Learning Semilinear Neural Operators: A Unified Recursive Framework for Prediction and Data Assimilation

A learning-based state-space approach for semilinear PDEs enables efficient prediction and data assimilation.