IRIS researchers are tuning in to how riverine infrastructure shapes the landscape–and soundscape–of ecosystems.
As part of a NASA-funded research effort, assistant professor Charles van Rees and his research team have deployed automated recording units, or ARUs, across study sites on the Missouri River to listen in on the local wildlife with the help of some ecologically-informed artificial intelligence tools.
“They’re essentially programmed to record wildlife sounds at certain times of day,” van Rees explained, “[for birds] at dawn and dusk, for singing frogs and toads in the middle of the night, and for bats at night.”
Why do they need all this noise?
The overarching goal of this project is to evaluate how levee setbacks, a form of natural infrastructure that helps prevent flooding, provide habitat for different species.
While traditional levees are built very close to riverbanks with the goal of containing high water levels, these levees are “set back” from the banks, widening the river’s floodplain to give water room to spread out. This reduces strain on levees, reducing the risk of failures, and helps reduce flood risk. It also restores important habitats that rely on regular flooding.
“We have very strong reason to believe that setbacks restore the ecohydrology of floodplain ecosystems,” said van Rees, “but the amount of full-on evidence we have that this benefits wildlife is kind of piecemeal.”
By monitoring and mapping species before, during and after a levee setback, the team and their collaborators at the University of South Dakota, University of Missouri, and the US Army Corps of Engineers intend to fill the gaps.
In addition to the numerous forms of data collection and active monitoring the team is enacting across their field sites, the automated recordings provide a passive source of species monitoring over long periods of time.
“It’s a great way to be able to collect a lot of data without having to have a person go out and actually spend hours and hours in the field,” said Master’s student Aurora Fowler, who spent this past summer on the ground in Nebraska, Iowa, and Missouri collecting data in person. “We’re starting to work on creating an analysis pipeline to work through analyzing the audio data and incorporating that data into the project.”
The hours and hours of recordings are a lot to listen to, which is why the team is utilizing AI algorithms, such as the one developed by the Cornell Lab of Ornithology, BirdNet: “Birdnet can analyze all those recordings and identify species, with a little help from human experts,” said van Rees.
These models use a graph of the audio frequencies in a recording to highlight known frequencies of different bird calls, identifying each species by their individual frequency signature.
“If you’re looking for, for example, an American Robin,” explained Fowler, “it knows the picture of that visualized sound, and can go through your audio files and highlight the subset where it sees American Robins. It’s a great way to collect data without having to go out and actually collect that data in-person.”
Over 30 ARUs were deployed across spring 2024 and 2025, providing a huge amount of data already. The project involves a full array of data collection methods with the goal of combining them into a tool for freshwater managers, helping predict the potential impacts and benefits of a levee setback before it’s implemented.
The interdisciplinary team includes geographers and engineers to help create a model for these predictions, which ecologists then use to help predict how wildlife will respond. “If you can tell us what the water is going to do, we can tell you what the wildlife might do,” said van Rees.
In addition to the wildlife data being collected on the ground and through the ARUs, the team is utilizing machine learning models to classify and map habitat types across the region. This map of habitat types will be used for predicting future conditions.
“Our team is using a Geospatial AI model,” explained Rabindra Parajuli, a postdoctoral associate leading the project’s computational analyses. “It learns the spatial relationships of existing vegetation and land covers, including the hydrological and topographical conditions, to predict future habitats under levee setback scenarios.”
Predicting the potential benefits of levee setbacks allows land managers, such as project partner US Army Corps of Engineers, to better make the case for this infrastructure solution in the future.
We can’t predict the future, but thanks to this team’s research, we’re listening–and responding–to the potential of levee setbacks.
This work is part of an ongoing project exploring the biodiversity impacts of levee setbacks. Learn more about this project here.