KAN-GCN: Combining Kolmogorov–Arnold Network with Graph Convolution Network for an Accurate Ice Sheet Emulator

We introduce KAN–GCN, a fast, accurate emulator for ice-sheet modeling that places a Kolmogorov–Arnold Network (KAN) as a feature-wise calibrator before graph convolution networks (GCNs). The KAN front end applies learnable 1-D warps and a linear mix, improving feature conditioning and nonlinear encoding without increasing message-passing depth. We employ this architecture to improve the performance of emulators for numerical ice sheet models. Our emulator is trained and tested using 36 melting-rate simulations with 3 mesh-size settings in the Pine Island Glacier, Antarctica. Across 2–5 layer architectures, KAN–GCN reproduces or outperforms the accuracy of pure GCN and MLP–GCN baselines; it is consistently better for 3–5 layers (with larger gains on velocity than on the easier thickness target) and slightly worse at 2 layers due to reduced aggregation depth. Despite a small parameter overhead, KAN–GCN improves inference throughput on coarser meshes by replacing one edge-wise message-passing layer with a node-wise transform; only the finest mesh shows a modest cost. Overall, KAN-first designs offer a favorable accuracy–efficiency trade-off for large transient scenario sweeps.