You are here
Marta Zaniolo: Applying Novel Models and Data Mining to Water Resource Management
New faculty member Marta Zaniolo pulls AI concepts typically reserved for robotics into complex models of local and regional water use
Marta Zaniolo will join the faculty of the Department of Civil and Environmental Engineering (CEE) in Duke University’s Pratt School of Engineering, beginning January 1, 2024. Working on water resources management at scales both large and small, Zaniolo combines an innovative mixture of hydrology and climatology with machine learning and data mining to make better decisions about water use.
Zaniolo earned an undergraduate degree in environmental and land planning engineering from Politecnico di Milano in Italy, where she also finished her PhD in information technology. As a postdoctoral researcher at Stanford University since 2020, she has married the two disciplines into research combining environmental, climate and hydrologic disciplines with machine learning and evolutionary computation.
“One of the novelties of my research is the idea of using control and optimization techniques more typical of robotics and applying them to issues in water resources,” Zaniolo said. “I also look for ways to incorporate new information that hasn’t been applied to these types of models before.”
“One of the novelties of my research is the idea of using control and optimization techniques more typical of robotics and applying them to issues in water resources. I also look for ways to incorporate new information that hasn’t been applied to these types of models before.”
For example, Zaniolo worked on a new dam construction project between Ethiopia and Kenya, where many different stakeholders would be impacted by its use. Her projections had to account for the amount of hydropower generated, the volume of water released downstream, the levels of water in the river’s terminal lake, and how all these factors would affect surrounding tribes and growing communities in both countries.
To balance these competing priorities, she trained her computer model through a reinforcement learning technique typically reserved for robotics. Her model was essentially free to explore many options in the operations of the dam and was “rewarded” every time its decisions resulted in positive outcomes. Through many iterations, the model eventually trained itself to be the best steward of the water resources that it could be.
Zaniolo also gave the model access to information not commonly incorporated into such systems, even going so far as to help it predict changes in that year’s precipitation estimates through oceanic water temperature sensors that underlie forecasts for the global effects of El Niño.
“But I can only mine that information if the control and optimization techniques in my model are sophisticated enough to be able to handle it in a meaningful way,” Zaniolo said.
“I love areas that have a rich academic and research environment that also feel quirky and fun—the combination is dynamic. I interviewed at several universities, but after visiting Duke, I knew it was the right place for me. I’m so happy to join this fast-growing department and community.”
On a smaller scale, Zaniolo has worked with the city of Santa Barbara, California to help plan its water supply portfolio, making it more resistant to drought in the future. That’s a topic that she sees as being important for the future of North Carolina, as well. While overall rainfall may remain relatively stable, she says many climate change models predict longer periods of drought for the state, with large rain events in short periods of time. Zaniolo believes local municipalities should begin working together toward finding ways to better capture that water and store it for longer periods of time.
At Duke, Zaniolo plans to continue her research by generalizing the tools that she’s created so that a larger number of cities and towns can make use of them. She also wants to work toward finding methods for determining the best use of emerging sensor technologies.
“Developing models to understand if cheaper but more widely distributed sensors would be more valuable than continuing to increase the accuracy of a more limited supply of sensors is a current blind spot in the field,” Zaniolo said.
Aside from the research opportunities presented by the interesting and complex relationships between the various municipalities and utilities in the Triangle, Zaniolo said she was drawn to Duke because the Triangle is a special place. Having long lived in earthquake-prone California, she loves the classic red brick architecture more suitable to the area’s tectonic history. She also plans to make use of the warm climate to enjoy the outdoor scenery through her love of biking and hiking.
“I love areas that have a rich academic and research environment that also feel quirky and fun—the combination is dynamic,” Zaniolo said. “I interviewed at several universities, but after visiting Duke, I knew it was the right place for me. I’m so happy to join this fast-growing department and community.”