Description
In this second tier of our MOOC on Machine Learning (ML) in Weather and Climate, we go deeper into the theory underpinning ML, and the tools that are used to implement it, using Python case studies in weather and climate.
This five-module course covers regression trees and random forests, explores workflows in ML, and dives deeper into the ingredients of deep neural networks, including architectures, training, tuning and evaluating models. We will also explore approaches to quantify uncertainty, including generative ML models and Bayesian neural networks. The last module covers physics-guided ML approaches with applications to post-processing.
This course is aimed at those with a technical/weather background, some experience in Python, and a solid understanding of statistics. Ideally you will have already taken Tier 1 of this MOOC or already be familiar with the concepts there.
This course features a mixture of interactive modules, webinar recordings, quizzes, podcasts, and contains a number of detailed code examples.
Please note that this course ran live in early 2023 and reflected the state of the art at that point in time.
Course Content
- Introduction to data handling
- Quiz 1: test your understanding of exercise 1
- Quiz 2: test your understanding of exercise 2
- Quiz 3: test your understanding of exercise 3
- Quiz 4: test your understanding of exercise 4
- Notebooks on linear regression and random forests
- Introduction to deep learning 🕓10 min
- Overview of ML architectures in weather and climate 🕓10 min
- Deep dive into the building blocks of modern Deep Learning 🕓15 min
- Understanding of Neural network training and hyperparameter tuning 🕓10 min
- Quiz
- Uncertainty & generative modelling 🕓25 min
- Quiz 🕓5 min
- Physics-Guided Machine Learning 🕓30 min
- Physically-constrained postprocessing of surface weather 🕓25 min
- Quiz 🕓5 min
- Physically-informed subgrid-scale parameterization for climate modeling 🕓25 min
- Quiz 🕓5 min