USC Researchers Unveil Low-Cost Machine Learning Tool to Map Urban Tree Canopy Using Free Aerial Imagery

In a paradigm-shifting development for urban climate resilience, researchers at the University of Southern California (USC) have unveiled a groundbreaking, low-cost machine learning tool that maps urban tree canopy using free aerial imagery instead of costly lidar surveys datainnovation.org . Published in the Center for Data Innovation's "10 Bits: The Data News Hotlist" on July 10, 2026, this conspicuous technological advancement provides municipalities with an affordable mechanism to target tree planting, expand shade, and make smarter investments in climate resilience datainnovation.org . Bypassing the Dilemma of Expensive Lidar Historically, producing fine-scale tree canopy maps has required exorbitant lidar surveys, which use lasers to produce highly detailed 3D landscape maps, or proprietary commercial satellite imagery datainnovation.org . The USC-developed tool eliminates this financial barrier by leveraging publicly accessible aerial photographs collected nationwide every two to three years through the U.S. Department of Agriculture’s National Agriculture Imagery Program (NAIP) datainnovation.org . By synthesizing these images with advanced machine learning algorithms, the researchers have drastically reduced the cost of producing detailed tree canopy maps, making the technology practical for cities that may lack the resources for specialized surveys datainnovation.org . Rigorous Testing Across Diverse Urban Landscapes To validate the model's accuracy, the researchers initially deployed the system in Boyle Heights and City Terrace, two densely populated, majority-Latino neighborhoods east of downtown Los Angeles that have historically experienced less tree cover than wealthier parts of the city datainnovation.org . The canopy-mapping model accurately identified tree cover, while the individual tree detection model—a more challenging task because tree crowns appear small and often overlap in aerial imagery—performed competitively with far more expensive lidar-based approaches datainnovation.org . Crucially, to test whether the approach could generalize beyond Southern California, the researchers applied the trained models, without additional retraining, to neighborhoods in San Francisco and Phoenix datainnovation.org . Despite the cities’ distinct climates and urban layouts, the tool produced consistently strong results, suggesting communities elsewhere may be able to use the model and avoid building one from scratch datainnovation.org . Ubiquitous Adoption and Future 3D Integration The enthusiasm for the technology is already burgeoning. The ArcGIS deep learning package developed through the project has been downloaded more than 12,900 times from Esri’s Living Atlas platform over the past six months, according to the research team datainnovation.org . The research, code, and a ready-to-use ArcGIS deep learning model are freely available online, making the tool accessible to municipalities without in-house machine learning expertise datainnovation.org . Looking ahead, the team’s next step is to pair its AI tool with freely available lidar data that captures the height and three-dimensional structure of tree canopies datainnovation.org . “Knowing both the height and extent of the canopy will allow us to estimate the shade trees provide today—and model how much additional shade new plantings could create,” the researchers articulated datainnovation.org . This prescient integration will allow communities to analyze individual street blocks, school playgrounds, and parks before expanding the approach to entire counties, giving them even more precise information for planning cooler, healthier, and more resilient urban environments datainnovation.org .