05.12.2023 Views

Environmental Internship Program - 2023 Booklet

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

OCEANS AND<br />

ATMOSPHERE<br />

Charlotte Merchant ’24<br />

COMPUTER SCIENCE<br />

Certificates: Applied and Computational<br />

Mathematics, Statistics and Machine Learning<br />

PROJECT TITLE<br />

Sensitivity Analysis of<br />

pCO 2<br />

Estimations and<br />

Code Migration for<br />

Enhanced Climate<br />

Modeling<br />

ORGANIZATION(S)<br />

Max Planck Institute<br />

for Meteorology<br />

LOCATION(S)<br />

Hamburg, Germany;<br />

Ostend, Belgium<br />

MENTOR(S)<br />

Peter Landschützer,<br />

Research Director,<br />

Flanders Marine Institute<br />

(VLIZ); Annika Jersild,<br />

Postdoctoral Researcher,<br />

Max Planck Institute for<br />

Meteorology<br />

I studied the influence of sea surface<br />

temperatures on the estimation of the<br />

partial pressure of carbon dioxide (pCO 2<br />

)<br />

globally. As a fundamental indicator of the<br />

ocean’s thermodynamic interactions, mixing<br />

phenomena, and air-sea interactions, sea<br />

surface temperature remains a key predictor of<br />

pCO 2<br />

in statistical, algorithmic and machine<br />

learning approaches. However, input sea surface<br />

temperature datasets are inconsistent across all<br />

pCO 2<br />

estimation methods due to differences in<br />

spatial and temporal focus, leading to sources<br />

combining different instrumental records<br />

and interpolation techniques. By evaluating<br />

the sensitivity of pCO 2<br />

predictions to different<br />

datasets, I aimed to distill the reliability of these<br />

estimations. In the program MATLAB, I used a<br />

previously described two-step neural network<br />

methodology for global pCO 2<br />

estimation. I also<br />

worked on migrating the MATLAB code into<br />

the program Python to enable execution within<br />

a high-performance computing environment.<br />

Engaging with an early implementation<br />

of machine learning in a climate science<br />

context motivated me to explore how other<br />

computational advancements can amplify the<br />

predictive capabilities of climate models. I also<br />

enjoyed the opportunity to engage in dynamic<br />

discussions with colleagues. The intellectually<br />

stimulating environment of the institute<br />

cemented my desire to pursue further study in<br />

climate computing.<br />

90

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!