Mapping informal settlements with artificial intelligence: challenges of morphological ambiguity in Monterrey Metropolitan Area

Mapping informal settlements remains a major challenge for urban planning due to limited reliable spatial data and heterogeneous morphologies of informal urbanization. Advances in remote sensing and artificial intelligence have expanded possibilities for large-scale detection; however, their effectiveness depends on clear morphological distinctions between formal and informal development. This study evaluates the transferability of a U-Net-based convolutional neural network model, originally trained in the Sula Valley (Honduras), to the Monterrey Metropolitan Area (Mexico), and assesses how morphological ambiguity between informal settlements and state-promoted serviced-lot developments complicates automated classification. A U-Net convolutional neural network was applied to Sentinel-2 satellite imagery and calibrated with institutional datasets, documentary sources, and ground-truth verification. Results show the model effectively identifies patterns of peripheral and topographically constrained expansion but struggles where informal settlements resemble state-promoted serviced-lot developments. Findings highlight the need for hybrid mapping approaches that integrate automated detection with contextual territorial knowledge and institutional data.