In order to precisely capture the spatiotemporal coupling characteristics of ground operation processes for turnaround flights at hub airports under high-density operations, a dynamic coupling prediction methodology combining multi-dimensional spatiotemporal features with a bidirectional long short-term memory (Bi-LSTM) network is proposed. The turnaround process is first deconstructed into 18 critical milestone nodes, where the spatial distance matrix of aircraft stands and multi-flight time overlap factors are extracted to quantitatively model resource competition and evolutionary dynamics under multi-process collaborative decision-making. On this basis, a multi-process spatiotemporal feature correlation framework is developed, incorporating a graph attention mechanism and a dynamic error feedback weight allocation scheme to enable the real-time capture of actual node states and rolling updates of model weights. Validation using records of 5,767 turnaround flights at Beijing Daxing International Airport, obtained through VariFlight, demonstrates that the model achieves an overall mean absolute error of 1.05 min and a root mean square error of 1.34 min across all 18 nodes. Under a 10% Gaussian noise perturbation, the mean absolute error escalates by merely 0.27 min. Ablation studies further confirm the dominant role of temporal and spatial correlation factors in residual compensation and accuracy enhancement. By jointly considering within-flight process evolution and interactions with neighboring flights, the proposed method can provide rolling prediction support for turnaround progress monitoring, coordination of ground handling tasks, and collaborative departure operations.