The design of long-span bridges is mainly driven by their wind-induced responses. Traditional design approaches involve conducting wind tunnel tests of a reduced set of deck cross-section configurations to extract their fluid-structure interaction (FSI) parameters, thereby limiting domain exploration due to budget constraints. Recent advances in CFD simulations have enabled an increase in the scope of design exploration, but they are still limited by computational resources. In contexts of limited data, effective and accurate interpolation becomes critical. This investigation explores the development of robust workflows for accurately interpolating FSI parameters across varying bridge deck shapes and reduced velocities. While traditional interpolation methods typically work in the frequency and design domains, previous works by the authors [1] demonstrated that using deeper information, such as force time series in the time domain before their integration in the frequency domain, improves prediction accuracy. Building on this direction, this investigation proposes emulating in the space-time domain by training a deep learning (DL) model with flow fields of bridge decks under forced vibration. The proposed workflow utilizes a custom Autoencoder (AE) architecture tailored for flow field image data. The model leverages physical information (including reduced velocities and bluff body shape geometry) and physics-informed loss functions (enforcing mass and moment conservation) to physically ground its predictions. The architecture integrates state-of-the-art techniques, including skip connections, pixel shuffle, and Feature-wise Linear Modulation (FiLM) [2], to maximize fidelity. We utilize SDF-biased sampling [3] to prioritize critical flow features near the bridge deck surface (the boundary layer) when constructing flow field imagery. This ensures that critical surface features are preserved rather than smeared, allowing for the precise extraction of non-dimensional self-excited forces. Preliminary results indicate that this method can surpass the accuracy of existing image-based flow field prediction techniques, thereby improving the prediction of FSI parameters.