Introduction

Cities are at the forefront of current climate challenges and will bear a substantial burden of future impacts. Home to more than half of the world’s population and emitting ~70% of global CO2 emissions, urban areas are taking action to both adapt to climate change and curb greenhouse gas emissions1,2. The COVID-19 pandemic demonstrated that cities have the ability to rapidly mobilize to enact solutions during times of global crisis3,4. Unfortunately, addressing climate change often takes a backseat in urban areas, stymied by roadblocks caused by siloed government agencies, inefficient financing, and unequal access to knowledge, as well as pressures to meet a range of competing policy demands and concerns5.

The Intergovernmental Panel on Climate Change Sixth Assessment Report (IPCC AR6)6 defines solution space as, “… a set of biophysical, cultural, socio-economic and political-institutional dimensions within which opportunities and constraints determine why, how, when, and who acts to reduce climate risks”. Following this understanding, what constitutes an urban climate solution is contingent on such a set of dimensions that shape and intersect within a given locality, therefore defining its feasibility.

Urban climate change solutions thus need to be developed on a case-by-case basis, as cities and city actors tailor suitable responses to their current and expected conditions, needs, priorities, and capabilities. Assessing case studies is essential to deepen understanding of how multiple and diverse climate actions may—or may not—translate into effective solutions on the ground. It also provides insights into how these solutions are operationalized within political, economic, and sociocultural contexts. How they may be scaled, i.e., how they become systemic solutions implemented across many different urban areas—is a central challenge that needs to be tackled in order to dramatically reduce vulnerability and greenhouse gas (GHG) emissions in sustainable, zero-carbon, resilient, and just cities.

Many cities and communities worldwide have emerged as centers of innovation and implementation for climate solutions. City networks, city-focused assessments, and city-generated reports are valuable mechanisms to generate and transfer new knowledge to researcher and practitioner communities. Urban-focused climate change research is rapidly growing7. In a machine learning-aided analysis of the urban climate literature, Montfort et al. 8 found a total of 55,000 articles, 85% of which were published between 2012 and 2022. The analysis revealed that this research was biased to high-emitting cities, typically leaving out small or medium-sized cities8. Overall, urban climate literature supports two widely-held findings: 1. City case studies provide a key mode of new knowledge generation and exchange, and are a critical base for researcher analysis and for practitioner action; and 2. Case studies in the research and practitioner communities are over-represented by those in the English language and those from larger cities in the Global North8.

The Urban Climate Change Research Network (UCCRN) addresses these issues through the development of the UCCRN City Solutions Case Study Atlas (City CSA), a dynamic, centralized, and searchable database of urban case studies, related metadata from multiple sources across wide geographies, and relevant data layers. This article presents the conceptual design of the UCCRN City CSA including key goals, objectives, and protocols used to integrate case studies from multiple sources and provide search functionality across a range of user groups (Fig. 1). It provides examples of how users can apply metadata filters as well as artificial intelligence/machine learning techniques to search for climate action solutions applicable to their own regions and sectors. By collating diverse urban experiences, priorities, and challenges, the UCCRN City CSA provides a rigorous evidence base for city policymakers, practitioners, and civil society groups as an enabling condition for development of meaningful on-the-ground urban climate solutions informed by the latest science.

Fig. 1
Fig. 1
Full size image

UCCRN City Solutions Case Study Atlas framework.

As an evolving tool, the UCCRN City CSA continues to grow through a set of development phases (see Supplementary Table 1). Building sequentially, each phase is designed to improve on the process and structure, collect, integrate, and respond to stakeholder feedback, and enhance its usability so that the City CSA can fulfill its potential to serve as a key resource for urban climate action.

Easily accessible case studies of climate change actions in cities or metropolitan areas serve two key purposes. They help researchers better understand the linkages between the urbanization process and climate change, and enable urban policymakers and practitioners to learn from each other about climate solutions, their scope, and potential co-benefits and trade-offs9. Case studies that report on these innovative climate actions provide a means to share solutions and challenges.

Many global and regional initiatives have emerged in recent years to accelerate local climate action and knowledge exchange. Platforms like the Climate Disclosure Project, C40 Knowledge Hub, Urban Sustainability Exchange, Climate-ADAPT, and the Indiana University Environmental Resilience Institute Case Study Database have helped to facilitate city leadership to support adaptation and mitigation planning and collaborative climate governance10,11,12,13,14.

