Experimenting With AI in Subglacial Hydrology Research

May 28, 2026

Thupstan Angchuk and Arindan Mandal in the Himalayas, Ladakh, India | photo credit: Karuna Sah

Aleah Sommers' research colleagues, Thupstan Angchuk and Arindan Mandal, in Ladakh, India | photo credit: Karuna Sah

How Machine Learning Models May Help Predict Glacier Melt and Movement

Subglacial hydrology requires a rugged explorer’s fortitude and an experimenter's mindset. Whether atop a glacier or along a glacial lake, in the lab or at a conference, subglacial hydrologists sit with many unknowns, particularly about what’s happening beneath the ice. Research scientist and lecturer Aleah Sommers, Thayer School of Engineering, embodies that fortitude and mindset to advance her research.

From Greenland to India, Sommers traverses high mountain glacier regions to collect data, including how glacier melt and movement impact downstream communities, and “the big implications of their environmental impact.” It's a gigantic undertaking, and what she and fellow researchers are attempting to solve is “tricky,” from navigating tough terrain to difficulties monitoring on the ground. This work is critical because the impacts, as Sommers explains, “can be catastrophic to infrastructure with lives lost downstream.”

Aleah Sommers in Ladakh, India | photo credit: Karuna Sah

A Shared Passion

Last November, Sommers gave a talk about her subglacial hydrology research at the Irving Institute. In it, she highlighted her recent fieldwork in Ladakh, India (where all these remarkable images were captured) with fellow collaborators from the Irving-affiliated Indian Institute of Science. She presented on the physics-based mathematical equations she uses to model how water runoff impacts glacier movements, emphasizing how these models contain many “unknowns because we can’t see below the ice.”

In the audience was Mashrekur Rahman, AI Research Data Science Specialist at the Libraries, who shares a background in hydrology and groundwater research. His interest and curiosity were piqued when listening to Sommers' talk, particularly about models. He grew excited about what could result if Sommers used machine learning to model glacier ice melt, its movements, and its impacts as a complement to her physics-based approach. So he asked her, and a “fun,” collaborative, and exploratory machine-learning AI modeling experiment followed.

These new [AI] tools and techniques open up a whole new avenue for us to understand coupled systems that are difficult to observe directly.

Aleah Sommers, Research Scientist and Lecturer at Thayer School of Engineering

From Aleah Sommers' fieldwork with research colleagues in Ladakh, India | photo credit: Dr Aleah Sommers

Creating Something Novel and Useful

With this research project, Sommers hopes to answer the question, “What can [Mash and I] do together to create something novel and useful?” while acknowledging that “we’re in the exploratory phase without knowing exactly what new insights will come out of this project, which makes it most exciting.”

“At the moment,” says Sommers, “physics-based models approximate interactions between ice and water in the complex natural world using mathematical equations to represent the relevant processes. But these new [AI] tools and techniques open up a whole new avenue for us to understand coupled systems that are difficult to observe directly.”

Rahman offers the additional reflection on the benefits of applying AI: “All the relationships that exist in nature and how humans interact with them are nonlinear; there are degrees of change across time and space. Machine learning can capture all those interdependencies without researchers having to explicitly dictate them, as they would have to in a physics-based model system.”

Sciences are going through a paradigm shift. I’m here to help researchers across Dartmouth explore AI and how to integrate AI into their research and workflows.

Mashrekur Rahman, AI Research Data Science Specialist at the Libraries

field researchers with their guide, far left, in the Hamalayas

From Aleah Sommers' fieldwork with research colleagues in Ladakh, India | photo credit: Karuna Sah

What a Successful Model Looks Like

Rahman is currently experimenting with how to translate Sommers’ “SHAKTI,” a subglacial hydrology model, to a multivariable, high-speed, machine learning-based one. He's using datasets—including observed ice velocities, different climate and weather variables, and how much water reaches below a glacier—from various glacier sites, not just one, to train the model.

They hope to produce a “stand-alone” model with a “very clear, reproducible workflow,” says Sommers. And maybe, offers Rahman, their work also results in “future researchers [referring] to the pipeline we applied and [using] that as a base for other Earth sciences.” Ultimately, a successful model will support “real-world water resource managers in glaciated regions from the Andes to the Arctic,” providing relevant and useful information to decision-makers who are in positions to effect policy and change.

Our work is a great example of what happens when you combine different disciplines to leverage Dartmouth's resources and expertise.

Aleah Sommers, research scientist and lecturer, Thayer School of Engineering

Rahman and Sommers’ generative partnership continues to grow. Their abstract for “Machine Learning-Driven Understanding of Subglacial Hydrology and Glacier Sliding Dynamics Across Temporal Regimes” was accepted to the International Symposium on AI in Glaciology. Organized by Mathieu Morlighem, Evans Family Distinguished Professor of Earth Sciences, Dartmouth will host this year’s symposium from June 23 to 27.

Rahman and Sommers will present how their physics-informed machine learning model is “critical for predicting water resources from High Mountain Asia's glaciers, which supply freshwater to nearly two billion people,” and demonstrate how it can “extract physically meaningful relationships from complex simulations.”

Dr. Aleah Sommers, left, with Dr. Mashrekur Rahman

Aleah Sommers and Mashrekur Rahman

In collaborating with Rahman, Sommers found a “wonderful resource.” She shares that working with library experts like him is “great for anyone who may not know where to begin [with AI], what tools to use, and how to apply it to their disciplines.”

Rahman notes that “the sciences are going through a paradigm shift” between physics-based models and machine learning ones. He hopes that Dartmouth researchers will connect with him to explore AI and machine learning to integrate these tools into their research and workflows. He specifies that general tools like a chatbot are limited in their research process capacity; instead, he sees AI's potential in how we tailor and build “optimized architecture that can solve a specific problem.” And having experts at Dartmouth who are “excited to collaborate on research problems and ideas,” says Sommers, “is a tremendous benefit.”

Explore the Symposium Program