KANs Layer Integration: Benchmarking Deep Learning Architectures for Tornado Prediction

Tornado occurrence and detection are well established in mesoscale meteorology, yet the application of deep learning (DL) to radar-based tornado detection remains nascent and under-validated. This study benchmarks DL approaches on TorNet, a curated dataset of full-resolution, polarimetric Weather Surveillance Radar-1988 Doppler (WSR-88D) radar volumes. We evaluate three canonical architectures (e.g., CNN, VGG19, and Xception) under five optimizers and assess the effect of replacing conventional MLP heads with Kolmogorov–Arnold Network (KAN) layers. To address severe class imbalance and label noise, we implement radar-aware preprocessing and augmentation, temporal splits, and recall-sensitive training. Models are compared using accuracy, precision, recall, and ROC-AUC. Results show that KAN-augmented variants generally converge faster and deliver higher rare-event sensitivity and discriminative power than their baselines, with Adam and RMSprop providing the most stable training and Lion showing architecture-dependent gains. We contribute (i) a reproducible baseline suite for TorNet, (ii) evidence on the conditions under which KAN integration improves tornado detection, and (iii) practical guidance on optimizer–architecture choices for rare-event forecasting with weather radar.