Using Cloud Computing and Software Container Technology for Interactive Classroom Learning in Numerical Weather Prediction

Numerical Weather Prediction (NWP) is at the core of weather research, modeling, and forecasting. Unfortunately, in meteorology and other atmospheric sciences, students often find themselves with little practical experience in actually performing NWP, due to the many software- and hardware-related challenges involved in setting up, designing, and running an end-to-end NWP system.

On the software side, computer programs necessary to perform NWP require extensive effort and knowledge to ensure all dependencies, libraries, and programs are properly installed. Procedures and dependencies vary widely across different platforms and operating systems, and even if successful, it is non-trivial to set up a proper end-to-end workflow that will easily move data from one portion of the system to the other. Containerized software allows for all the components (including the operating system, source code, library dependencies, and executables) required to build and run a software system to be bundled together and deployed on any platform.

On the hardware side, adequate computer resources (processors, memory, storage, etc.) are hard to come by, especially in a classroom environment, where significant compute power is required to allow many students to run simulations at the same time in a way that can be completed in a reasonable amount of time. It can be very cost-ineffective to purchase hardware just for classroom exercises, which will see very little utilization outside of a few hours a week where students are actively using them. Cloud computing provides a solution in the form of on-demand computing resources, with essentially unlimited power, yet you only pay for the time and resources you actually use.

Combining the accessibility of cloud computing platforms with the portability of software containers offers a solution for atmospheric science programs that may not otherwise have the computing resources to allow students to use and experiment with a full NWP system. This year the Developmental Testbed Center (DTC) has partnered with the Metropolitan State University of Denver to design a meteorology laboratory class based around running full, end-to-end NWP simulations using Docker software containers and Amazon Web Services (AWS) cloud computing. This end-to-end system of containers makes use of Gridpoint Statistical Interpolation (GSI) for data assimilation, the Weather Research and Forecasting model (WRF) and the Unified Post-Processor (UPP) for producing NWP model output, Model Evaluation Tools (MET) for evaluating model performance, and NCAR Command Language (NCL) scripts and METviewer for visualizing model output and statistics.

This presentation will demonstrate the classroom exercises performed by students in this year’s “Forecasting Laboratory” class at Metro State University, demonstrating not just the ease of use but also the ability for students to make changes to settings, input data, and other parameters to gain a better understanding of how NWP systems produce forecasts.