How Python Can Help Us to Create the Physical Data Scientists of the Future (Core Science Keynote)

Atmospheric scientists are increasingly faced with an overwhelming deluge of data. As computational power increases, more frequent and higher resolution model runs are available. As new sensing systems come online, high resolution observations are available for forecasters and researchers alike. However, humans can only process so much data in a limited amount of time. Machine learning and data science can provide tools to help atmospheric scientists improve predictions, models, data processing, and understanding. However, atmospheric science students are not typically trained with multiple years of programming (although that is changing).

Python provides a feasible entry point that enables atmospheric scientists to combine their physical understanding of the atmosphere with the mathematical sophistication needed to successfully use machine learning. Not only is python a relatively easy language to learn, there are a plethora of machine learning toolkits available that enable atmospheric scientists to get started quickly on machine learning. These kits include scikit-learn, which provides many of the traditional machine learning methods, and keras and pytorch which provide deep learning methods. In this talk, I will discuss how I have trained atmospheric scientists to become physical data scientists. This includes a discussion of the classroom training in machine learning which is specifically aimed at an interdisciplinary audience as well as research training.

Finally, python also enables us to train the next generation of scientists by exciting K-12 students about STEM majors. I will also talk about how I combine my love of programming with my love of the atmosphere to excite K-12 students in workshops and camps about becoming STEM majors. I will share results of using python to program autonomous drones in a variety of K-20 settings.