However, case studies on how cities are addressing climate change are currently distributed in a wide variety of sources, platforms, and databases, hindering both research efforts and uptake of tested solutions. As the world grapples with the 1.5 °C overshoot, research to support climate action in cities is essential15. A centralized, open-access repository for a diverse range of urban climate actions is needed via a framework that centers on equity, regional contextualization, and practical usability for academics, policymakers, and practitioners.

The UCCRN City CSA establishes its distinctiveness by addressing limitations found across other urban climate case study database platforms. The UCCRN platform employs a rigorous, established set of protocols for acceptance from all sources. It operates outside of institutional membership requirements, i.e., one does not need to be a UCCRN member to submit a case study, unlike some other platforms. Case studies are also actively solicited and curated, differentiating from self-reported or exclusive network-generated content. This establishes the UCCRN City CSA as a globally inclusive and nonmembership-based resource.

The City CSA encompasses integrated climate solutions, presenting both qualitative and quantitative content with maps of interactive data layers so that users can query both types of information. This “one-stop shop” enables the City CSA to be a valuable resource for both the scientific and practitioner communities. While other urban climate case study databases focus primarily on either mitigation or adaptation, the City CSA Metadata Template captures a wide range of critical dimensions such as climate solution drivers, hazards, funding sources, and case study provenance, permitting a more nuanced understanding of urban climate actions.

Case study sources and subsequent submission protocols inform the City CSA structure and functionality, which in turn support a range of user communities. Data flow and functionality is driven by a participatory and co-production approach that prioritizes the inclusion of diverse voices including city, regional and global climate change networks, practitioners, policy and decision-makers, civil society, and underrepresented groups such as local, migrant, and Indigenous urban communities. UCCRN invites researchers as well as all other key providers of urban climate change case studies to participate in this evolving, solution-oriented initiative. By encompassing the multi-dimensional characteristics of urban areas, the City CSA seeks to gather and present solutions and plans by cities from all regions to support informed, science-based assessments and decision-making.

The ongoing development of the City CSA is also informed by high-level scientific assessments such as UCCRN’s Third Assessment Report on Climate Change and Cities (ARC3.3) 7 and the Global Research and Action Agenda on Cities and Climate Change Science (GRAA)16. The platform is also being developed in parallel with the IPCC Special Report on Climate Change and Cities (IPCC SR-Cities).

The UCCRN City CSA emphasizes the myriad actions and plans that are already implemented or are being developed in cities and their corresponding metropolitan regions to address climate change challenges and advance urban transformational pathways that improve human and non-human quality of life. Case studies provide city stakeholders with valuable insights into strategies and interventions being executed by other cities, including the challenges and opportunities they entail, thereby enabling UCCRN City CSA users to adapt solutions to fit their own urban contexts in a more robust and informed manner. Figure 2 illustrates the design of the UCCRN City CSA Prototype, highlighting the visual interface for accessing case studies and data layers.

Fig. 2
Fig. 2
Full size image

Prototype of UCCRN City CSA with case studies in Copenhagen, Durban, Kano, Los Angeles, Melbourne, Mexico City, Monterrey, Naples, New York, Port Vila, Rio de Janeiro, Shanghai, and Singapore.

The UCCRN City CSA seeks to reduce gaps in information and data on existing actions being implemented at local city levels, particularly in the Global South. It brings together new and existing case studies, expanding on the >250 expert-reviewed case studies in the UCCRN ARC3 Case Study Docking Station (CSDS), a searchable digitized database that provides benchmarked, expert-reviewed examples of how cities from around the world have been working to mitigate and/or adapt to climate change17.

The platform includes case studies that present climate actions that have been evaluated as effective as well those that explicate barriers to effectiveness. The goal is to promote learning about which interventions are viable and can be implemented effectively, are ambitious enough to promote transformative change, and address equity and justice concerns. Criteria for assessing effective climate solutions and identifying barriers to success were co-developed by the UCCRN City CSA Core Workgroup and the Guidance Group. The Core Workgroup of the City CSA consists of researchers from Columbia University, Hunter College, the National Autonomous University of Mexico (UNAM), Aalborg University, and the University of Southern Denmark (SDU). A 25-member Guidance Group composed of academics, city policymakers, practitioners, networks, and civil society based at institutions from around the world was formed in 2024 to advise the development of the City CSA.

