User- and Data-Centric GeoAI for Scalable Mapping of Deprived Urban Areas: The IDEAtlas Approach for SDG 11.1.1 Monitoring

Up-to-date spatial information on deprived urban areas (DUAs) is essential for evidence-based urban policy and for monitoring Sustainable Development Goal (SDG) indicator 11.1.1 on slums and informal settlements. Yet, many cities in low- and middle-income countries (LMICs) lack reliable, spatial data on the location, extent, and dynamics of settlements. The IDEAtlas project addresses this gap by combining Earth Observation (EO), artificial intelligence (AI) (here referred to as GeoAI), and a user-centred design framework to deliver scalable, transparent, and policy-relevant DUA maps. This is achieved through a data-centric AI approach utilizing multi-modal EO inputs, primarily Sentinel-2 multispectral imagery, supplemented by ancillary geospatial layers such as building density, topography, and proximity to infrastructure. These datasets are fused within a lightweight Multi-Branch Convolutional Neural Network (MB-CNN) architecture (~33,000 parameters) designed for efficient, city-scale processing of Sentinel data. The model produces two main outputs at 10 m resolution, which are resampled to 100 m to protect vulnerable groups: (1) binary maps of deprived urban area extent and (2) a continuous deprivation severity index building on the IDEAMAPS Domain of Deprivation Framework. Multi-temporal processing provides annual DUA maps (2019–2023), capturing settlement expansion, densification, eviction, and upgrading dynamics. Thus, IDEAtlas adopts a user- and data-centric GeoAI approach. Through Living Labs in eight pilot cities (Nairobi, Lagos, Mumbai, Jakarta, Salvador, Medellín, Mexico City, and Buenos Aires), and an ongoing expansion to a larger number of Latin American cities, local governments, national statistical offices, NGOs, and community groups co-design data needs, validated outputs, and contributed to the creation of reference data. The interactive web-based IDEAtls Portal allows users to inspect model predictions, digitise settlement boundaries, correct misclassifications, and provide contextual feedback. These user-generated annotations are reintegrated into the model, improving performance and trust. In several cities, stakeholder-driven refinement increased F1-scores by up to 13%, demonstrating the value of participatory data curation in complex urban environments.

The IDEAtlas outputs provide several policy-relevant indicators, such as total DUA area, population living in deprived areas, and temporal change metrics directly linked to SDG 11.1.1. By integrating scalable GeoAI methods with user-in-the-loop validation and open-source infrastructure, IDEAtlas demonstrates how user- and data-centric GeoAI can bridge urban data gaps. The approach strengthens local capacity, enhances transparency, supports inclusive and evidence-based urban policy, and outlines a pathway towards a transferable framework for global SDG 11.1.1. monitoring.