Wednesday, 31 January 2024: 11:30 AM
320 (The Baltimore Convention Center)
Robert Bartel, NWS, Tuscaloosa, AL; Lynker, Leesburg, VA; and T. C. Flowers, N. Frazier, A. Raney, C. O. Tubbs, and M. Williamson
The Office of Water Prediction’s (OWP) Distributed Model On Demand (DMOD) platform provides optimization tools that automate many of the tedious processes in model setup and execution, reducing the cognitive and manual burdens of developing, testing, and experimenting with models. The DMOD platform accounts for many of the practical concerns beyond the underlying domain science, such as data management, model configuration, and compute environment considerations. Technical and organizational tasks such as these impose a considerable mental load, much of which falls to domain scientists. DMOD allows users to instead leverage its capabilities to carry much of this load.
We accomplish this reduction in the mental burden via a combination of abstracting the compute environment infrastructure and providing component services that manage and simplify common model execution, evaluation, and development tasks. These enable functionality such as orchestrating the availability of input and output data for executing tasks and users; dynamically assembling and generating model configuration files; provisioning and executing model processes; and integrating model execution with tools for calibration, evaluation, and visualization. Here, we present an overview of DMOD, demonstrating several examples of DMOD’s capabilities and discussing how these features can aid in developing, testing, and executing models.

- Indicates paper has been withdrawn from meeting

- Indicates an Award Winner