Climate actions are evaluated as effective based on whether they have measurable mitigation and/or adaptation outcomes, can be implementable at larger city scales (e.g., for multiple buildings, neighborhoods, cities, and metropolitan regions), are capable of delivering climate benefits on longer temporal scales (e.g., decades), and are equitable. Best practices derived from analyses of the case studies can guide financial investments, support peer-to-peer learning, foster local capacity-building, and inform the development of robust policy and planning frameworks.

Barriers to effective climate action include a lack of enabling government and policy conditions, inadequate technical capacity to implement at larger spatial and longer temporal scales, and insufficient financing. Including city case studies that reflect challenges, partial failures, or plans still in development highlights the critical importance for users to learn from limitations, setbacks, and evolving contexts, so they can anticipate barriers and strengthen future project design.

Methods

This section outlines the development of the UCCRN City CSA and the analytical methods used to evaluate its search functionality and case study content. It outlines the case study development process, metadata framework, user typologies, and analytical approaches, including metadata-based search and LLM-assisted discourse analysis applied to the City CSA prototype.

Case study development process

The development of the UCCRN City CSA follows a diverse and inclusive process that includes researchers, practitioners, and stakeholders from multiple geographies and institutional settings. Through five virtual convenings during 2024 and 2025, Guidance Group input was iteratively provided to aid with strategy, outreach, case study protocols, metadata development, data collection, and prototype interface. Along with academic and scientific expertise, the Guidance Group brought valuable practitioner insights and real-world perspectives to ensure the relevance and applicability of the UCCRN City CSA across diverse urban contexts. In 2024, UCCRN convened a webinar, co-hosted with the IPCC to conduct outreach on both SR-Cities, as well as the City CSA. Attendees included individuals from the five user groups. A survey was shared with webinar participants to gather insights on existing gaps and needs, and to understand how the City CSA can help address them. (See Supplementary Table 2 on the ways the Guidance Group and the results of the co-hosted IPCC webinar informed the development of the City CSA.)

Inclusion of case studies from underrepresented geographies and highly vulnerable sites and groups is a special focus of the City CSA. UCCRN’s 15 Regional Hubs are taking a leading role in their own geographies to ensure local participation and representation in the City CSA. The Regional Hubs are located in Aalborg, Ahmedabad, Bogotá, Cairo, Durban, Kano, Mexico City, Melbourne, Nassau, Paris, Philadelphia, Rio de Janeiro, São Paulo, Shanghai, and Tel Aviv. Hubs are linked to universities, research institutions, and governments.

The UCCRN City CSA highlights how differential exposure and intersectional vulnerabilities, such as those related to income, informality, and geography, shape urban climate risks and responses. Additionally, by casting a wide net for case studies from multiple geographies, urban contexts, and languages, the City CSA can help to address biases in existing research on climate change and cities, such as the overemphasis on peer-reviewed literature and the limited or scattered availability of case studies in other languages besides English.

The City CSA also emphasizes the inclusion of case studies from the Global South by actively seeking submissions that highlight innovative climate solutions and adaptation strategies implemented in these regions. By showcasing successful initiatives and lessons learned from these cities, the City CSA aims to foster South-South and South-to-North knowledge exchange, addressing the distinct vulnerabilities and capacities across the diverse urban contexts of the Global South while inspiring similar city actions globally.

The platform seeks to expand inclusion of case studies on climate solutions in towns and small and medium-sized cities, as well as in rapidly growing cities. Currently, urban areas make up ~55% of the global population—a proportion which is projected to grow to 68% by 205018,19. Building cases to harness the knowledge about the challenges that these smaller urban places experience as well as the solutions they are innovating is thus of great importance for building a sustainable and climate-resilient urban future. Knowledge sharing through the UCCRN City CSA on the challenges faced by large and megacities can also support small and medium-sized cities in moving beyond business-as-usual practices and fostering innovation.

