ClimateSOM: A Visual Analysis Workflow for Climate Ensemble Datasets.

Area of Science:

• Climate Science
• Data Visualization
• Artificial Intelligence

Background:

• Ensemble datasets are crucial for climate projections, capturing variability under different emission scenarios.
• Analyzing the magnitude and patterns of variability among ensemble model runs is vital for climate scientists.
• Existing methods may not fully capture the complex spatial-temporal dynamics within ensemble climate data.

Purpose of the Study:

• To introduce ClimateSOM, a novel visual analysis workflow for exploring and interpreting climate ensemble datasets.
• To leverage self-organizing maps (SOMs) and Large Language Models (LLMs) for enhanced sensemaking of climate model variability.
• To enable interactive exploration, pattern identification, comparison, and clustering of ensemble model runs.

Main Methods:

• Utilizes a self-organizing map (SOM) to abstract spatiotemporal time series into a 2D distribution, capturing ensemble variability.
• Integrates Large Language Models (LLMs) to aid in the interpretation of the SOM-defined 2D space.
• Applies the workflow to precipitation projection ensemble data for California and the Northwestern United States.

Main Results:

• ClimateSOM effectively visualizes and abstracts climate ensemble variability into an interpretable 2D space.
• LLM integration assists users in making sense of the SOM-generated data representations.
• Case studies demonstrate the workflow's utility in exploring precipitation projections and identifying patterns.

Conclusions:

• ClimateSOM provides a powerful tool for climate scientists to interactively explore and understand ensemble dataset variability.
• The combination of SOMs and LLMs offers a promising approach for analyzing complex climate model outputs.
• Expert reviews confirm the utility and effectiveness of ClimateSOM in generating climate insights.