Charlotte Merchant ’24
Sensitivity Analysis of pCO2 Estimations and Code Migration for Enhanced Climate Modeling
Certificate(s): Applied and Computational Mathematics, Statistics and Machine Learning
I studied the influence of sea surface temperatures on the estimation of the partial pressure of carbon dioxide (pCO2) globally. As a fundamental indicator of the ocean’s thermodynamic interactions, mixing phenomena, and air-sea interactions, sea surface temperature remains a key predictor of pCO2 in statistical, algorithmic and machine learning approaches. However, input sea surface temperature datasets are inconsistent across all pCO2 estimation methods due to differences in spatial and temporal focus, leading to sources combining different instrumental records and interpolation techniques. By evaluating the sensitivity of pCO2 predictions to different datasets, I aimed to distill the reliability of these estimations. In the program MATLAB, I used a previously described two-step neural network methodology for global pCO2 estimation. I also worked on migrating the MATLAB code into the program Python to enable execution within a high-performance computing environment. Engaging with an early implementation of machine learning in a climate science context motivated me to explore how other computational advancements can amplify the predictive capabilities of climate models. I also enjoyed the opportunity to engage in dynamic discussions with colleagues. The intellectually stimulating environment of the institute cemented my desire to pursue further study in climate computing.
Oceans and Atmosphere
Max Planck Institute for Meteorology - Hamburg, Germany; Ostend, Belgium
Peter Landschützer, Research Director, Flanders Marine Institute (VLIZ); Annika Jersild, Postdoctoral Researcher, Max Planck Institute for Meteorology