🌦️ hybmkg-pycast

hybmkg-pycast is a hybrid machine learning and deep learning framework for climate and weather forecasting. It is designed to integrate multiple forecasting approaches — from classical statistical models to modern deep learning architectures — and support research and operational applications at BMKG (Meteorological, Climatological, and Geophysical Agency of Indonesia).

🚀 Key Features

Comprehensive Forecasting Models

Statistical: ARIMA, SARIMA, ANFIS, Wavelet-ARIMA, Wavelet-ANFIS, etc.

Machine Learning: Random Forest, XGBoost, LightGBM, SVR, KNN, MLP using multi-stacked approach.

Deep Learning: RNN, LSTM, GRU, CNN, Transformer, and hybrid approaches using multi-stacked approach.

Flexible Data Input

Supports both NetCDF and CSV formats for climate and environmental datasets.

Hybrid Framework

Combine traditional time series models with machine learning and deep learning methods for improved forecast accuracy.

Visualization and Evaluation

Built-in utilities for plotting time series, model diagnostics, and forecast verification (correlation, RMSE, R²).

Project-Oriented Directory Structure

hybmkg_pycast/

├── config/ # JSON configuration files

├── data/ # Input data (CSV, NetCDF)

├── model/ # Model scripts (statistical, ML, DL)

├── plots/ # Generated plots (PNG)

├── results/ # Output results (CSV, trained models)

├── run_all.ipynb

├── hybmkg_pycast.yaml # Conda environment specification

🧩 Installation

You can install the package using pip:

pip install hybmkg-pycast

Or from source:

git clone https://github.com/yourusername/hybmkg_pycast.git

cd hybmkg_pycast

pip install .

If you prefer Conda, use the provided environment file:

conda env create -f hybmkg_pycast.yaml

conda activate hybmkg_pycast

📊 Applications

Seasonal and sub-seasonal climate forecasting

ENSO (El Niño–Southern Oscillation) prediction

Rainfall variability and extreme events analysis

Climate change impact studies

👥 Authors and Acknowledgment

Developed by researchers at BMKG to support data-driven climate prediction and research collaboration.

📄 License

This project is licensed under the MIT License.