Anomaly-Informed Model Simplification: Using Unexpected Behaviors to Guide Climate Model Reduction

While Proske et al. (2023) discuss the complex balance between model fidelity and interpretability—sometimes simplifying cloud microphysics where process sensitivities are weak—this idea flips the script: What if we systematically use unexpected model behaviors (i.e., anomalies) as diagnostic cues for where to simplify or, conversely, where more detail is needed? Building on anomaly detection frameworks like the scVARMA model (Ma et al., 2024) and autoencoder-based methods (Chesnokov et al., 2020), this research would develop pipelines that monitor model outputs for statistically significant deviations from both observations and physical expectations. Regions or processes with frequent unexplained anomalies could then be flagged for either simplification (if they're shown to have little impact on key outputs) or for targeted refinement. This approach would create a feedback loop between anomaly detection and model structure, leading to more interpretable, efficient, and physically grounded climate models.

References:

1. Sparsity-Constrained Vector Autoregressive Moving Average Models for Anomaly Detection of Complex Systems with Multisensory Signals. Meng Ma, Zhongyi Zhang, Zhi Zhai, Zhirong Zhong (2024). Mathematics.
2. Addressing Complexity in Global Aerosol Climate Model Cloud Microphysics. Ulrike Proske, S. Ferrachat, Sina Klampt, Melina Abeling, U. Lohmann (2023). Journal of Advances in Modeling Earth Systems.
3. Detection of Structural Behavior Anomalies in Hybrid Roof Systems. A. Chesnokov, V. Mikhailov, I. Dolmatov (2020). Summa.