AI for Climate Science seminar series: AI applications for Climate Science

In this first seminar, Yi-Ling Hwong and Sandeep Chowdhary will present insights into AI applications for Climate Science and share their professional experiences in the field.
Kai Kornhuber will introduce the seminar and moderate the session.
For online participation, a registration is necessary.

First presentation: From Clouds to Clicks: Machine Learning for Convective Parameterisation and Climate Science Communication by Yi-Ling Hwong

This talk will present two applications of ML in distinct but similarly complex domains: cloud modelling and science communication. The first part introduces how ML can be used to represent subgrid-scale convective processes in general circulation models. High resolution cloud-resolving models capture these processes accurately but are computationally expensive. ML methods such as deep neural network and random forest, trained on a multiscale model, can replace traditional parameterisations in a climate model, preserving key atmospheric dynamics at a fraction of the cost.
The second part presents a ML-driven analysis of public engagement and trust in climate science on social media. Using supervised and unsupervised methods, the factors driving audience engagement and trust in science are analyzed. Audience engagement (clicks) is driven by visual elements, while trust in science is shaped by similarity. Authenticity is the only factor that influences both engagement and trust, suggesting that audiences value personal, honest, and genuine messages.

Second presentation: Integrating AI tools Individually and Institutionally by Sandeep Chowdhary

This talk will begin by categorizing AI tools for writing, reading, summarization, podcast-style paper reviews, and more. It will examine how these tools accelerate research by making information retrieval (and writing) more seamless.
Next, the need for high-accuracy, domain-specific models instead of broad generalized Large Language Models (LLMs) will be discussed. Another focus will be how LLMs, as a soft technology, enhance query ability and significantly speed up research by bridging the gap between fuzzy human commands (natural language) and the precise language machines require. As a result, interactions become smoother and faster, increasing the bitrate of human input. Finally, Cursor AI will be showcased, in particular, how it improves efficiency in coding and research.