Purdue AI urban tree monitoring and analysis initiative to improve city life

WEST LAFAYETTE, Ind. — With 89% of the U.S. population and 68% of the world population projected to live in cities by 2050, concerns mount about how to address health and environmental issues such as excessive heat, poor air quality and rainwater runoff. Urban trees have the potential to improve these issues, but first, accurate tree inventories and information must be obtained, analyzed, shared and regularly updated to make effective data-driven decisions.

This is the ambitious goal of a project led by Daniel Aliaga, Purdue computer science associate professor and principal investigator, with Songlin Fei, professor and director of the Purdue Institute for Digital Forestry, as the forestry lead.

Aliaga and Fei have received a $5 million fund from the National Science Foundation (NSF) for uTREE (Urban Tree Resilience and Environmental Equity), a project that focuses on using AI to obtain information about tree density, species and location, and changes in tree counts over time and events. The uTREE team will achieve this goal using the first cloud-based cyberinfrastructure (CI) to support a national (and potentially worldwide) urban tree inventory. The CI will be used in multidisciplinary urban science, engineering and planning to create and maintain safer and more livable cities.

“The uTREE CI is scalable to serve a large and multidisciplinary community of researchers and urban practitioners,” Aliaga said. “In collaboration with the Institute for Physical Artificial Intelligence, a Purdue Computes initiative, we will also demonstrate how an AI-based approach can be successful in operating in previously unseen urban areas.”

The CI fills a major gap. “Researchers — armed with the increased computational power and advancement of recent novel urban canopy models, urban ecological models and urban social science linkages — can now provide improved predictions,” Aliaga said. “However, the lack of high-resolution tree data, particularly in urban and under-resourced communities, have previously remained a significant barrier.”

Disadvantaged communities disproportionately suffer from the negative fallout of rapid urbanization, industrial development, lasting environmental inequities, and absent or misinformed codes and ordinances in cities and towns, said Fei, professor and Dean’s Chair in Remote Sensing in the Department of Forestry and Natural Resources. 

“uTREE, in partnership with local organizations already working on these issues, will advance environmental justice and grow local capacity to ensure lasting impacts in urban and community forests nationwide,” Fei said.

Other Purdue team members working on uTREE are: co-principal investigator Brady Hardiman, an associate professor of forestry and natural resources and environmental and ecological engineering; professor Zhao Ma from forestry and natural resources; professors Melba Crawford and Panagiota Karava from civil engineering; Matthew Huber, professor and director of Purdue’s Institute for a Sustainable Future; and co-principal investigator Rajesh Kalyanam, a computer engineer and senior research scientist at the Rosen Center for Advanced Computing.

Aliaga and his colleagues have already tested their CI. The team developed an initial tree localization foundation model yielding a preliminary dataset comprised of more than 278 million trees, covering 8.2 million acres in 330 U.S. cities. Aliaga estimates the uTREE CI will serve a community of 60,000 researchers and half a million urban practitioners.

The innovation and technology are novel, but it’s the potential impact on which Aliaga and his colleagues stay focused.

“Using our AI-based method, we can now update urban tree inventories with 92.5% count accuracy,” Aliaga said. “Instead of relying solely on high-spatial resolution, the uTREE generative AI modeling approach draws from data collected repeatedly over time. It’s automatic and repeatable so the monitoring for any city can be computed in a few hours and quickly repeated at will. We also definitely anticipate timely tree inventory and localization data will be used to improve weather/climate modeling.”

Aliaga explains why this innovation has deep roots at Purdue. “Our team worked together on prior NSF projects that laid the groundwork for this larger effort and we are members of the Institute for Digital Forestry at Purdue, which draws on faculty expertise from multiple colleges. Our project also makes use of the MyGeoHub utilities at Purdue, which permit jump-starting such a large-scale software infrastructure providing data and tools catered to communities ranging from domain scientists, to CI users/experts, to practitioners.”

The prototype work was completed through a multidisciplinary collaboration. Adnan Firoze, a computer science PhD student, led the graduate student development team that included computer science student Akshaj Uppala and forestry and natural resources student Lindsay Darling. The collaboration also included computer science assistant professor Raymond Yeh and professor Bedrich Benes.

“It is extremely gratifying to work in such an eclectic domain and know it will have an incredible impact on the natural world,” Firoze said. “Seeing the state of trees so precisely and on a national scale across time is fascinating and the first work of its kind. It is like a time machine for trees!”

While they collaborate with professionals across disciplines and the country, the team is also preparing the next generation of urban computational and ecological researchers and managers as they leverage the Purdue Data Mine, a learning community for undergraduates, to address the educational and experiential gap in urban computational and forest management tools.