A KAN-SAE with a nonlinear encoder but linear decoder directions discovers climate features in a real AI weather model and causally steers physically coherent forecasts

Cheon (2026) trains a KAN-SAE (linear decoder, nonlinear per-feature B-spline encoder gating) on the layer-5 residual stream of Sonny, a real pretrained hierarchical weather transformer trained on ERA5 reanalysis, finding markedly more alive, less redundant features than a matched linear-encoder SAE baseline (975/1024 alive vs. 566/1024; median inter-feature correlation 0.076 vs. 0.092) and much tighter feature localization for a heatwave feature (2 degrees vs. 51 degrees error) [cheon-2026-beyond-linear-superposition-discovering-climate-features-kan-sae] Steering along one feature's decoder direction causally produces a dose-dependent, physically coherent regional temperature anomaly (EU-mean T2m +1.42K at steering strength 2, r^2>0.99) with coupled pressure-field changes consistent with real blocking-anticyclone dynamics -- the decoded features remain genuine linear directions in the model's own activation space [cheon-2026-beyond-linear-superposition-discovering-climate-features-kan-sae]