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

2026-06-13 · Source: Machine Learning · Field: Science & Research — Environmental Science & Earth Systems, Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

A Hybrid NARX-LLM framework has been developed to predict Greenland iceberg discharge, addressing the complex nonlinear dynamics and limited observability that challenge traditional models. This framework integrates a Nonlinear Autoregressive model with eXogenous inputs (NARX) with a large language model (LLM) for residual correction. A key innovation is the Physics-Informed Prompt (PIP) method, which converts unstructured physical knowledge into structured prompts for the LLM's zero-shot in-context reasoning. The NARX component handles intrinsic temporal dependencies, while the LLM, guided by PIP, encodes glacier dynamics and environmental drivers to correct systematic prediction errors and reason about unmodeled factors. This approach aims to enhance overall predictive accuracy, particularly for extreme events, by fusing structured time-series modeling with knowledge-driven foundation AI. The code for this framework is available, published on 2026-06-13.

Key takeaway

For research scientists modeling complex environmental systems with limited data, you should consider integrating hybrid NARX-LLM frameworks. This approach, particularly with Physics-Informed Prompts, offers a scalable and interpretable method to correct systematic prediction errors and reason about unmodeled factors, especially for extreme events. Explore the provided code to adapt this fusion of structured time-series modeling with knowledge-driven foundation AI for your specific climate forecasting challenges.

Key insights

A hybrid NARX-LLM framework uses physics-informed prompts to correct iceberg discharge predictions, enhancing accuracy for complex climate dynamics.

Principles

• Hybrid models improve complex system predictions.
• LLMs can correct systematic prediction errors.
• Physics-informed prompts enable zero-shot reasoning.

Method

The framework combines NARX for temporal dependencies with an LLM for residual correction. A Physics-Informed Prompt (PIP) method transforms physical knowledge into structured prompts for the LLM's zero-shot reasoning.

In practice

• Apply LLMs for residual correction.
• Use PIP for physics-informed reasoning.
• Integrate foundation AI for climate forecasting.

Topics

• Hybrid Models
• NARX
• Large Language Models
• Iceberg Discharge
• Climate Forecasting
• Physics-Informed Prompts
• Time Series Analysis

Best for: AI Scientist, Research Scientist

Related on AIssential

• The next chapter in flood resilience: Open sourcing Google’s hydrology framework — Open-sourcing Google's AI hydrology framework democratizes advanced flood forecasting, enabling local adaptation and global resilience.
• EpiEvolve: Self-Evolving Agents for Streaming Pandemic Forecasting under Regime Shifts — LLM forecasters can adapt to streaming data and regime shifts via memory-based self-evolution, not parameter updates.
• Physics-Guided Dual Decoding and Spectral Supervision for Global 3D Hydrometeor Prediction — PredHydro-Net uses physics-guided dual decoding and spectral supervision to improve global 3D hydrometeor prediction, overcoming smoothing issues.
• Enhancing AI and Dynamical Subseasonal Forecasts with Probabilistic Bias Correction — Probabilistic Bias Correction (PBC) significantly improves subseasonal weather forecast accuracy and extreme event prediction.
• EVIL: Evolving Interpretable Algorithms for Zero-Shot Inference on Event Sequences and Time Series with LLMs — LLM-guided evolution can discover compact, interpretable algorithms for zero-shot dynamical systems inference, outperforming complex neural networks.
• Comparing LLM and Fine-Tuned Model Performance on NVDRS Circumstance Extraction with Varying Prompt Complexity — LLMs, guided by a context-aware "Complexity Score" for prompt selection, significantly outperform fine-tuned models on inferentially complex, low-prevalence data extraction.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.