Extreme heat events intensified by climate change pose a severe threat to public health. Tackling these challenges calls for an advanced climate digital twin that merges physical systems with virtual environments, enabling bi-directional information exchange and supporting decision-making under extreme heat conditions. However, physical urban climate models are computationally demanding, which limits their rapid response capabilities in urgent scenarios. Conversely, data-driven AI models enable faster predictions and enhance digital twin functionality.
Here, we propose an innovative climate digital twin framework embedding the Spatiotemporal Vision Transformer (ST-ViT) model using a Texas campus as a testbed. Specifically, we utilized the physical microclimate model, high-resolution urban 3D model, and meteorological data to simulate the Universal Thermal Climate Index (UTCI) during a 2022 heatwave with a spatial resolution of 1 meter and hourly temporal resolution. These multimodal data, including the generated fine-scale UTCI maps, were integrated to develop the ST-ViT model, which enables rapid and accurate human heat stress forecasting. Building on this real-time capability, we further created a digital twin platform to empower stakeholders and the public with effective tools for responding to neighborhood-level heat exposure.
This hybrid digital twin framework facilitates the efficient mitigation of extreme heat's negative impacts and supports informed decision-making to enhance thermal comfort and build more climate-resilient cities.