OceanCBM: New AI model makes ocean heatwave predictions interpretable

Extreme ocean phenomena like marine heatwaves are notoriously difficult to predict and diagnose—standard machine learning models often deliver accurate forecasts but remain black boxes, failing to reveal the underlying physical drivers. Enter OceanCBM, a novel concept bottleneck model introduced by researchers Sanah Suri, Kieran Ringel, and Maike Sonnewald on arXiv (paper 2605.12639). OceanCBM is the first such model designed specifically for spatiotemporal prediction and mechanistic interrogation of ocean dynamics. It uses mixed supervision to predict mixed layer heat content, a key precursor of marine heatwaves, by routing information through an intermediate layer of prescribed concepts derived from geophysical fluid dynamics, plus a 'free' concept that captures residual physical processes. This design imposes soft physical structure without over-constraining the model, explicitly balancing interpretability and performance.

Across ensemble initializations, OceanCBM consistently yields mechanistic representations, whereas prediction-only and prescription-only baselines learn highly variable latent structures despite similar predictive performance. The model achieves interpretable, physically grounded representations without sacrificing skill, effectively characterizing the interpretability-performance trade-off. For climate scientists and oceanographers, this means they can now trust AI forecasts for extreme ocean events by understanding the specific physical mechanisms driving predictions. The 17-page paper includes 9 figures and 4 tables, with code and data access links provided via arXiv. OceanCBM marks a significant step toward reliable, explainable AI in climate science, offering a template for embedding physical concepts into deep learning pipelines.