Title: Climate Diagnostic Retrievals using a GPT Language Model
Presenting Author: Michael Hendrickson
Organization: NASA Goddard Institute for Space Studies
Co-Author(s): Alexander Herron and Maxwell Elling

Abstract:
With the increasing urgency of understanding climate change impacts, there is a critical need for accessible and user-friendly tools to interact with complex climate data. In this work, we propose a novel approach utilizing a language model, specifically GPT (Generative Pre-trained Transformer), to streamline the process of accessing, analyzing, and visualizing climate model output. Central to our approach is the model's capability to handle human-readable queries, facilitating intuitive data interaction for users. The model bridges the gap between user queries and desired analyses by initially identifying pertinent climate model outputs and variables. We intend to leverage NLP (Natural Language Processing) tools like LangChain to utilize metadata linked to models and variables, ensuring precise interpretation of user queries. Subsequently, with the integration of the GPT advanced data analysis plugin, the climate model data can undergo automated analysis and visualization. As an illustrative example, suppose a user inputs, ""Show me how temperatures will increase over the next 100 years under a moderate CO2 emissions scenario."" In response, our model would curate a list of temperature datasets from a relevant climate model experiment, such as the Shared Socioeconomic Pathway (SSP) 2-4.5 experiments, which assume moderate future CO2 emissions. The model then conducts data analysis and generates visualizations, such as the global mean temperature time series. By offering a user-friendly interface to climate model data, our approach aims to widen access to climate science knowledge. Leveraging a language model enables non-technical troubleshooting, empowering users to guide the model toward their desired output. This tool has the potential to expand the audience of climate researchers beyond those with traditional data visualization and analysis expertise, thus facilitating greater understanding and informed decision-making regarding climate change.