Apparatuses, methods, and systems for generating simulations of physical variables of a physical system are disclosed. A method includes fusing, by the one or more computing devices, the observation data and the numeric simulation data, including training a spatial-temporal emulator model for a physical numerical model using the observation data, the numeric simulation data, and the domain interpretable data. The method further includes estimating one or more physical parameters based on the trained spatial-temporal emulator model, compressing the trained spatial-temporal emulator model, including generating candidate mutations of an architecture of the trained spatial-temporal emulator model, evaluating a performance of each of the candidate mutations of the architecture on validation data using metrics, retaining a subset of candidate mutations exhibiting best performance on the metrics, iterating until convergence to a desired reduction in size, and deploying the compressed trained spatial-temporal emulator model to one or more edge devices.