VHR-GSIS: A Deep Learning Based Very High-Resolution Dataset for Informal Settlements in 37 Global South Cities (2010 & 2025) Using Harmonized Google Earth Imagery

This study develops a harmonized, very‑high‑resolution (VHR) informal settlement mapping product covering 37 major cities across Africa, Asia, and Latin America. Building on a detailed assessment of regional morphological typologies, we operationalize two core UN‑Habitat criteria—housing durability and building density—as universal principles for informal settlement mapping. Nine representative cities are selected to construct independent segmentation models, each trained on a newly compiled dataset totally 12,053 manually curated image–label pairs derived from 0.59‑m Google Earth imagery, forming one of the largest VHR informal settlement datasets to date.

The products are provided as vector (shapefile) products derived from post‑processed raster classifications. Each shapefile contains a categorical “status” field with three values: 0, 1, and 2. A value of 0 denotes areas where informal settlements extent remained unchanged between 2010 and 2025; a value of 1 represents areas that existed only in 2010 and had disappeared by 2025; and a value of 2 represents areas that newly appeared in 2025 relative to 2010. This structure enables users to directly quantify stable, lost, and newly emerged informal settlements patches across the two epochs.