📖 Overview
This repository contains an enhanced implementation of PyForTraCC with:
✅ Support for NetCDF/HDF5 input files (not available in the original PyForTraCC).
✅ A physics-aware machine learning forecasting module (Wide & Deep hybrid model).
✅ End-to-end pipeline for tracking, analyzing, and forecasting convective cloud clusters.
It is designed for meteorologists, researchers, and data scientists working with satellite data who want both tracking and short-term forecasting capabilities.
✨ Key Features
Extended Data Input
Direct support for NetCDF & HDF5 datasets.
Automated variable detection (temperature, latitude, longitude).
Tracking
Improved PyForTraCC-based detection of convective systems.
Handles splits, merges, growth, decay, and dissipation.
Outputs system lifecycles, trajectories, and movement vectors.
Forecasting (Newly Added)
Physics-aware Wide & Deep ML architecture.
Forecasts brightness temperature evolution.
Constrained by physical laws → realistic predictions.
Visualization
System trajectory animations.
Diagnostic plots: size evolution, intensity changes, movement dynamics.
Statistical summaries (DuckDB-based).
📂 Repository Structure ├── Pyfortracc_Tracking_&_Enhanced_Forecasting.ipynb # Main notebook ├── data/ # Place NetCDF/HDF5 input files here ├── output/ # Tracking and forecast outputs ├── requirements.txt # Dependencies └── README.md # Project documentation
Run Jupyter Notebook / Colab Open the main notebook:
jupyter notebook Pyfortracc_Tracking_&_Enhanced_Forecasting.ipynb
- Upload NetCDF/HDF5 Files
Place your .nc or .hdf5 files in the data/ folder.
The notebook automatically detects variables (temperature, lat, lon).
- Configure Parameters
Set thresholds:
Warm cloud = 235K
Cold cloud = 200K
Adjust cluster size filtering if needed.
- Run Tracking
Detects cloud systems across time steps.
Outputs lifecycle statistics, trajectories, and movement vectors.
- Forecasting (New)
Uses physics-aware Wide & Deep ML model.
Produces short-term temperature forecasts.
Ensures outputs remain meteorologically realistic.
📊 Results (Example)
Tracking: Successfully identified both short-lived convective events and long-lived mesoscale systems.
Forecasting: Achieved 2–4K RMSE with <5% physical violations (temperature outside 150–350K).
Visualization: Generated animations and diagnostic plots for lifecycle analysis.
📦 Requirements
Main dependencies:
pyfortracc (core algorithm)
xarray, numpy, pandas (data handling)
matplotlib, ipywidgets (visualization/UI)
duckdb (tracking result queries)
torch, lightgbm (forecasting models)
Install all with:
pip install -r requirements.txt
🚀 Your Contributions vs Original PyForTraCC
🔹 Original PyForTraCC: Limited to specific formats, tracking-only.
🔹 This Repo:
Added NetCDF/HDF5 input compatibility.
Added physics-aware ML forecasting.
Built an end-to-end workflow (tracking → analysis → forecasting → visualization).
📌 Future Work
Extend forecasts to spatial fields (not just point-based).
Add uncertainty quantification in forecasts.
Optimize for real-time streaming satellite data.
🤝 Acknowledgements
Original FORTRACC methodology.
PyForTraCC library.
Open-source ML & meteorological data analysis community.
📜 License
MIT License.