Plume: A Plugin Mechanism for Numerical Weather Prediction Models

Plume (Plugin Mechanism) is a plugin system for Numerical Weather Prediction models (NWP) developed by ECMWF as part of the EU Destination Earth initiative. Plume loads plugins at runtime and lets them access model data in memory, through a well-defined interface. Plume plugins offer scientists and third-parties a controlled and flexible execution environment which allows them to easily extend specific functionalities of the model. Plume complements a collection of ECMWF tools that address the demand for scalability, given the very high resolution of the upcoming Destination Earth digital twins (DT) and the continuous increases in resolution of ECMWF's operational forecasts. 

Examples of plugin applications include regional sub-models, light-weight data extraction and analysis and even machine learning algorithms. Building these applications as plugins with direct access to in-memory model data, rather than as separate applications coupled via the IO system, creates a more efficient and scalable system for very-high resolution forecasts. Plume is designed to allow coupling of plugins to different NWP models, including ECMWF's IFS, using a common interface for NWP data (Atlas). Plume is primarily written in C++, but supports plugins written in C++ and Fortran.  

This presentation focuses on Plume architecture, implementation and initial interface design. Some plugin examples are used to demonstrate how Plume can access and interact with the ECMWF Integrated Forecasting System (IFS) and extend some specific functionalities.