While some Indigenous Peoples continue to reside in their traditional lands—some of which have since been urbanized— many others have migrated and now live in urban or periurban settings where their social practices and cultural identities persist, hybridize, and evolve20,21,22. Indigenous cultures, values, knowledge, and traditions have been mobilized to revitalize urban and peri-urban ecologies, advance more inclusive and equitable land-use planning, and enhance political, social, and cultural agency23. This includes fostering ecosystem health, supporting social and solidarity-based local economies, community-based urban regeneration projects and service provision, as well as upholding local and Indigenous Peoples' rights, including their right to the city. The City CSA is committed to including Indigenous voices and migrant communities, highlighting the numerous climate-related solutions these communities are leading in cities around the world while advancing human rights. To implement CARE Principles (collective benefit, authority to control, responsibility and ethics), an international Indigenous Knowledge and Local Knowledge (IKLK) Advisory Committee will be integrated for active engagement of IKLK actors24.

The aim of the UCCRN City CSA is to be a resource for a diverse set of stakeholders, including local and subnational decision-makers and practitioners. Now in Phase Two, the City CSA is advancing the development of case study sub-platforms in main regional languages. The first sub-platforms are in Spanish and Portuguese, covering case studies from Latin America and the Caribbean (LAC sub-platform), as well as Scandinavian languages, covering Nordic cities. Selected case studies of the LAC sub-platform are translated to English for amplifying outreach and feeding comparative analyses. The LAC sub-platform, as well as other regional sub-platforms, are embedded within the main English-based UCCRN global platform, thus sharing the same data science, search, and analytical tools. However, these sub-platforms may have their own entry web portals to provide customized spaces in those languages, focused on enhancing regional outreach and communication.

UCCRN Regional Hubs are anticipated to serve as the responsible entities managing such regional sub-platforms in coordination with the UCCRN Secretariat. In parallel, case study authors are encouraged to submit case studies in their regional languages, including the six official United Nations languages (Arabic, Chinese [Standard Mandarin], English, French, Russian, and Spanish) to support broader knowledge exchange across regions.

All case studies included in the City CSA are expert-reviewed and linked to the Global Research and Action Agenda on Cities and Climate Change Science (GRAA)16. This rigorous, multi-step, regionally grounded review and verification process distinguishes the UCCRN City CSA from other climate solution platforms that often lack independent evaluation, local contextualization, or alignment with global urban climate research agendas. To ensure comparability across regions, a standardized review form consists of whether the author(s) demonstrate a clear understanding of the case study topic; present a well-defined climate change solution within a city; and provide adequate detail and analysis regarding implementation. Prior to conducting reviews, UCCRN Regional Hub Directors and members receive guidance documents and participate in virtual orientation sessions to establish common understanding of the City CSA goals, protocols, and review process. (For more information on the expert review process, see Supplementary Material.)

City Case Study Atlas structure

The UCCRN City CSA contains both mapping and search functions. Case study cities are linked to an interactive global atlas, with cities represented via clickable icons (see Fig. 2). The map contains data layers that can be selected based on user choice. When users apply metadata variables through the search function, city icons on the map are illuminated related to the associated case studies of the search. This allows for easy visual and text-based navigation. Case studies are available as PDF documents or through direct links to their original web sources.

By integrating analyses of climate trends, downscaled projections, and remote sensing data alongside the detailed case studies, the CSA offers valuable insights into urban landscapes and environmental conditions in a rapidly changing climate. City governments can use the projections to prepare for climate risks such as increased temperature, droughts, or heavy rainfall events and integrate them into risk assessments and urban resilience planning. (See Supplementary Table 12 for 2050 temperature, precipitation, and sea level rise projections in prototype case study cities.)

The City CSA interactive global atlas provides a comprehensive set of city-accessible global data layers, empowering stakeholders with critical information to address climate change challenges. Each data layer within the City CSA offers insights for planning, decision-making, and long-term sustainability for both climate mitigation and adaptation. Key data layers are shown in Fig. 3. The platform serves as a centralized and user-friendly hub where global datasets relevant to urban climate resilience are consolidated for streamlined access and practical application. (For a list of all data layers included in the City CSA, see Supplementary Table 9.)

Fig. 3: Key data layers in the UCCRN City CSA.
Fig. 3: Key data layers in the UCCRN City CSA.
Full size image

a Hyper-resolution (3 Arc sec ~100 m) Global Human Settlement population density layer; b Hyper-resolution (300 m) YCEO-V4 urban heat island layer; c Hyper-resolution (300 m) MODIS land cover map; d Köppen–Geiger Climate Zones.

