KFTD: Koopman-Fourier Time-Differentiable Network for Continuous Ocean Spatiotemporal Forecasting

The Koopman Fourier Time-Differentiable (KFTD) Network has been introduced for continuous ocean spatiotemporal forecasting, addressing the challenges of complex dynamical systems and computational efficiency. This two-stage model employs Koopman linearization and Fourier analysis for continuous time interpolation, achieving a 4x speedup compared to traditional methods and a 5.6% average reduction in mean squared error (MSE) across four ocean datasets. The introduction of a Differentiable Physics-PDE (DPP) Loss enhances physical consistency, making it a significant advancement for practitioners focused on scalable and accurate forecasting in ocean dynamics.