistic estimate of the coupled Earth system. My background is interdisciplinary, and I enjoy integrating methods from the world of mathematics and computer science to a variety of applications, from medium-range weather forecasting to reanalysis production.
Before joining the Nansen Center, I worked in the Modeling and Data Assimilation Division of NOAA's Physical Sciences Lab in Boulder, Colorado, USA. At PSL I worked on a variety of projects aimed at designing Machine Learning weather emulators for weather forecasting and Data Assimilation applications. Most recently, I led the development of a Machine Learning based global weather emulator which has a high resolution (6km) refinement over the Contiguous United States. The model produces forecasts with lower mean squared error than NOAA's traditional weather prediction systems, including GFS and HRRR , but currently suffers from excessive blurring due to its deterministic training. Work is ongoing to extend the model to produce stochastic, higher temporal resolution predictions. Prior to that, my work was centered on building and using neural network emulators to produce inexpensive ensemble members for Data Assimilation methods. For these projects I worked with idealizations of weather dynamics, ranging from Lorenz-type models to models of Surface Quasi-Geostrophic turbulence.
I obtained my Ph.D. in Computational Science, Engineering, and Mathematics in 2021 from the Oden Institute at UT Austin. My graduate work focused on quantifying uncertainties that are inherent to ocean models, and I implemented a generic, adjoint-based framework to propagate these uncertainties onto predictions from the MIT general circulation model. I used this framework to show the degree to which sparse hydrographic observations reduce uncertainty in simulation-based estimates of ocean driven melting underneath the Pine Island ice shelf, which is fed by one of the fastest flowing glaciers in Antarctica. The results showed how valuable observations in this region are for constraining modeled quantities.