Peter Van Katwyk

The year is 2100. The ocean has risen by over a meter. In the US, this means dedicating a massive amount of resources to building flood walls. The cost is astonishing, but few are displaced. These are the fortunate. Hundreds of thousands lose their homes in third-world countries. Entire islands are swallowed up. If we want to help future generations by preventing these catastrophes, we need to act now by tracking and predicting the ongoing sea-level rise.

Peter Van Katwyk is a BYU alumnus who graduated in 2017 with his undergraduate degree in geology, which he explained was a fantastic experience. His grandpa was a geology professor here and inspired Peter to become a geologist himself. He had zero skills when he began but became involved in undergraduate research almost immediately, setting the course for the rest of his career. After BYU, he attended Brown University, where he is currently working on a PhD. His thesis centers around using artificial intelligence to create projection models of ice melt and sea-level rise. He came to BYU on January 22 to speak about the intersection between geology and technology.

Monitoring sea-level rise and predicting its progression is an extremely tedious process. To create just a single prediction, estimates need to be taken from large groups of climate models and compiled into what are classified as "MIPs" (Model Intercomparison Project). These MIP's use physics-based computational models to simulate ice sheet flow, melt, and retreat. Though incredibly useful, these models can take over 1000 CPU hours to make.

To create a greater quantity of models, Peter creates emulators—small-scale models to approximate the larger supercomputer models. He explained different methods of developing these emulators using reduced complexity, statistics, and the Gaussian function. His main project is called ISEFlow: Flow-based Ice Sheet Emulator. This emulator looks at aggregate sectors in Antarctica and simulates where water is flowing, resulting in the quicker production of projections while also preserving some of the quality of more complex models.

After talking about his specific project, Peter moved on to the intersection of science and AI in general. Using AI to compute data helps shift the bottleneck from creating models to actually studying them. It is transforming scientists' relationship with climate models as they are more accessible and more queries are possible. In this sense, AI is a powerful tool. That being said, Peter emphasized that AI does not replace the scientific process or domain expertise. Geology experts are still very much needed to help interpret and contextualize the data that AI compiles.

To enter this space, you will need skills in three areas: the domain, machine learning, and computer science. Peter emphasized that students should stick with their domain. Most problems don't require AI, but having a solid scientific foundation is always helpful. When he was at BYU, and to expand his opportunities, he took a minor in applied stats and learned machine learning and programming. Those skills have been incredibly handy in his work.

Thanks to the work of Peter Van Katwyk and others like him, we can rest assured that our future is in good hands.