Deep learning-based thermal mapping for enhanced urban heat management and cooling energy reduction in Arid Saudi Arabian environments

Urban energy consumption in arid regions presents substantial challenges for sustainable development, with cooling demands representing a major component of building energy loads. The urban heat island (UHI) phenomenon intensifies these pressures by raising surrounding air temperatures, which in turn amplifies cooling needs. Robust energy management approaches depend on precise representation of land surface temperature (LST) at spatial scales fine enough to distinguish individual buildings and infrastructure, although practical thermal mapping systems are constrained by fundamental trade-offs between spatial resolution and temporal frequency. Although deep learning (DL) methods have shown promise for enhancing the resolution of thermal imagery, prior research has largely focused on temperate regions, resulting in limited insight into thermal behavior in arid and semi-arid urban environments. This study addresses this limitation by adapting DeepLabV3 + architecture for LST super-resolution in Saudi Arabian cities. The model incorporates atrous spatial pyramid pooling to capture multi-scale thermal patterns and employs specialized feature-fusion modules designed to preserve thermal boundaries at building edges and vegetation interfaces. The training data consisted of spatially aligned Landsat 8 thermal images and high-resolution UAV-based thermal observations collected over representative urban forms in Riyadh and Medina, allowing assessment of the model's spatial transferability across contrasting climatic regimes. Model accuracy achieved RMSEs below 1.5 °C in Riyadh and Medina, validating building-scale thermal assessments. The results indicate that vegetation and high-albedo surfaces play a substantial role in lowering LSTs. The resulting building-scale thermal products facilitate energy audits, cooling-demand mitigation, targeted retrofit prioritization, and energy-efficient urban planning in arid regions.