Responding to the current lack of climate projections for cities, which can significantly hinder the development and implementation of their adaptation strategies25, the City CSA includes downscaled climate projections for mean temperature and precipitation for all cities included in the UCCRN ARC3.3. These are based on the NASA Earth Exchange (NEX) Global Daily Downscaled Projections (GDDP) dataset26. The NEX GDDP includes 35 global climate models (GCMs) from the Coupled Model Intercomparison Project Phase 6 (CMIP6) for the baseline period from 1950-2014 and the future period 2015-2100. Localized climate projections for case study cities provide essential content for urban adaptation strategies. (See Supplementary Material for more information on the downscaled climate change projections.)

Metadata for each case study presented in a uniform template streamlines search functionality and enhances user experience. Key thematic categories and data fields were identified by the Core Workgroup and refined to ensure usability and effectiveness. The Metadata Template was then vetted by the broader City CSA Guidance Group. Five different metadata categories were defined: Source; Location, Timeframe, and Language; Climate, Population, and Socio-Economic Data; Content; and Financing. (See Supplementary Table 7 for complete Metadata Template.)

The City CSA Core Workgroup developed the metadata categories to create the taxonomy for the City CSA. The taxonomy expands upon and refines metadata used and tested for the case studies compiled as part of UCCRN’s CSDS, developed to support UCCRN’s Second and Third Assessment Reports on Climate Change and Cities7,27. Other taxonomy input included the GRAA and feedback provided by the Guidance Group. For example, the financing category was developed in direct response to comments by practitioners in the Guidance Group, in addition to input from authors of the UCCRN ARC3.3 Finance Element28. This process of developing the metadata balanced the goals of being comprehensive and exercising restraint, so as not to add too many sub-categories of metadata, responding to feedback that the Metadata Template could become cumbersome. (For more information on the Metadata template and structure, see the Supplementary Material.)

Potential user communities

The intended user base of the UCCRN City CSA is broad, offering impactful use cases for urban actors and stakeholders who may draw on them in varying ways (Table 1). The Guidance Group, which is made up of all six user groups, elucidated their primary objectives and how the platform contributes to their knowledge needs. The platform’s tools are designed to evolve to support these flexible and practical uses, i.e., case study searches through data map visualization, metadata filters, and Large Language Learning Model (LLM)-assisted discourse analysis.

Table 1 UCCRN City CSA user groups, primary objectives, and contributions
Full size table

Analytical approaches to search functionality

For the potential user groups, the City CSA currently supports two complementary approaches for accessing and analyzing case study content: (1) Direct User Metadata Search; and (2) Intentional AI Discourse Analysis, i.e., LLM-assisted discourse analysis search. These approaches facilitate the mobilization of knowledge, demonstrating distinct ways in which the platform content can be used in applied and research settings. The City CSA complies with three repertoires of knowledge: brokering or knowledge mobilization; supplying (with relevant expertise or appropriate experts); bridging (translating and communicating knowledge questions and answers); and facilitating (integrating knowledge production and use)29.

To compare the two approaches, a user type was defined based on discussions with the Guidance Group: Flood Manager in the Global South (in a rapidly growing city in a low-income country). Two other user types were defined in the LLM-assisted analysis: Climate Resilience Practitioner in the Global North responding to extreme heat (in a developed city of a high-income country), and Policymaker in East Asia responsible for energy transitions and GHG emissions reduction (in a rapidly growing city in a middle-income country).

Direct User Metadata Search

Accessing metadata through the keyword-based search allows users to filter case studies by sets of pre-defined variables that include categories on climate hazards, locale, population, socioeconomic status, and financing. Viewing case studies within the context of relevant temporal and spatial resolution datasets has been shown to support long-term development of urban areas by providing qualitative and quantitative datasets relevant for decision-making30. When metadata filters are applied to the search, criteria matching case studies appear, allowing for the rapid appearance and collation of case studies fitting these criteria. Similar forms of metadata search databases, such as sustainability knowledge-action platforms, also support local urban climate action planning and implementation.

Intentional AI Discourse Analysis

The use of the City CSA has the potential to be augmented by LLM-assisted discourse analysis to enhance the scale and scope by which users can search and interpret language patterns across case studies. Qualitative data analysis software (QDAS) with semi-supervised AI coding functions, where the user defines the question and codes for analysis, makes this blended approach possible31. For the City CSA, this approach was applied through ATLAS.ti’s intentional AI coding function, which utilizes generative pre-trained transformer-based LLMs trained on large general bodies of text and human-centered parameters32.

