This presentation focuses on our implementation of containerizing the full workflow of SHiELD (System for High-resolution prediction on Earth-to-Local Domains), a unified modeling system for weather-to-subseasonal prediction developed at the Geophysical Fluid Dynamics Laboratory (GFDL). We have created publicly available images for each of the three workflow stages: pre-processing, model execution, and post-processing. Utilizing Docker’s multi-arch build feature, these images support both ARM64 and AMD64 architectures. They are also compatible with HPCs using Apptainer (formerly Singularity) as the container software.
The containerized SHiELD provides a standardized environment, eliminating the need for any extensive setup work by the user. We will demonstrate live the ease of installation and execution by showcasing a global-nest simulation of a hurricane case deployed from scratch on an Apple M1 laptop. Additionally, we will explain how existing development workflows can be seamlessly adapted to incorporate container technology. Finally, we discuss the challenges faced during our design process and our efforts to improve the security of distributing the containerized SHiELD. Our complete procedure and code are available at github.com/jzhang512/Containerized_SHiELD_Workflow.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner