An Operational Uncertainty-Aware Framework for Urban Heat Mapping in City-Scale Digital Twins

We present an operational data-fusion framework for high-resolution urban heat island analysis, producing hourly near-surface air temperature fields at 20 × 20 m² spatial resolution over complex urban environments. The framework is demonstrated over the Thessaloniki metropolitan area, a dense Mediterranean city characterized by warm summers and mild winters. It integrates heterogeneous data streams, including meteorological forecasts, land-use and land-cover indicators, building height and urban morphology descriptors, and low-cost temperature sensor measurements. Inputs are harmonized in space and time, while automated gap-filling procedures ensure spatial completeness under real-world data availability constraints.

Τhe framework employs an uncertainty-aware ensemble combining Universal Kriging and Gaussian Process regression models. Both methods are executed in parallel to generate temperature estimates along with their predictive uncertainty fields. These uncertainty maps are explicitly used to construct a weighted ensemble, where model contributions are modulated according to local and temporal uncertainty, allowing the final temperature field to reflect the predictor expected to be more reliable at each location and time. This approach preserves fine-scale thermal gradients associated with land use and built structure while maintaining spatial coherence.

Results from the operational deployment, running continuously over several months, demonstrate the system’s ability to deliver stable and spatially consistent high-resolution temperature fields in a fully automated manner (Figure 1). The produced maps capture persistent intra-urban thermal contrasts linked to urban morphology and land-use patterns. Importantly, these results are derived from a live operational pipeline rather than post-processed reanalysis, highlighting robustness under real-time data availability and sensor sparsity constraints. Although the current evaluation period corresponds primarily to winter conditions, the outputs already provide valuable insight into urban thermal variability and establish a reliable baseline for forthcoming warm-season heat assessments.

.

Figure 1. Example hourly realization of the operationally produced near-surface air temperature field at 20 × 20 m² resolution over the Greater Thessaloniki Area.

The resulting temperature fields are designed to act as a core thermal layer within a city-scale digital twin, supporting integrated analysis of urban heat patterns, micro-climatic variability, and heat exposure, and providing a scalable backbone for urban heat studies and climate-adaptation planning.