Urban restoration is particularly difficult because cities are simultaneously damaged ecosystems and densely inhabited human environments. Rivers may be channelized, soils sealed beneath asphalt, wetlands replaced by buildings and natural habitats fragmented into isolated patches. At the same time, urban residents depend on the same spaces for housing, transport, cooling, recreation and economic activity. A restoration project that improves biodiversity but increases flooding risk, displaces vulnerable communities or restricts access to public space can produce new problems while solving old ones. AI offers a way to process vast and rapidly changing information about these systems. Satellite imagery, drone surveys, environmental sensors, weather records, mobility data and community observations can be combined to reveal patterns that would be difficult to detect through conventional fieldwork alone.

Technically, the value of AI lies in its ability to identify relationships within complex, high-dimensional datasets. Machine-learning models can classify land cover, detect changes in vegetation, estimate surface temperatures and map habitat fragmentation from images. Time-series algorithms can analyze air pollution, rainfall, soil moisture and water quality as they fluctuate across neighborhoods. More advanced models may simulate how different restoration strategies influence ecological and social outcomes over time. For example, an urban planning system could compare the likely effects of restoring a stream, expanding tree canopy or converting vacant land into a wetland. Such models do not “understand” ecosystems in the human sense, and they do not eliminate uncertainty. Instead, they calculate patterns and probabilities from available data, allowing decision-makers to explore possible futures before committing resources on the ground.

The study’s central contribution is its emphasis on social–ecological–technological interactions. This perspective treats AI not as a neutral machine placed above society, but as part of the urban ecosystem it is intended to manage. The data used to train an algorithm are collected through institutions, technologies and human choices. If some neighborhoods have dense sensor coverage while others are poorly monitored, an AI system may produce more accurate recommendations for affluent areas and weaker conclusions for communities already exposed to environmental risks. Historical data can also reproduce earlier planning inequalities. A model trained on past decisions may interpret unequal access to parks, clean water or cooling infrastructure as a normal pattern rather than a problem requiring correction. In this context, technical accuracy alone is not enough; the quality, representativeness and governance of data become ecological and political questions.

AI could also transform the way restoration is monitored after a project is completed. Traditional assessments may rely on periodic surveys that capture only brief snapshots of an ecosystem. Automated image analysis and sensor networks could provide continuous information about plant survival, invasive species, wildlife activity, soil conditions, water flows and heat patterns. This would make restoration more adaptive. If a newly planted corridor fails to support expected biodiversity, managers could adjust species selection, irrigation or habitat design rather than waiting years for a final evaluation. Real-time monitoring could be especially important as climate change intensifies heatwaves, extreme rainfall and drought. However, the study’s framing implies that more data should not automatically lead to more intervention. Ecosystems are dynamic, and managers must distinguish meaningful ecological change from short-term variation, sensor errors or artifacts created by the algorithms themselves.

The most visible promise of AI may be its ability to connect ecological information with public decision-making. Digital platforms could help residents visualize how a proposed green space might reduce local heat, absorb stormwater or support pollinators. Natural-language systems could translate technical assessments into accessible explanations, while participatory mapping could allow communities to identify flooding, pollution or unsafe conditions that official datasets overlook. These tools may broaden participation in restoration planning, but they can also create the illusion of inclusion if public feedback is collected without influencing final decisions. Community-generated data raise questions about privacy, consent and ownership, particularly when information reveals movement patterns, health conditions or the locations of culturally significant sites. A socially responsible AI system must therefore be designed not only to gather more information, but also to determine who controls it and who benefits from its use.

The researchers’ focus also highlights a danger that accompanies technological enthusiasm: the temptation to treat AI as a substitute for ecological knowledge and public institutions. Algorithms can optimize a selected objective, but the choice of that objective is a human judgment. A system designed to maximize carbon storage may favor fast-growing vegetation while overlooking native species or water demand. A model focused on reducing urban heat might recommend tree planting in locations where underground infrastructure, land ownership or maintenance capacity make the proposal unrealistic. A platform intended to identify restoration priorities could rank sites according to measurable indicators while missing historical, cultural or emotional values that communities consider essential. AI can support decisions, but it cannot determine what a city ought to value. That responsibility remains with residents, scientists, planners and policymakers.