Adapting Ozuem et al.’s33 qualitative coding procedures for thematic analysis with AI-assistance, several quality control measures were implemented to promote reliability and transparency for the coding process of the City CSA user type. First, all intentional AI-generated codes, code groups, and coded quotations were reviewed by the research team through an iterative process. This provided a human validation step by specifically adapting the iterative qualitative coding protocol that has been established and used in general qualitative code procedures research: (1) review individual codes and associated quotations for accuracy and contextual relevance; (2) assessment of preliminary intentional AI-generated code groups; and (3) evaluate if codes and quotations are consistent with case study document content.

Validation of the intentional AI-assisted coding follows established protocols for AI-assisted qualitative analysis of text34. Since text coding is a qualitative process, validation of the AI-assisted codes relied on systemic human review to assess the relevance of the context and themes, rather than quantitative metrics for accuracy. This was done through review of the text codes and removing ones not relevant to the sentence, phrase, or paragraph. The iterative human review process is a validation approach that aligns with qualitative research standards for text coding, where human judgment is the calibration standard35,36.

A question typology was developed for the intention prompt and was replicated across all scenario user groups (Table 2). The questions set the parameters for the intention prompt to interrogate the case studies in a systematic way. The process also enables the user to be at the center of the analysis process, so results are tailored to the context in which the user is working, while maintaining analytic rigor. Tier 1 questions produce manifest content where basic identification allows the user to understand frequency of interested terms. Tier 2 questions focus on relational content where the user is focused on the patterns of terms used across text. Tier 3 questions produce latent content where the user focuses on synthesis or summary of ideas across the text.

Table 2 Question typology approach
Full size table

Across the three user types tested in the Intentional AI Discourse Analysis, the same set of 20 case study documents was used to evaluate all equally. This enabled a comparison of how different analytical lenses illustrate different thematic emphasis across the documents. This can be seen in the most-coded documents across the user types (See Supplementary Tables 4–6). In addition, we conducted complete documentation of the input questions, AI-generated question refinements, code categories, quotation frequencies, and thematic hierarchies to provide transparency and replicability.

Based on discussions with the UCCRN Guidance Group, three sets of questions following the typology were formulated to create scenarios of different user profiles. The goal was to have questions about different hazards and concerns that could be determined systematically across the text of the case studies, and across geographic regions. The goal here is that a user could extract learnings or inquire about strategies instead of being limited to a region. From the questions inputted for the intention prompt, a set of questions and codes are generated for the user to review and revise based on what they are looking for. From here, the prompt produces a set of questions that further breaks down the original set of three questions asked, so each question is split into finer detail to inform code categories.

The questions and code category can be edited to user needs, where additional questions can be added or removed at user discretion. Once established, the user confirms these questions and categories, and ATLAS.ti uses these parameters to check the documents and text, paragraph by paragraph, and uses the questions to find the relevant codes. Once completed, major code groups and their sub-code categories are delineated throughout the documents which can be edited and deleted by the user.

Results

The results summarize patterns that emerged from using a metadata search filter approach to discover relevant case studies, as well as an LLM-assisted discourse analysis to analyze common themes within case studies in the City CSA prototype. Outcomes are presented by search approach, highlighting similarities and differences in the types of insights generated.

Direct user metadata search filter approach

To demonstrate the Direct User approach for accessing case study content in the City CSA, the user type “Flood Manager in the Global South” was applied. All flood-related case studies in the prototype from both the Global North and Global South were selected (Table 3).

Table 3 Flood-related case studies in UCCRN City CSA prototype
Full size table

Salient metadata that appeared in each of the case studies was identified to determine how a Global South flood manager could effectively navigate the City CSA to identify case studies relevant to their local context. Figure 4 presents the metadata search filters for Global North and Global South flood-related case studies and outlines a potential pathway for how a Global South flood manager could use these functions to identify the relevant cases.

Fig. 4
Fig. 4
Full size image

Decision tree illustrating the pathway a Flood Manager in the Global South user type might take when utilizing the metadata search filters, based on flood-related Global South and Global North case studies from the UCCRN City CSA prototype.

