The Indonesian Throughflow (ITF) serves as a critical conduit linking the Pacific and Indian Oceans, with its interannual variability extensively modulated by the El Niño–Southern Oscillation (ENSO) and the Indian Ocean Dipole (IOD). Due to the complex basin topography and scarcity of in-situ observations in the Indonesian seas, traditional methods face difficulties in directly revealing the specific propagation pathways of interannual signals into the interior Indonesian seas.
To address this issue, this study first utilizes small-scale and remote lag correlation analyses to identify network nodes. We depict the propagation pathways of signals from different source regions (i.e., the Pacific or Indian Ocean) to the Indonesian seas. Grid points exhibiting significant correlations with the source signals are identified as key nodes for the graph network, and an adjacency matrix is constructed based on spatial topological relationships. Subsequently, we apply the Deep Learning Model known as Spatial- Temporal Graph Convolutional Network (STGCN). To date, the results demonstrate that the STGCN model can identify high-weight connections between nodes, explicitly revealing the propagation pathways of interannual signals through key passages in the Indonesian seas, as well as variations associated with the climate modes. This investigation provides a novel data-driven paradigm for exploring oceanic wave propagation mechanisms in regions characterized by complex topography and sparse observational coverage.