The AI ​​system accurately maps urban green space, highlights environmental division

A research team led by Rumi Chunara – an NYU Associate Professor, with appointments in both Tandon School of Engineering and School of Global Public Health – has unveiled a new Artificial Intelligence (AI) system that tracks urban green space To do satellite uses imagery. Important to ensure healthy cities, compared to previous methods.

To validate its approach, researchers tested the system in the largest city in Karachi, Pakistan, where several team members are based. Karachi proved an ideal test case with a mixture of dense urban areas and separate vegetation conditions.

Accepted Publication From ACM Journal on Computing and Sustainable SocietyThe team analysis exposed an environmental division: some areas enjoy trees and interested roads while many neighborhoods have almost no vegetation.

Cities have long struggled to track their green spaces properly, from parks to individual road trees, traditional satellite analysis is missing from about 37% urban vegetation.

As cities face climate change and rapid urbanization, especially in Asia and Africa, accurate measurements have become important. Green space can help reduce urban temperature, filter air pollution and provide the necessary space for exercise and mental health.

But these benefits can be disbursed unevenly. Low-income areas often lack vegetation, causing them to become warm and more polluted than in trees and rich rich neighborhoods.

The research team developed its solution by increasing the AI ​​segmentation architecture, such as Deeplabv3+. Using the high-resolution satellite imagery from Google Earth, they trained the system by increasing their training data to include various versions of green vegetation in various lighting and seasonal conditions-a process that they say “green growth” Are. This technique increased the accuracy of vegetation detection by 13.4% compared to existing AI methods – a significant advance in the area.

When measuring how often the system correctly identifies the vegetation, it achieved 89.4% accuracy with 90.6% credibility, much better than traditional methods that achieve 63.3% accuracy with only 64.0% credibility.

“The previous methods depended on the simple light wavelength measurements,” Chunara said, who serves as the director of the NYU Center for Health Data Science and is a member of the Visulesation Imaging and Data Analysis Center (VIDA) of NYU Tandon.

“Our system learns to recognize more subtle patterns that distinguish trees from grass, even in challenging urban environment. This type of data is required for urban planners to identify the neighborhood in which vegetation in which vegetation Lack of them so that they can develop new green locations that will provide the most benefits, without accurate mapping.

The Karachi analysis found that the average per capita of the city is only 4.17 square meters of green space, which less than half the World Health Organization (WHO) recommended a minimum 9 square meter per person. Inequality within the neighborhood is dramatic: while some external union councils – the smallest local government bodies of Pakistan, a total of 173 were included in the study – more than 80 square meters per person, five union councils per capita less than 0.1 square meters per person per person Is.

Studies have shown that areas with more paved roads – usually a marker of economic growth – is for more trees and grasses. More importantly, studied in eight separate union councils, regions with more vegetative surface temperatures were shown to be low, with the role of green spaces in cooling cities.

Singapore provides a contrast, showing what is possible with the intentional plan. Despite having a population density similar to Karachi, it offers more than the WHO target, 9.9 square meters per person.

Researchers have made their functioning public, although it will require to re -form the system on local satellite imagery to implement it in other cities.

This study connects the body of the development of computational and statistical methods including data mining and machine learning to understand social determinants of health and health inequalities. Precious studies include using social media posts to map neighborhood-level systemic racism and homophobia and assess their mental health effects, as well as record electronic health to understand telemedicine access disparities during covid-19 Analysis involves.