AI Unlocks Hidden Patterns In Ecosystems From Africa To The Amazon

Comparing Biological Networks

One project introduces a new way to compare biological networks—the webs of interaction among species that underpin every ecosystem. The goal is to identify structural similarities between ecosystems in different regions, even when they comprise completely different species. Such comparisons can inform large-scale monitoring of ecosystem health and guide conservation priorities. Traditional methods, however, often struggle with data this complex.

Uribe, together with Lydia Beaudrot at Michigan State University and other colleagues, applied a new class of mathematical tools known as optimal transport distances to analyze over a hundred African mammal food webs across six different regions on the continent.

Optimal Transport Distances

Optimal transport describes the minimum amount of work needed to transform one object into another. If each object is represented by a mound of dirt, then optimal transport, or "earth mover's" distance, represents the most efficient way to move dirt around so that the two mounds become analogous.

In ecology, each network of species interactions can be thought of as one of those "mounds." Optimal transport distances enable researchers to align the overall structure of two networks, illustrating how their patterns of connection compare, even when the networks comprise different species.

Using these tools, the researchers analyzed data from multiple sources and identified functionally equivalent species, i.e., different species that play the same ecological role in their respective ecosystems.

"This allows us to determine, for example, if the lion in this food web plays the same role as the jaguar in this other one or the leopard in this other one," Uribe said.

Former Rice undergraduates Kai Hung, now a doctoral student at the Massachusetts Institute of Technology, and Alex Zalles, pursuing a doctorate at the University of California, Berkeley, led the effort to quantify the ecological data.

"They did so well here at Rice that they were recruited into the top programs in the nation," Uribe said. "It really speaks to the caliber of training and undergraduate research experience we provide."

Bioacoustics in Tropical Forests

An earlier project focused on the tropical forests of Colombia and used sound to map biodiversity. Led by Maria Guerrero, a doctoral student from Colombia who will join Rice this fall as a visiting scholar on a Fulbright Scholarship, the study deployed 17 microphones across a range of habitats within a Colombian oil palm plantation. Over the course of 10 days, the team recorded hundreds of hours of sound, capturing the calls of frogs, birds, and insects.

Through AI analysis, the researchers created what Uribe called a "tropical forest connectome," borrowing a term from neuroscience to describe the interconnectedness of different areas of the forest through sound.

"Instead of connections inside the brain, we were looking at the connections in the tropical forest, how information and energy flows," Uribe said. "We were using bioacoustics data as a proxy to understand the health status of an ecosystem. The novelty here is being able to automatically identify and segment the sounds."

The results showed that habitat matters more than distance: Two patches of intact forest can sound alike even when far apart, while a forest and a nearby region planted with oil palms may be completely different. The study confirmed that converting native forests to monoculture plantations drastically reduces biodiversity, highlighting how bioacoustics can serve as a low-cost tool for large-scale monitoring.

Personal and Global Impact

For Uribe, who is from Colombia, the project carried special weight.

"It is personally meaningful because I am doing research that has global impact, using techniques that I am developing here in the United States with many local, regional and international collaborators," Uribe said. "In terms of impact, both papers are meaningful because the research entails applying AI for something other than maximizing profit or gaining a competitive edge: This is AI for ecology and conservation."

Publications and Support

Both papers are published in Methods in Ecology and Evolution, the leading journal in the field.

For the study of African mammals' food webs, the research was supported by the National Science Foundation (2211815, 2213568, 2443064) and Google. For the bioacoustics study, the research was supported by the Universidad de Antioquia, the Alexander von Humboldt Institute for Research on Biological Resources, the National Science Foundation (2213568, 2443064), and Rice, with data from Puerto Wilches funded by the Universidad de Antioquia, SGI, and Ecopetrol under contract FOGR09.