11th - 15th July, Reuben College Common Room
This hackathon will focus on the interface of climate science and machine learning and introduce students to the automated analysis of satellite imagery that can be used across disciplines.
Clouds play a key role in the climate system via their role in the hydrological cycle and by modulating Earth’s radiation budget. Clouds scatter radiation back to space, cooling the Earth system, but also interact with outgoing infrared radiation emitted from the surface, warming the Earth system. The net effect of clouds is to cool Earth by about ten times as much as the warming radiative effects of man-made CO2. Even a small response of clouds to a changing climate could have significant effects, termed cloud feedbacks. Unfortunately, clouds are difficult to represent in climate models as many key processes occur on scales that cannot be explicitly resolved. For example, droplets form at about 10-6 meters and small clouds may only be 100s of meters deep and wide – while current climate models have a typical resolution of around 100 kilometres. Hence, clouds remain the key uncertainty in current climate models.
Recent work in Reuben Fellow Philip Stier’s Climate Processes Group in the Department Physics, based on machine learning (gradient boosting decision trees, neural networks & k-means clustering), showed that clouds in current climate models respond very differently to their environmental controls than observations suggest, which could significantly limit climate models’ ability to accurately represent the response of clouds in a changing climate. In this hackathon we will try to identify the type of clouds dominating these climate model uncertainties from satellite observations (as shown above), combining labelling of cloud regimes in satellite datasets with supervised and unsupervised machine-learning based cloud classification techniques.
No prior knowledge required.
||Introduction to clouds and climate – Philip Stier (Physics, Reuben)
||Introduction to the hackathon and the underlying datasets – Alyson Douglas (Physics)
Hands on tutorials on clouds and climate grouped by interests:
- No programming experience or background
- Python & ML beginners
- Python & ML experts
||Independent work as time allows
Daily virtual office hours with tutors
||Presentation of results by each group.
||Group lunch and discussions: what have we learned overall?
Sign up for the hackathon on Inkpath