Statistical Learning for Atmospheric and Climate Sciences
- Understanding the theoretical basis of machine learning
- Getting familiar with overarching concepts such as bias-variance trade-off, cost-functions, hyper parameters, cross-validation
- Having good command of the theoretical basis of selected machine learning tools
- Ability to select the appropriate statistical learning tools to tackle atmospheric and climate research problems
- Applying methods of statistical learning in atmospheric and climate research
The course will consist of overview lectures, hands-on practical exercises on (1) the basics of statistical learning and (2) with a focus on applications for atmospheric and climate science. Lectures will cover theoretical basics of statistical learning (advanced regression, nonlinear methods) and an overview of applications of statistical learning in the atmospheric and climate sciences.
Specific topics:
- Data in atmospheric and climate research (data types, observations, models)
- Exploring properties of atmospheric and climate data (data in space and time, multivariate data)
- Concepts of supervised learning (bias variance trade-off, overfitting, cross-validation)
- Advanced linear regression (multiple linear regression, regularization)
- Non-linear regression (local linear regression, regression trees, gradient boosting, random forests, neural networks)
- Bootstrapping
- Keynote speakers showcasing recent topics in statistical learning (Nicolai Meinshausen, ETH) and high-level applications (Mark Liniger, MeteoSwiss) for atmospheric and climate research
- Hastie, T., Tibshirani, R. & Friedman, J. (2009). The elements of statistical learning (Ed. 2). New York: Springer series in statistics.
(external page Link to book, external page book homepage) - James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning. New York: springer. (external page Link to book, external page book homepage)
- Knowledge of introductory statistics
- Basic overview of the climate system
- Basic experience in a programming language
Course limited to 30 participants.
Exercises will be conducted in the R environment (external page https://www.r-project.org/) for most of the sessions and in Python (external page https://www.python.org/) for deep learning.