ABSTRACT
This practice offers guidance on how to use the PROcess-guided deep learning and DAta-driven modeling (PRODA) approach to integrate observations with the biogeochemical module of the Community Land Model version 5 (CLM5) to best represent regional soil organic carbon distributions. Over three exercises, we focus on how to build, train, and tune a deep learning model in the PRODA approach to predict parameters estimated from site-level data assimilation. Readers can use the CarboTrain platform to explore different deep learning options and to understand and modify the optimization methods.
