Hybrid NARX-LLM for Greenland Iceberg Discharge: Prompt-Driven Residual Correction

The article presents a Hybrid NARX-LLM framework that integrates a nonlinear autoregressive model with exogenous inputs (NARX) and a large language model (LLM) for enhancing predictions of Greenland iceberg discharge. It introduces a Physics-Informed Prompt (PIP) method, which converts unstructured physical knowledge into structured prompts for zero-shot reasoning, allowing the model to correct systematic prediction errors by incorporating intrinsic temporal dependencies and environmental factors. This approach addresses the challenges of modeling extreme and nonstationary events in climate forecasting, providing a scalable solution that merges time-series analysis with physics-informed AI methodologies.