Researchers from The University of Osaka use AI techniques and street view images to visualize the structural and seasonal characteristics of urban vegetation, allowing city planners to enhance the year-round benefits of urban green spaces

Osaka, Japan – The benefits of urban green spaces in cities, in terms of ecological sustainability, climate modification, and human well-being, have been known for decades. More recently, additional economic and restorative payoffs from diverse and colorful plantings have been recognized. Now, a research team from Japan has developed a new method to identify vegetation color, structure, and seasonal changes in urban settings.

In a study published in Landscape Ecology, researchers at The University of Osaka reveal an innovative approach to capture seasonal changes in urban plant species. This method combines artificial intelligence (AI) techniques and street view imagery to allow planners to improve the visual appeal of urban green spaces throughout the year.

“Diversity, in both plant color and species, seems to enhance the ‘feel-good’ factor of urban green space for city residents and visitors,” says Anqi Hu, lead author of the study. “Our aim was to develop a method to visualize urban vegetation configuration and seasonality in much greater detail than before.”
The method integrates AI in the form of deep learning and 3D reconstruction technology with street view imagery, which significantly improves the accuracy and consistency of urban vegetation analysis. The effectiveness of the technology was tested on streets in Suita City, Osaka Prefecture, and applied to a virtual park design.

The Seasonal Species-Specific Plant View Index can distinguish between 51 urban plant species with an average accuracy of 82.17%. Plants with highly seasonal visual impacts such as cherry blossoms in spring and maple leaves in autumn can be singled out. This level of detailed modeling and identification is almost impossible with conventional green view analysis.

“Our approach removes the distortion and gaps in coverage from street view images, allowing the automatic generation of standardized viewpoints in space and over time,” explains senior author Tomohiro Fukuda. “The technique will assist in the restoration of brownfield sites or improve existing parks using a diversity of plant shapes, colors, and growth patterns.”

The framework offers a new perspective on 4D urban design and forms a foundational technology to support future urban green space evaluation and planning.

“Urban planners can extend the economic, ecological, and well-being benefits of vegetation using a range of plants that add pops of color and interest throughout the year,” says Hu.
###
The article, “Multi-temporal analysis of urban vegetation using deep learning and 3D reconstruction", has been published in Landscape Ecology at DOI: https://doi.org/10.1007/s10980-025-02090-4.