Overview
This project tackles the challenge of understanding and predicting Arctic Ocean dynamics in the face of rapid climate change. Using over 70 months of ICESat-2 satellite altimetry data, I developed a comprehensive machine learning framework to analyze Sea Surface Height Anomalies (SSHA) across the Arctic region.
The Challenge
The project addresses a critical gap in Arctic oceanography: traditional physical models struggle to capture the complexity of polar ocean dynamics, particularly in regions with seasonal ice cover and rapidly changing environmental conditions. Understanding these dynamics is essential for climate change prediction and Arctic navigation.
Technical Approach
I implemented multiple complementary machine learning approaches:
- K-means Clustering: Pattern identification across spatial and temporal dimensions
- Random Forest: Feature importance analysis to identify key oceanographic drivers
- U-Net CNN Architecture: Spatial prediction achieving 16cm accuracy across 448×304 grid
- LSTM Autoencoder: Temporal pattern analysis identifying distinct modes of variability
Key Findings
I was responsible for the entire pipeline, from data acquisition and preprocessing through model development, training, and evaluation. This work ultimately uncovered counterintuitive relationships between sea ice extent and surface height that challenge conventional freeze-melt cycle explanations.
The findings reveal five distinct temporal patterns in Arctic SSHA dynamics and demonstrate how machine learning can provide new insights into climate change impacts in polar regions.
Technical Stack
Research Poster
Click the poster to view it fullscreen.