The Direct User metadata search applies filters drawn from predefined categories developed by UCCRN and selected by the case study authors. This approach is designed to help users efficiently identify relevant case studies to aid local implementation of specific urban climate actions, inform policy recommendations, support city networks in evaluating case studies that demonstrate best practices for effective climate action and potential interventions for funders evaluating specific interventions.

LLM-assisted discourse analysis with user profile approach

We experimented with the three user profiles to observe how the CSA content could be used for inductive coding by different sets of content interests. In each we assume that a Global South user could gain practical knowledge from Global North experiences and vice versa. In some situations, the size of the city and its location/geography are as equally important for users and knowledge transfer as its position in the global development economy.

In the examples given, a user from the Global South finds content from case studies in both the Global North and the Global South. The same applies for the case of a user from the Global North. A further study could examine if there is a significant variation in results if case studies analyzed are limited to the Global South or for the Global North. Across the three user simulations, six code groups emerged that identified the most commonly occurring themes based on quantification of text. Within these code groups, there were subcodes that categorized text based on presence of terms, phrases and their context (Table 4).

Table 4 Three user type scenarios using ATLAS.ti intentional AI coding function with tiered questions and coding group results for all case studies in the City CSA prototype (N = 20 case studies)
Full size table

Flood Manager in the Global South. The LLM-assisted discourse analysis showed that buyout programs emerged as the most frequently coded theme with 25 quotations across two documents, including subcodes for buyout committees, context, options, policies, and community dynamics. Buyout programs expectedly showed a heavy emphasis on the New York City Staten Island post-Hurricane Sandy case study.

Results related to community resilience contained 29 quotations across nine documents, covering adaptation strategies, collaborative decision-making, community engagement, and disaster response patterns.

Flood management included 39 quotations across nine documents, addressing coastal floods, drainage systems, stormwater management, and infrastructure approaches across cities including New York City, Shanghai, Naples, and Durban. Management patterns encompassed 37 quotations across 10 documents, identifying themes of community engagement, collaboration, adaptation practices, and stakeholder involvement. The analysis also highlighted contextual associations linking community action, climate change, and decision-making processes, in addition to terminology locations that mapped specific flooding and infrastructure concepts to geographic contexts.

Through the iterative examination of the buyout code group, some buyout codes had to be removed from the Durban, Kano, Port Vila, and Mexico City case studies as these quotes did not contain relevant information or text related to buyouts. Across all these user scenarios, the research team retained, modified, or removed quotes through a systematic review of 916 codes to make sure that the thematic interpretations made were contextually accurate within the scope of the case study instead of relying solely on automated pattern matching from the intentional AI coding process. As such, the approach did not lessen the time spent on qualitative analysis, but shifted the labor and concentration37.

Climate Resilience Practitioner in the Global North. Mitigation strategies represented the most identified cluster, with 91 quotations across 16 documents, containing subcodes that categorized infrastructure development, energy efficiency, renewable energy systems, green infrastructure, and community-based adaptation approaches. Transportation infrastructure patterns followed with 67 quotations across 16 documents, linking electric buses, green transportation, and infrastructure resilience to specific geographic contexts including Mexico City’s electrification efforts and Shanghai’s urban renewal initiatives. Sustainable energy management included 56 quotations across 15 documents, with subcode themes on cooling systems, temperature regulation, and HVAC efficiency.

Urban sustainability themes included 39 quotations, linking heat management to ecological and social resilience efforts encompassing, but not limited to, subcodes on urban green spaces, vertical greening, rain gardens, and nature-based solutions across the case study cities. Heat mentions provided context with 33 quotations across 10 documents, including subcodes on urban heat islands, rising temperatures, heatwaves, and heat mitigation approaches, with geographic concentration in Shanghai, Mexico City, and Naples. The equity association theme included 18 quotations across 6 documents, with subcodes capturing disadvantaged communities, environmental justice, gender equity, and vulnerable populations, predominantly present in LA County transportation planning.

Practitioner in East Asia Responsible for Energy Transitions and GHG Emissions Reduction. Themes from the text focused on energy transformation approaches and financing strategies across the case studies. Energy mentions included 120 quotations across 13 documents, containing subcodes that categorized renewable energy systems, energy efficiency measures, emissions reductions strategies, clean energy deployment and energy governance frameworks. Climate action patterns followed with 113 quotations across 15 documents, linking green energy initiatives, carbon emissions reduction, energy transition projects to specific geographic contexts including Rio De Janeiro’s renewable energy procurement and Shanghai’s low-carbon development strategies.

