Automate and scale shelter detection using artificial intelligence and high-resolution satellite imagery 

Overall objectives

This activity will test whether artificial intelligence (AI) can improve the availability, quality, and timeliness of geospatial data on settlements in displacement contexts. It will assess whether automated shelter mapping offers a scalable, cost-effective, and sustainable alternative to the manual mapping methods currently used by UNHCR and its partners. 

By identifying shelters and settlement boundaries from high-resolution satellite imagery, the AI model pipeline will enable the monitoring of settlements, the planning household surveys, and optimization of assistance delivery. The activity will also generate evidence on the accuracy, effectiveness, and practical value of AI-based shelter mapping to inform sustainable responses to displacement at scale in the future. 

Activity description

​The activity will use a cloud-based AI model, developed by Google Research and made available through Data Insights for Social & Humanitarian Action (DISHA). As the model has already been developed and trained, the work will focus on testing, validating, and applying it in operational displacement settings. 

A range of settlements will be selected to test the model in different geographical and environmental conditions. For each site, satellite imagery will be acquired and analyzed to produce maps of shelters and settlements. The outputs will undergo quality checks, and a sample will be validated using manually labelled data. 

The data generated will support operations and programs in UNHCR and through its partnerships with humanitarian, development, and government organizations. Secondary outputs from the models, including shelter centroids, will be made available through UNHCR’s GeoServices platform as global public goods.  

Engagement with partners

The activity will be led by UNHCR’s Geographic Information Systems (GIS) Support Unit in the Global Data Service in partnership with DISHA, Google Research, the United Nations Satellite Centre (UNOSAT), and the UNHCR Innovation Service. DISHA will provide and manage the AI platform, while Google Research will provide the shelter mapping model. UNOSAT will support the acquisition and processing of satellite imagery and help assure the quality of results. 

Background and Context

Approximately 8.7 million forcibly displaced people live in settlements worldwide. Nearly 1,000 of these settlements are managed by UNHCR and its partners. Effective planning, service delivery, and monitoring depends on up-to-date data about shelters, infrastructure, and settlement boundaries. Yet, producing and analyzing this data is time-consuming, resource intensive, and difficult to scale, making regular updates challenging. At a time when humanitarian funding is declining, faster and more affordable ways to produce and update these data are increasingly important. Artificial Intelligence could help extract information on shelters and settlement boundaries from high-resolution satellite imagery more efficiently and at greater scale. More timely spatial data could strengthen operational decisions, support survey implementation, and improve planning and service provision in displacement settings.