Growing up in Chengdu, China, Chaopeng Shen remembers the rivers of his hometown transforming during his childhood.
"They became very unappealing," Shen said. "I remember thinking, when I grow up, I want to fix these problems for the betterment of both the environment and human well-being."
That childhood realization planted a seed that would grow into a career devoted to safeguarding water. Today, Shen, a professor of civil and environmental engineering at Penn State, is recognized as a leading innovator in blending artificial intelligence (AI) with water and Earth systems research. For these transformative contributions, the Institute of Energy and the Environment (IEE) has named him Person of the Year for 2025.
Finding connections in Earth systems
As a postdoctoral researcher at Lawrence Berkeley National Laboratory, Shen discovered Earth system models that wove together carbon, water and energy cycles.
"I was amazed that we could put equations - the ones you see from textbooks about the energy cycle, vegetation cycle and water cycle together - and they worked naturally," he said. "You could see how one system influences another, and it was fascinating."
That appeal shaped his approach to Earth science. Instead of isolating processes, he sought their connections and designed tools that could capitalize on those connections.
Turning to AI out of necessity
When Shen arrived at Penn State in 2015, hydrological and Earth system models were becoming unwieldy. They grew more complex with every attempt to capture new processes. However, they often failed to match observations and had no easy way to absorb information from data.
"Our models were becoming too complex, too hard to diagnose," Shen said. "We needed to start from scratch and relearn from data. When I saw how deep learning could discover principles on its own, I realized that this is exactly what we need in water science."
Shen began experimenting with long short-term memory (LSTM) neural networks, a type of deep learning architecture. The equations reminded him of how hydrologists step through processes and solve differential equations. But instead of being programmed, the AI learned the dynamics directly from data.
It wasn't easy, Shen said.
"We spent eight months in the dark," he explained. "Things didn't make sense, and progress was uncertain. It was unclear what we could get out of deep neural networks, and it was even unclear which were the right targets to pursue. Then one day we fixed some bugs, and everything started to work."
Suddenly it was clear: AI could capture natural processes and mimic how the world behaves, he said.
Neural networks as "question marks"
Shen describes his method as using AI not to replace physics but to extend it.
"I think of neural networks as question marks," he said. "When you're not sure about a process but have data, you put a neural network there with other parts of the model that can be trusted and predefined. It's like a flexible glue that learns the useful yet missing relationships from big data."
This approach allows his models to integrate massive amounts of observational data, from river gauges to satellites, absorbing information that was previously impossible to handle. The result is higher-resolution predictions that are locally relevant, whether for a farmer deciding what crops to plant or an emergency manager preparing for floods.
From research to real-world impact
Shen's models now support the scientific foundation of the next-generation National Water Model, which provides real-time flood forecasts across the United States, as well as the next-generation global water model. What once required years of computational time can now be done in minutes, an advance with life-saving implications.
"We are able to do things that were unfathomable before, including building a national hydrologic model that is far more accurate and reliable than past generations," Shen said.
Globally, his work helps anticipate water crises and avoid the mistakes of the past. He points to examples of such past mistakes as Saudi Arabia's unsustainable groundwater use in the 1970s and the collapse of the Aral Sea after water diversions.
"Those were mismanagements of water that I hope our tools can help prevent from happening again," he said.
Learning from criticism
While Shen has been at the forefront of bringing AI into hydrology, he said, he has also taken critiques seriously. Concerns about AI's inability to handle unseen extreme events or its "black box" nature pushed him to develop new hybrid methods.
"Some of the criticism put serious dampers on our efforts to make progress," Shen said. "But it also pushed me forward and forced me to think about harder problems."
His solution was a physics-embedded model, also known as differentiable modeling. This is a framework where physical equations and neural networks are trained side by side and work together.
"Now the neural network becomes your friendly neighborhood helper," he said. "It doesn't do everything. It just brings you the right tool when you need it. There are predefined physical equations that guarantee some base behaviors. That makes the models both reliable and interpretable."
Shaping the field
Shen's influential 2018 review paper on deep learning in water research has been cited more than a thousand times and introduced a new generation of scientists to the field. At first, some reviewers questioned his insistence on including examples from physics, chemistry and biomedicine. But Shen held firm.
"I insisted because I get a lot of inspiration from other domains," he said. "At their foundations, many engineering sciences are very similar. We all have theories, parameters and observations that we want to reconcile. We can learn from each other."
Shen's lab has published articles in the proceedings of top-tier AI conferences, bringing unique insights from his domain to all others. That cross-disciplinary mindset helped shape the narrative around AI in environmental science and positioned Shen as a leader in the space.
Looking to the future
Shen said he envisions an AI "foundation model" for the environment, a comprehensive knowledge repository that could integrate global data, understand complex relationships and forecast changes across ecosystems.
"I hope we can build a tool that grasps all of the knowledge about Earth systems, including how they evolve, how they respond and how they connect," he said. "It would be like a knowledge repository for the planet, guiding both local and global decisions."
He also envisions that the tools and ideas he has developed will be useful for the studies of water quality, ecosystems, agriculture, geohazards, fluid mechanics, landscape evolution and many other science and engineering domains.
Bruce Logan, director of the Institute of Energy and the Environment, said Shen exemplifies the spirit of the Person of the Year Award.
"Chaopeng is pushing the boundaries of what is possible in environmental science," Logan said. "His pioneering use of AI in hydrology is not just advancing research, it's turning knowledge into action that benefits society, ecosystems and future generations."
For Shen, the recognition is both humbling and motivating.
"Water touches everything we care about," he said. "If my work can help protect it and help people make better decisions, then that's what matters most."
