The integration of advanced monitoring and interaction systems in industrial and urban environments is critical for ensuring safety and facilitating rapid responses to potential hazards, such as accidental gas releases at chemical plants. For this purpose, digital twins (DTs) have emerged as a powerful tool, supporting reactive decision-making and response in real-world scenarios. However, the accuracy of predictions within a DT heavily depends on the chosen approach for modeling the system dynamics. For the problem at hand, a common approach leverages Gaussian plume models, providing fast yet rough estimates for different wind conditions. In contrast, computational fluid dynamics based simulations might lead to more accurate predictions by providing valuable information related to the complex flow patterns that arise in built environments. However, their computational demands, particularly for complex geometries and large domains with many degrees of freedom, can be prohibitive, making them difficult to apply in time-critical scenarios or multi-query analyses. To address this challenge, Model Order Reduction (MOR) techniques offer a viable solution approach by reducing computational demands without significantly compromising the accuracy of predictions. This enables the use of a reduced-order model instead of the original high-fidelity, yet computationally expensive, full-order model. Consequently, this presentation will focus on the benefits and limitations of various MOR techniques applied to simulations of gas dispersion scenarios under different wind conditions. Additionally, the potential of multi-fidelity approaches will be evaluated to enhance the predictive capability of the developed models by incorporating high-fidelity data. Finally, practical pathways to integrate the resulting models into a DT framework will be discussed. To this end, the presented approach uses centralized data storage to make simulation results and metadata accessible to various services, such as interactive visualization tools, in order to facilitate improved decision-making capabilities.