User and data-centric artificial intelligence for mapping and benchmarking urban deprivation for a global sample of cities
Rapid urbanization across many regions worldwide has significantly contributed to the growth of deprived urban areas (DUAs), often called slums or informal settlements. The lack of reliable geospatial information on their location and extent in many cities continues to hinder efforts aimed at improv...
| Autores: | , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Tipo de recurso: | artículo |
| Estado: | Versión publicada |
| Fecha de publicación: | 2026 |
| País: | España |
| Institución: | Universidad Pública de Navarra |
| Repositorio: | Academica-e. Repositorio Institucional de la Universidad Pública de Navarra |
| OAI Identifier: | oai:dnet:academicae__::f662c7fba9044754587afd93c644300a |
| Acceso en línea: | https://hdl.handle.net/2454/56891 |
| Access Level: | acceso abierto |
| Palabra clave: | Urban deprivation Informal settlements Slums Earth observation Deep learning Data-centric AI |
| Sumario: | Rapid urbanization across many regions worldwide has significantly contributed to the growth of deprived urban areas (DUAs), often called slums or informal settlements. The lack of reliable geospatial information on their location and extent in many cities continues to hinder efforts aimed at improving living conditions. This study addresses this critical information gap by exploring a User- and Data-centric Artificial Intelligence (AI) approach for accurately mapping these areas to support Sustainable Development Goal (SDG) Indicator 11.1.1. In collaboration with local communities, governments, and international stakeholders, we co-designed an AI-driven strategy leveraging open Earth Observation (EO) and geospatial data acrosseight cities worldwide. Instead of relying solely on algorithmic precision, our method prioritizes local knowledge, iterative validation, and adaptive data collection. To achieve this, we developed a tailored multi-branch encoder-decoder convolutional neural network capable of integrating multi-modal data sources. Our approach incorporates an agile and iterativemodel refinement process, ensuring continuous feedback loops between AI design, data collection, and validation. Recognizing the importance of stakeholder engagement, we developed the IDEAtlas collaborative data collection platform - https://portal.ideatlas.eu/ - to enhance data quality and inclusivity. The resulting dataset (IDEABench) is publicly available at https://doi.org/10.17026/PT/X4NJII to facilitate continued research and development. Findings indicate that fusing multi-spectral EO data with urban morphometric features, particularly Sentinel-2 imagery and built-up density, provides the highest accuracy for identifying DUAs. Furthermore, improvements in reference data quality through the IDEAtlas platform led to increased mapping precision. However, the significant variability in accuracy across cities underscores the complexity of the task and suggests the need for supplementary geospatial data to complement EO-driven analysis. The code used in this study is available at https://github.com/IDEAtlas/ai-dua-mapping. |
|---|