Using Large Language Models to Produce Literature Reviews: Usage Patterns of Microphysics Parameterizations in WRF Simulations from 2699 Publications

Abstract Large language models (LLMs) afford opportunities for using computers for intensive tasks, providing research opportunities that have not been considered before. One such opportunity is a systematic interrogation of the scientific literature. Here, we show how an LLM assists in constructing a literature review, focusing on feasibility, process design, and resulting outcomes. We constructed a database of 2699 publications associated with precipitation simulations in the Weather Research and Forecasting (WRF) Model, with the goal of learning how microphysics parameterizations were used and reviewing forecast errors across a diverse array of precipitation simulations. The large language model GPT-4 Turbo was used to extract information about model configurations and performance from the text of 2699 publications identified using Web of Science and Scopus searches. Our results reveal the landscape of how nine of the most popular microphysics parameterizations have been used around the world: Lin, Ferrier, WRF single-moment 3-class, 5-class, 6-class, Goddard Cumulus Ensemble, Morrison, Thompson, and WRF double-moment. More studies used single-moment parameterizations before 2020 and double-moment parameterizations after 2020. Except for simulations using Lin, Ferrier, and Goddard that tended to underestimate precipitation over almost all locations, the remaining six parameterizations tended to overestimate precipitation, particularly over China, Southeast Asia, the western United States, and central Africa. This method could be used by other researchers to help understand how the increasingly massive body of scientific literature can be harnessed through the power of artificial intelligence to solve their research problems. Significance Statement Microphysics parameterization is used to represent unresolved cloud and precipitation processes and hydrometeor populations at subgrid scales in numerical weather prediction models. However, to our knowledge, no systematic literature review, meta-analysis, or other large study exists on the usage and performance of different microphysics parameterizations. This study used a large language model to analyze 2699 scientific publications associated with precipitation simulations in the Weather Research and Forecasting (WRF) Model. The most popular parameterization was the WRF single-moment 6-class microphysics scheme (WSM6) before 2020 but Thompson after 2020. Seven out of nine parameterizations tended to overestimate precipitation. Our results synthesize previous studies on microphysics parameterizations, summarizing their reported performance in precipitation simulations. Moreover, this study demonstrates a novel approach to solving scientific problems through the large-scale analysis of existing scientific publications.