AI-based multi-temporal analysis of urban dynamics using Sentinel-2 data. A case study over Osmaniye, Turkey
TL;DR
This study presents a multi-temporal framework for urban monitoring that integrates Sentinel-2 multispectral imagery with semantic segmentation techniques, and highlights the potential of Artificial Intelligence and Remote Sensing for continuous urban monitoring, disaster impact assessment, and data-driven urban planning.
Abstract
Abstract. Monitoring urban dynamics in hazard-prone regions is essential for understanding long-term urban growth and assessing the impact of disruptive events. This study presents a multi-temporal framework for urban monitoring that integrates Sentinel-2 multispectral imagery with semantic segmentation techniques. A U-Net architecture trained using World Settlement Footprint (WSF) reference data was employed to automatically extract built-up areas and reconstruct the temporal evolution of urban expansion in the city of Osmaniye (Türkiye) between 2015 and 2025. The trained model was applied to the full Sentinel-2 time series to generate yearly built-up maps and analyse changes in the urban footprint over the decade. The results reveal a clear and sustained expansion of built-up areas throughout the study period. The multi-temporal analysis also captures a distinct disruption in the urban trajectory associated with the 2023 earthquake, followed by a rapid rebound likely related to post-disaster reconstruction activities. By combining semantic segmentation with multi-temporal satellite observations, the proposed framework enables the detection of both gradual urban expansion and abrupt disaster-related changes using openly available Earth Observation data. The results highlight the potential of Artificial Intelligence and Remote Sensing for continuous urban monitoring, disaster impact assessment, and data-driven urban planning.