AI learns to identify buildings in under-mapped regions

A new mapping initiative has been launched by the UN and Google to provide high-resolution building data in Africa and other under-mapped regions from 2016 to 2023. This project aims to address the challenges posed by low-quality mapping in these areas, which has hindered understanding of settlement changes and living conditions. The AI model developed for this purpose can identify various building types from low-resolution satellite images taken every five days, allowing for real-time updates on even temporary structures like refugee camps. Despite its capabilities, the AI has limitations, such as its inability to detect improvised shelters or tents, and its reliance on clear weather conditions for accurate identification. This can affect the reliability of the mapping in certain regions, particularly during rainy or cloudy periods. Accurate mapping is crucial for various practical applications, including emergency response, healthcare service delivery, and resource allocation. In conflict-affected areas, where census data is scarce, understanding population sizes becomes increasingly important. For instance, Somalia has not conducted a census since 1979, making it difficult to ascertain the number of residents. The WorldPop research team at the University of Southampton is utilizing the new dataset to improve population estimates globally, which is essential for equitable resource distribution. The overarching goal of this initiative is to empower policymakers in the global South with the same mapping tools available in the global North, ultimately working towards reducing inequality in access to vital information and services.