Infrastructure included 110 quotations across 16 documents, focused on climate investments to financing instruments, including but not limited to subcodes on green bonds, public-private partnerships, project financing, and municipal funding mechanisms. Coordination patterns provided context with 62 quotations across 16 documents, including subcodes on renewable energy coordination, project development, financing partnerships, and energy efficiency programs, with geographic concentration in Rio de Janeiro, Kano, and eThekwini Municipality in Durban. Policy association themes included minimal quotations across one document with subcodes capturing industrial land transformation, predominantly present in Shanghai’s urban renewal initiatives.

Discussion

The two search approaches serve different technical requirements, purposes, and scales (geographic and analytic) to accommodate different needs that occur when deriving urban climate action knowledge from case studies38 (Table 5). The Direct User metadata search utilizes filters based on pre-defined categories developed by UCCRN and selected by the case study authors to enable users to gather basic relevant identification of case studies of particular sets of urban climate actions. The intentional AI-assisted discourse analysis illustrates a research-focused methodology that explores the depth of cross-cutting patterns found across the case studies through systematic forms of text review. While the metadata search functionality of the City CSA helps users identify which case studies are most relevant to their needs, an LLM-assisted discourse analysis reveals what those cases actually say.

Table 5 Comparison of metadata search filters and discourse analysis search functions
Full size table

Based on the metadata analysis for a Flood Manager in the Global South user type, the metadata revealed differences between relevant themes or keywords presented in the LLM-assisted discourse analysis. For example, flood-management measures such as buyout programs were salient coding group results that appeared in Global North case studies like New York, but are largely absent from Global South contexts due to constraints such as governance, fiscal capacity, and informality. While Global South practitioners may draw from Global North examples, some interventions may not be directly transferable. A direct-user approach allows users to filter metadata by applying their own contexts, recognizing that some interventions may be discursively frequent but practically irrelevant. By comparison, an LLM-assisted discourse analysis lacks built-in awareness of social, economic, political contexts and can overgeneralize unless explicitly guided.

Discourse analysis is used to understand the main themes that emerge across the text of the entire body of case studies. In this approach, codes are applied to label text that summarize content, which when examined together, enables the user to delineate the meaning of a piece of textual data. At the same time, in working with large volumes of textual data, human-centered coding alone does not allow for efficient use of the City CSA datasets. In addition, the different user types or audiences for the case studies have a range of interests in terms of what they want to learn from the City CSA. Since the volume of urban climate solutions documents is rapidly growing, the potential for enhancing the customization of users provides an opportunity for more discourse analysis users to augment their coding with AI-assisted qualitative analysis with LLMs39.

At the same time, the use of AI with qualitative research presents concerns about transparency and validity40. In this way, we critically attended calls to accentuate the role and agency of humans for QDAS and wider coding methodologies in AI-assisted qualitative research41. For example, the use of an iterative human review process for this study resulted in modification of intentional AI-generated codes across all three user scenario groups.

Conclusion

The UCCRN City CSA is a critical step toward scaling up equity-centered, science-based urban climate action. By grounding its approach in regional representation, inclusive participation, and actionable knowledge, the Atlas aims to support cities and relevant actors worldwide in developing and implementing adaptation and mitigation strategies. A comparison of two search approaches has shown that they serve different technical requirements, purposes, and scales to accommodate diverse needs that arise when deriving urban climate action knowledge from case studies. While the metadata search functionality of the City CSA helps users identify which case studies are most relevant to their needs, an LLM-assisted discourse analysis reveals thematic information from the text itself.

As the UCCRN City CSA continues to develop, the goal is to flexibly support the integration of additional analytical tools, such as downloadable datasets, open API services, metadata tools, Web Map Services for dynamic geospatial visualization from regional and local sources, and crowdsourced geo-data libraries. These tools and services can provide more granular insights and support the monitoring of local initiatives. Through its phased development and continuous engagement with diverse stakeholders, the City CSA serves not only as a repository of climate solutions but also as a dynamic platform for the transformative learning and collaboration essential to climate-resilient urban futures.