This theme integrates machine learning, reservoir computing, physics-guided modeling and critical-phenomena analysis to study the structure, dynamics and predictability of complex systems.
Artificial intelligence provides new tools for representation, inference and prediction in complex-systems research. Complex systems often involve high-dimensional variables, nonlinear coupling, multi-scale evolution and noisy observations, making them difficult to capture fully with traditional analytical models. Machine learning can learn effective representations from observations and physical priors, supporting state identification, forecasting and mechanism discovery.
This theme focuses on the two-way integration of AI and complex-systems science. On one side, methods such as reservoir computing, deep echo state networks and graph learning are applied to climate modes, ENSO prediction and complex dynamical systems. On the other side, machine-learning models themselves are examined through critical phenomena, universality, stability and interpretability. The related work emphasizes synergy between physical constraints and data-driven learning, so intelligent models serve not only prediction accuracy but also mechanistic understanding.
Learning the coupled dynamics of global climate modes
This work uses machine learning to characterize coupled dynamics among global climate modes, providing a data-driven framework for understanding and predicting complex climate systems.
Enhancing the Predictability Limits of ENSO with Physics-Guided Deep Echo State Networks
An ENSO real-time forecasting method based on ORAS5 reanalysis data and physics-guided deep echo state networks (DESN), combining climate-mode index construction, seasonal-cycle and trend removal, ensemble training, member selection, and Niño3.4 forecast results for 2026-2027.
