Loading paper data...
Corpus Overview
Urban climate research papers with LLM-extracted structured metadata
Thematic Landscape
How Much We Collapse
Creating a stable middle layer for figures and a small top layer for the manuscript narrative.
How Themes Evolved Over Time
Publication trends across super-themes and thematic breadth.
Theme Browser
Click a theme to explore its sub-themes. Tile sizes show paper-to-theme assignments; the table below shows de-duplicated papers.
About the Atlas
Background
Since 2003 the International Association for Urban Climate (IAUC) Bibliography Committee has catalogued urban climate publications worldwide. This atlas distils that two-decade effort into a searchable database of 9,284 research papers—spanning heat islands, air quality, building energy, and climate adaptation—across 121 countries.
Methodology
How we built the corpus
Literature identification. The IAUC Bibliography Committee has catalogued urban climate publications since 2003. For this atlas, we queried the Web of Science Core Collection across 111 journals, plus two journals screened manually (Nature Cities and Cities and the Environment). An initial broad query yielded approximately 127,000 records; an iterative Delphi process refined the search terms and inclusion criteria. The final query—(urban* OR city OR cities) AND (atmospher* OR *climat* OR *meteorolog*)—returned 36,483 records, screened by experts on title and abstract, yielding 8,502 papers.
LLM-based extraction. Each paper was processed by Google Gemini 2.5 Pro with native PDF support. The model extracted 51 structured metadata fields—bibliographic information, study location and period, methodology flags, and content summaries. Every extracted value is evidence-grounded with page numbers and verbatim quotes from the source PDF.
Expert validation. Forty domain experts each reviewed 25 papers in a blind protocol—no LLM predictions were shown. The stratified sample of 999 papers (12% of the corpus) confirmed 86% agreement on inclusion decisions (Cohen’s κ = 0.43), 96–98% accuracy on content summaries, and 91–95% on methodology flags.
Enrichment analysis. Beyond the base extraction, each paper was further classified along several dimensions: geographic and climate-zone typology, contribution type (empirical, fundamental, applied, or synthesis), and alignment with IPCC AR6 and SR Cities chapter themes. These enrichment layers enable the thematic, geographic, and policy-relevant views in the atlas.
From evidence to atlas. The validated and enriched metadata feeds this searchable interface. Browse by geography, methodology, or research theme; filter by year or inclusion confidence; and explore individual papers with their extracted evidence.
Team
How to Cite
Data & Code
The extraction protocol, structured metadata for all 9,284 papers, anonymised expert validation data, and the full analysis pipeline are archived on Zenodo:
Contact
For questions, data access requests, or to report issues, please contact the corresponding authors: chenghao.wang@ou.edu or ting.sun@ucl.ac.uk.