Traditional 3D geographic scene reconstruction methods rely on single data sources for geometric restoration, struggling to effectively integrate non-geometric information like text and semantics. This severely hinders reconstruction feasibility in scenarios involving sparse data or non-geometric descriptions. Advances in generative artificial intelligence (AI) for multi-modal data understanding and generation offer new solutions. However, applying generative AI to this task still faces two core challenges: the conflict between model output randomness and reconstruction accuracy, and between model generality and the generation of diverse, professional content within geographic scenes. To address these challenges, this study proposes a framework guided by geographic entity semantics and powered by multi-modal generative models. It first integrates multi-modal inputs for semantic enhancement, then generates 3D entities under geometric and domain-enhancement constraints, and finally completes scene integration using spatial pose constraints. A case study demonstrates the method’s versatility across various data combinations. It not only reconstructs scenes from text-only inputs, addressing the feasibility challenge, but also generates high-precision professional entities via rapid domain enhancement. Furthermore, adding geometric constraints yields significant improvements in reconstruction accuracy and visual realism, with user evaluations confirming its comprehensive superiority over traditional parametric methods.

Augmented reality (AR) scene modeling with virtual-real integration is an effective way to enhance users’ perception and understanding of geographic spaces. However, the existing modeling methods focus on precise virtual-real alignment in static scenes using single-frame images, leading to inefficiencies in dynamic scene modeling and low accuracy in virtual-real integration. This paper proposes a fast modeling method for AR dynamic scenes with spatio-temporal semantic constraints. By thoroughly analyzing spatio-temporal semantic constraint rules in AR dynamic scene modeling, a keyframe extraction algorithm based on a synchronized spatio-temporal semantic distance measurement model was designed. A rapid spatio-temporal interpolation model for AR dynamic view poses with spatio-temporal semantic association was established, and a real 3D scene-driven fast twin modeling method for AR dynamic scenes was proposed. Experimental results show that the proposed method reduces redundant image matching computations by 87.53% while maintaining virtual-real registration accuracy above 1°. This method enables accurate sampling of keyframes with spatio-temporal homogeneity, avoids redundant transmission of large volumes of frame image data, and improves AR dynamic scene virtual-real registration efficiency while maintaining accuracy. Furthermore, the spatial semantic information in real 3D scenes effectively guides fast AR dynamic scene modeling.

An important step in implementing geospatial digital twins is to enhance the expressiveness of virtual geographical scenes for the physical world. However, the existing virtual geographical scenes cannot quickly express the dynamically changing geographic environment for remote users due to the inefficient handling of modeling processes, user perception, and remote sharing. The research analysed the concept and characteristics of geospatial digital twins, and constructed the virtual geographical scene ontology, based on which we developed geographical spatiotemporal semantic rules and designed a dynamic annotation algorithm to enhance the representation of virtual geographical scenes. Finally, we investigated a real-time transmission method of panoramic video based on 5 G and used immersive virtual reality (IVR) to realize the user experience of remote immersion in geographical scenes. We selected a specific geographic environment containing multiple typical geographic entities to develop three prototype systems for experimental analyses. The results showed that the proposed method enabled users to view the virtual geographical scene on a VR device. The average latency for this process was 14.72 seconds. Compared with the virtual geographical scenes constructed by traditional methods, the experiments showed the proposed method advantageous in comprehensiveness, timeliness, and photorealism and abilities to enhance the user’s geographical scene perception.

Public’s rational flood mitigation behaviors depend on accurate perception of flood risks. The use of natural language for flood risk perception is an effective approach, and it is critical to ensure the accuracy and comprehensibility of the flood information provided by the system in natural language dialogues. This study presents a framework for large language model (LLM) that is constrained by flood knowledge and can interact with geographic information system (GIS), aimed at enhancing the public’s perception of flood risks. We tested the performance of LLM within this framework and the results demonstrate that LLM can generate accurate information about floods under the constraints of entities and relationships in the knowledge graph, and interact with GIS to produce personalized knowledge through real-time coding. Furthermore, we conducted flood risk perception experiments on users with different cognitive levels. The results indicate that using natural language dialogue can narrow the differences brought about by cognitive levels, allowing the public to equally access knowledge related to flood events.

The detail and representation of a spatial layout varies with scale. This affects an individual’s learning effectiveness and understanding, in turn directly influencing their behavior in a fire evacuation. However, the impact of layout learning methods with different spatial scales on fire evacuation behavior, and the relationship between spatial cognition and evacuation effects, remains unclear. We conducted spatial layout learning across three scales with 81 participants and simulated a fire evacuation scenario in a mobile virtual reality for groups. We collected evacuation decision-making and user experience questionnaires as supplementary data. The results demonstrate that small-scale learning objects are the easiest for participants to understand in terms of spatial layout and relationships, but their performance in fire evacuation is poor. Large-scale learning objects significantly improve participants’ evacuation efficiency. Spatial layout learning plays a crucial role in fire evacuation outcomes, but traditional spatial knowledge acquisition measurement methods cannot predict fire evacuation performance. This study sheds light on how spatial cognition influences fire evacuation behavior and provides a more reliable fire evacuation simulation method based on mobile virtual reality (MVR).