Machine Learning Meets Geophysics: Image Segmentation and Inversion Tools with Johnathan Kuttai

How do we map the subsurface without digging? It is finally time we explore geophysical inversions—the math of working backward from surface data to Earth's hidden structures. Jeff Zurek and researcher Jonathan Kataj discuss using Image Segmentation and foundational AI models (like Meta’s "Segment Anything") to resolve "fuzzy" data into precise geological maps.From the Athabasca Basin to remote mineral exploration in China, we break down the ill-posed math and the messy reality of the scientific research process.HighlightsThe Inversion Problem: Solving mathematical equations with infinite solutions.AI & Machine Learning: Repurposing self-driving car tech for geological faults.Tech-Mining: The transition from academic theory to industry application in the Athabasca Basin.The Human Element: PhD career paths, remote logistics, and field stories from northern China.Chapters(00:00) Geophysics Pickup Lines(01:50) Machine Learning in Scientific Applications(03:15) What is a Geophysical Inversion?(05:10) The Logistics of Remote Data Collection(06:50) Introducing Jonathan Kataj(08:15) Image Segmentation Methods in Geophysics(11:00) The Winding Path from Engineering to PhD(13:00) Defining "Ill-Posed" Problems & Null Space(15:20) Building on the Oldenburg & Li Legacy(17:40) Jargon Alert: Gaussian Mixture Markov Random Fields(21:40) Why Standard Inversions Create "Fuzzy" Images(25:45) Foundational Models: Training on the Internet(29:30) Case Study: The Athabasca Basin & Unconformity Deposits(35:00) Magnetotellurics vs. DC Resistivity(38:30) Closing the Gap Between Industry and Academia(41:50) The Future of "Tech-Mining" and Prospectivity Mapping(47:15) Field Story: Party and Hot Pot -China(51:00) The Best Segmentation Science JokeLinksUBC Geophysical Inversion Facility: https://gif.eos.ubc.ca/Support: PateronSocials: Bluesky | Instagram | Facebook | [email protected] Wavelengths: Deep-dive conversations where a working scientist unpacks how we know what we know, one paper, one idea, or whimsical detour at a time. Hosted by Dr. Jeffrey Zurek (P.Geo).