The article arrives as cities worldwide search for strategies that can address biodiversity loss, climate stress and environmental inequality at the same time. Its message is likely to resonate because it places artificial intelligence inside a much larger transformation: the shift from isolated restoration projects toward continuously managed urban ecological networks. Yet the success of that shift will depend on safeguards as much as on computational power. Transparent models, open methods, independent evaluation and clear accountability will be needed when AI-supported recommendations influence land use or public spending. Restoration systems should be tested across different climates, neighborhoods and social conditions rather than validated only in data-rich locations. Human expertise must remain central, especially when models confront unfamiliar ecological conditions or conflicting community priorities. Used carefully, AI could help cities see hidden connections and respond faster; used carelessly, it could automate old biases at unprecedented speed.

By reframing urban restoration as a partnership among ecological science, social knowledge and digital technology, Zhai, Bach, Ghazoul and their co-authors point toward a future in which cities are managed less like collections of infrastructure and more like evolving ecosystems. The challenge is not to make nature obey an algorithm, but to use computation to improve how societies observe, discuss and repair the environments on which they depend. That distinction may determine whether AI becomes another layer of urban control or a tool for more resilient and inclusive restoration. As the technology races forward, the most important question is not whether artificial intelligence can reshape cities. It is whether cities can shape artificial intelligence around ecological limits, democratic participation and the long-term well-being of both human and nonhuman life.

Subject of Research: Artificial intelligence and urban ecosystem restoration

Article Title: Artificial intelligence for urban ecosystem restoration: reshaping social–ecological–technological interactions

Article References: Zhai, X., Bach, P.M., Ghazoul, J. et al. Artificial intelligence for urban ecosystem restoration: reshaping social–ecological–technological interactions. npj Urban Sustain (2026). https://doi.org/10.1038/s42949-026-00453-7

Image Credits: AI Generated

DOI: 10.1038/s42949-026-00453-7

Keywords: Artificial intelligence, urban ecosystem restoration, urban sustainability, social–ecological–technological interactions, machine learning, biodiversity, climate resilience, environmental governance

The model distinguishes itself by estimating the operational carbon emissions of individual buildings at vast city-wide scales, surpassing previous methodologies that often depended on proprietary datasets, which hindered flexibility and global applicability. Department of Architecture PhD candidate Winston Yap, who led the study, emphasized the model’s adaptability, explaining how it can be utilized seamlessly across cities with varying data availability. This breakthrough opens new avenues for cities worldwide, especially those lacking comprehensive carbon accounting infrastructures, to track their building emissions with rigor and precision.

Applying this model to over 500,000 buildings across five diverse urban environments – Singapore, Melbourne, Manhattan in New York City, Seattle, and Washington DC – the researchers achieved remarkable explanatory power, accounting for as much as 78% of emission variations within these cities. This achievement represents a major technical milestone, harnessing a blend of open data sources such as satellite imagery, street-level photos, population maps, and climate data. These diverse inputs fuel a sophisticated graph neural network, a cutting-edge deep learning technique capable of capturing complex spatial interdependencies between urban elements.

What sets this AI-driven model apart is its ability to dissect the intricate interplay between urban form, socioeconomic factors, and energy consumption, highlighting nuances that more simplistic analyses often overlook. The research reveals that building emissions are influenced far beyond just physical size or density. Instead, they are intimately shaped by localized factors — including urban planning legacies, microclimates, and economic conditions — changing how energy is consumed in unique urban fabrics and neighborhoods.

