Robust Spatiotemporal Forecasting Using Adaptive Deep-Unfolded Variational Mode Decomposition

--Accurate spatiotemporal forecasting is critical for numerous complex systems but remains challenging due to complex volatility patterns and spectral entanglement in conventional graph neural networks (GNNs). While decomposition-integrated approaches like variational mode graph convolutional network (VMGCN) improve accuracy through signal decomposition, they suffer from computational inefficiency and manual hyperpa-rameter tuning. T o address these limitations, we propose the mode adaptive graph network (MAGN) that transforms iterative variational mode decomposition (VMD) into a trainable neural module. Evaluated on the LargeST benchmark (6,902 sensors, 241M observations), MAGN achieves an 85-95% reduction in the prediction error over VMGCN and outperforms state-of-the-art baselines. Accurate spatiotemporal forecasting is a foundational task for understanding and managing complex systems characterized by interconnected entities, such as transportation networks, environmental monitoring grids, and financial markets. A prime example is accurate spatiotemporal traffic forecasting, which is fundamental to intelligent transportation systems for enabling route optimization [1], congestion mitigation [2] and emission reduction [3].