Hagelslag

Storm tracking, machine learning, and probabilistic evaluation

Hagelslag is an object-based severe storm forecasting system that utilizing image processing and machine learning tools to derive calibrated probabilities of severe hazards from convection-allowing numerical weather prediction model output. The package contains modules for storm identification and tracking, spatio-temporal data extraction, and machine learning model training to predict hazard intensity as well as space and time translations.

Citation

If you employ hagelslag in your research, please acknowledge its use with the following citations:

Gagne, D. J., A. McGovern, S. E. Haupt, R. A. Sobash, J. K. Williams, M. Xue, 2017: Storm-Based Probabilistic Hail
Forecasting with Machine Learning Applied to Convection-Allowing Ensembles, Wea. Forecasting, 32, 1819-1840. 
https://doi.org/10.1175/WAF-D-17-0010.1. 

Gagne II, D. J., A. McGovern, N. Snook, R. Sobash, J. Labriola, J. K. Williams, S. E. Haupt, and M. Xue, 2016: 
Hagelslag: Scalable object-based severe weather analysis and forecasting. Proceedings of the Sixth Symposium on 
Advances in Modeling and Analysis Using Python, New Orleans, LA, Amer. Meteor. Soc., 447.

If you discover any issues, please post them to the Github issue tracker page. Questions and comments should be sent to djgagne at ou dot edu.

Requirements

Hagelslag is compatible with Python 3.6 or newer. Hagelslag is easiest to install with the help of the Miniconda Python Distribution, but it should work with other Python setups as well. Hagelslag requires the following packages and recommends the following versions:

• numpy >= 1.10
• scipy >= 0.15
• matplotlib >= 1.4
• scikit-learn >= 0.16
• pandas >= 0.15
• arrow >= 0.8.0
• pyproj
• netCDF4-python
• xarray
• jupyter
• ncepgrib2
• pygrib
• cython
• pip
• sphinx
• mock

Install dependencies with the following commands:

git clone https://github.com/djgagne/hagelslag.git
cd ~/hagelslag
conda env create -f environment.yml
conda activate hagelslag

Installation

Install the latest version of hagelslag with the following command from the top-level hagelslag directory (where setup.py is): pip install .

Hagelslag will install the libraries in site-packages and will also install 3 applications into the bin directory of your Python installation.

Use

A Jupyter notebook is located in the demos directory that showcases the functionality of the package. For larger scale use, 3 scripts are provided in the bin directory.

• hsdata performs object tracking and matching as well as data processing.
• hsfore trains and applies machine learning models.
• hseval performs forecast verification.

All scripts take input from a config file. The config file should be valid Python code and contain a dictionary called config. Custom machine learning models and parameters should be contained within the config files. Examples of them can be found in the config directory.

Documentation

API Documentation is available here.