Climate Pattern Recognition

An ML system for identifying atmospheric teleconnection patterns and extreme weather precursors from reanalysis data.

Overview

This project applies machine learning techniques to recognize and classify large-scale atmospheric patterns from ERA5 and NCEP/NCAR reanalysis datasets. The system identifies teleconnections (El Nino-Southern Oscillation, North Atlantic Oscillation, Arctic Oscillation, Pacific Decadal Oscillation) and their relationships to extreme weather events.

Features

• Automated download and preprocessing of ERA5 reanalysis data via CDS API
• Sea level pressure and geopotential height pattern extraction
• Unsupervised pattern discovery using clustering algorithms
• Supervised classification of known teleconnection phases
• Extreme event correlation analysis using EM-DAT disaster records
• Interactive visualization of atmospheric patterns

Project Structure

climate-pattern-recognition/
├── src/
│   ├── data/           # Data download and preprocessing
│   ├── features/       # Feature extraction and engineering
│   ├── models/         # ML model implementations
│   └── visualization/  # Plotting and interactive displays
├── tests/              # Unit and integration tests
├── notebooks/          # Exploratory analysis notebooks
├── docs/               # Documentation and implementation plans
└── data/
    ├── raw/            # Original downloaded data
    ├── processed/      # Cleaned and transformed data
    └── external/       # Climate indices and validation data

Data Sources

Dataset	Description	Access
ERA5	Gridded atmospheric variables (0.25 deg, 1950-present)	Copernicus CDS
NCEP/NCAR	Coarser reanalysis (1948-present)	NOAA PSL
NOAA Climate Indices	Pre-computed ONI, NAO, AO, PDO	NOAA PSL
EM-DAT	International disaster database	EM-DAT

Requirements

• Python 3.9+
• xarray, dask (NetCDF handling)
• cdsapi (ERA5 access)
• scikit-learn (ML algorithms)
• matplotlib, cartopy (visualization)

Installation

git clone https://github.com/Sakeeb91/climate-pattern-recognition.git
cd climate-pattern-recognition
pip install -r requirements.txt

Quick Start

from src.data.download import ERA5Downloader
from src.features.patterns import PatternExtractor

# Download monthly mean sea level pressure
downloader = ERA5Downloader()
data = downloader.get_monthly_slp(years=[2020, 2021, 2022])

# Extract dominant patterns
extractor = PatternExtractor(n_components=10)
patterns = extractor.fit_transform(data)

License

MIT License

Contributing

Contributions welcome. Please read the implementation plan in docs/IMPLEMENTATION_PLAN.md before starting.