Utilizing Scientific Python Tools for the Application of Data Science Techniques to High Impact Weather Prediction

Date and Time: 
2015 April 13 @ 3:00pm
Location: 
FL2-1022 Large Auditorium
Speaker: 
David Gagne

The developments and optimizations provided by Python’s scientific libraries have enabled the development of real-time high-resolution forecast post-processing systems primarily in Python. Numpy, Scipy, Matplotlib, and a set of newer scientific libraries have made this development possible. The Pandas library introduced efficient ways to load, analyze, manipulate, and merge large datasets. Scikit-Image provides a diverse array of image processing tools, which are useful for filtering and extracting information from gridded data. Scikit-Learn makes training and running machine learning models both easy and fast. The IPython Notebook has made it possible to perform interactive data analysis on both local and remote machines through a browser-based code interface. These tools have been incorporated into recent projects on severe hail prediction and the development of a gridded forecasting system for solar irradiance. The severe hail prediction system utilizes an object-finding method to identify potential hailstorms from storm-scale numerical weather prediction model output and then applies machine learning regression techniques to those storms to predict the distribution of possible hail sizes. The solar forecasting system ingests high-resolution numerical model output and point observations of solar irradiance then produces a calibrated gridded irradiance forecast. This talk will showcase examples of how these tools have been applied to optimize computationally intensive tasks, simplify complex processes, and unify analysis tasks once done in multiple languages under a common Python framework.

SLIDES: http://nbviewer.ipython.org/github/djgagne/ucar_sea_2015/blob/master/Gag...

CODE: https://github.com/djgagne/ucar_sea_2015

Speaker Description: 

David John Gagne is a doctoral candidate in meteorology at the University of Oklahoma and a visiting graduate research assistant with the NCAR Research Applications Lab. His main research interests involve the application of machine learning techniques to numerical weather models and observations in order to improve the prediction of high impact weather. He has developed frameworks for improving the prediction of hail, solar energy, wind energy, heavy rain, aircraft turbulence, and tornadoes. He is an active Python developer and has contributed to packages for weather data visualization, forecast verification, and gridded forecast correction.

Video recorded: 

If you use a non-flash enabled device, you may download the video here

Event Category: