Gaussian Process Regression for Ocean Temperature Anomaly Detection

This project explores the detection of climate anomalies using Gaussian Process Regression (GPR) on real satellite-derived oceanographic data. The goal is to model the expected behavior of a key variable (tdrop, representing the temperature drop across the ocean's cool skin layer) and identify regions or times where observations deviate significantly from this model.

---

Project Overview

• Data: NASA satellite reanalysis data from the MERRA-2 dataset, available at Source: Kaggle – OCEAN DATA / CLIMATE CHANGE / NASA
• Model: Gaussian Process Regression (scikit-learn)
• Target: tdrop (temperature drop across cool layer)
• Features:
  • tbar – average temperature of the interface layer
  • tskinice – skin temperature over sea ice
  • rainocn – ocean rainfall
  • delts – surface skin temperature change
• Training: August 1st, 2018
• Testing & Anomaly Detection: August 1st, 2021

---

Requirements:

Python 3.8+ NumPy, pandas, matplotlib, xarray scikit-learn joblib

---

How to run it:

Train the model (optional, will be auto-triggered if missing):

• Run train_model.ipynb or
• Call train_and_save_gpr_model() from train_model.py Run the analysis:
• Open analyze_results.ipynb
• or run the script python analyze_results.py

---

Repository Structure

├── train_model.ipynb           # Notebook to train GPR model
├── analyze_results.ipynb       # Notebook to evaluate predictions and detect anomalies
└── README.md                   # You are here!