Precipitation Rediagnosis: A Use Case for exploring MLOps for research machine learning projects in Weather & Climate
As adotpion of machine learning in weather and climate research increases, we want to move to turning ML outputs into reliable products. To facilitate this, we need to explore how we run projects using analysis ready datasets and that are open and reproducible, and what tools and practices are needed to support that goal.
To support this we have chosen a particular use case machine learning project, in this case Precipitation Rediagnosis. Precipitation is a challenging variable to predict because it can vary so quickly on a local scale, so even advanced NWP models can struggle to predict precipitation at the right time or location. This project aims to explore whether machine learning (ML) can rediagnose precipitation rates based on other NWP outputs. In creating an intial pipeline we will explore tools to recommend for data sharing, scaling reproducible workflows and tracking and sharing parameters and reasults of ML experiments acordding to MLOps best practices. This poster was presented at the 2023 January ESIP Meeting held virtually Jan. 23-27, 2023.