SURF 2026

I have two possible projects for a SURF student in the area of data-driven methods for application in the earth system. If interested, please contact me indicating which project(s) you would want to work on. See here for more information about SURF.

Projects

• Analog forecasting methods for use in attribution: Analogue forecasting tools have been used in data-driven climate attribution tasks (see 10.1073/pnas.2111875118) as well as for the forecasting for which they are named. This project will analyze and compare various forms (linear, kernel) of analogue forecasting methods for use in climate attribution problems, asking when these methods can be confidently applied and where they differ. The student working on this problem will compare results from these methods in both simple systems (where the answer is known) and in climate data (where the true answer is unknown).
• Diagnosing changing oscillations from data: Koopman operator-theoretic methods translate problems of nonlinear dynamics to equivalent linear (but infinite-dimensional) problems. Data-driven approximations of these operators are well suited to the analysis of nearly periodic phenomena in climate such as the Quasi-Biennial Oscillation, and an eigendecomposition gives modes that are quasi-oscillatory with an associated frequency. Many of these methods are applied under the assumption that the dynamics are stationary, which is unrealistic under climate change or other outside forcings. The goal of this project is to ask how, and under what limitations, we can diagnose changes in oscillatory phenomena from data.

Student Qualifications

Students will need strong background in linear algebra, (ordinary) differential equations, and coding proficiency — either in Python or Julia. It will be helpful (but not necessary) to have coursework in dynamical systems, optimization, climate dynamics, and data-driven methods.