Porting Noah-MP Model into Energy Research and Forecasting Model Using a Novel Large Language Model Paradigm

The DOE Wind Energy Technology Office has funded the development of the Energy Research and Forecasting (ERF) model, which is designed to enhance the performance of atmospheric simulations traditionally conducted with the Weather Research and Forecasting (WRF) model. ERF is built on AMReX, a block-structured Adaptive Mesh Refinement (AMR) library that provides performance portability, enabling ERF to efficiently run atmospheric simulations on modern computing architectures, including both CPU and GPU platforms. To address challenges similar to those tackled by WRF, ERF requires a robust land surface model. This presentation focuses on the integration of the Noah-MP land surface model within ERF’s framework. Rather than directly porting the native Noah-MP code from WRF into ERF, we explore an innovative approach that uses prompting strategies to guide large language models (LLMs) in generating APIs for FORTRAN-C++ interoperability between ERF and Noah-MP. Customized chat templates were developed to instruct LLMs on code conversion rules, facilitating the conversion of source files.