A particularly striking finding of the study pertains to the complex relationship between density patterns and emissions. The data suggests that while taller, densely clustered buildings generally achieve better energy efficiency per square meter, dense urban cores are also subject to intensified cooling demands linked to urban heat island effects. Intriguingly, suburban regions characterized by sprawling low-rise developments contribute disproportionately to total carbon emissions, rivaling city centers in their environmental impact. These insights challenge traditional assumptions and signal the need for multifaceted urban sustainability policies.

Furthermore, the investigation uncovered alarming disparities in emissions intensity across socioeconomic strata. In most cities assessed, affluent neighborhoods demonstrated significantly higher per capita emissions compared to lower-income areas. Manhattan’s data was telling; a mere handful of large, luxury buildings accounted for over fifty percent of all building-related emissions in the city. This deepens ongoing policy debates about environmental justice, underscoring the risk of uniform carbon pricing accidentally placing disproportionate burdens on economically vulnerable communities that often reside in less efficient housing stock.

Assistant Professor Biljecki elucidated the stakes of these inequities: “Uniform carbon pricing or blanket regulations risk placing an unfair burden on lower-income communities that may already be struggling with older, less efficient infrastructure.” This recognition compels a shift toward place-based strategies that account simultaneously for carbon intensity and social vulnerability, promoting climate action that is both effective and socially equitable.

Technically, the integration of various geospatial datasets is orchestrated through graph neural networks (GNNs), which excel in modeling relational and spatial data. GNNs enable the model to capture not just isolated building attributes but also the relational context within the urban landscape—such as proximity to roads, neighboring building types, and the configuration of green spaces. This holistic spatial understanding allows the model to predict emissions more precisely by considering the multifaceted interactions within densely interconnected city systems.

The open-source nature of the model is a deliberate and meaningful choice by the researchers, aiming to democratize access to advanced urban carbon accounting tools. By relying only on publicly available data and releasing their codes openly, the team paves the way for cities worldwide—including those with limited resources or restricted data environments—to participate in global decarbonisation efforts. This openness resonates with the principles of open science, fostering transparency, collaboration, and acceleration of research impact.

In practical terms, city governments and urban planners equipped with this model can perform detailed emissions audits, pinpointing hotspots and identifying the specific drivers of carbon intensity at the building level. This spatially precise intelligence facilitates the design of targeted interventions, such as prioritizing energy retrofits in identified high-emission districts or adjusting zoning codes and urban planning regulations to mitigate heat island effects and optimize building efficiency.

Moreover, the inclusion of socioeconomic data helps to align climate action with social equity objectives, enabling policymakers to tailor incentives or support mechanisms where they are most needed. Such data-driven, localized approaches are vital to craft fair decarbonisation pathways that avoid exacerbating existing social inequalities while achieving ambitious sustainability goals.

The collective implications of this research underscore a critical narrative: urban sustainability is inherently complex and context-dependent, requiring sophisticated analytical tools that move beyond aggregate city-wide metrics to embrace fine-grained spatial heterogeneity. The fusion of AI, geospatial data, and urban science in this project represents a paradigm shift toward smarter, fairer, and more actionable carbon management.

Looking forward, the potential applications of this framework extend beyond carbon accounting alone. The modeling techniques could be adapted to estimate other environmental burdens of urban environments or integrated into digital twins of cities for dynamic, real-time sustainability planning. The research team’s commitment to open science invites continual refinement and collaborative expansion, promising a future where cities can not only understand their carbon footprints but actively navigate the complexities of sustainable transformation.

For stakeholders invested in combating climate change at the urban scale, this research is a beacon of innovation and equity—demonstrating how open data and AI can unlock new frontiers in sustainability science. As cities continue to grow and evolve, such tools will be indispensable for meeting climate commitments in ways that are not only efficient but just.

Subject of Research: Not applicable

Article Title: Revealing building operating carbon dynamics for multiple cities

News Publication Date: 15-Aug-2025

Web References: http://dx.doi.org/10.1038/s41893-025-01615-8

Image Credits: College of Design and Engineering at NUS

Keywords: Urban planning, Climate change mitigation, Architectural design, Carbon emissions