J55.2 Improving the Performance and Scalability of the Colorado Fire Prediction System (CO-FPS) Using Dynamic Cloud Resources

Wednesday, 15 January 2020: 3:15 PM
212 (Boston Convention and Exhibition Center)
James Cowie, NCAR, Boulder, CO; and W. Petzke, J. Boehnert, D. Brucker, N. Chartier, and J. Knievel

The Colorado Fire Prediction System (CO-FPS) couples a WRF-based atmospheric model with a fire-behavior model (WRF-Fire) to predict the evolution of wildfires. Fire simulations are initiated using a web browser application by Department of Public Safety officials who can then direct resources to combat the fire if it will impact public safety or property. With funding from the state of Colorado, the Research Applications Laboratory (RAL) of the National Center for Atmospheric Research (NCAR) began development of the CO-FPS starting in 2015. The project is in the final year of a 5-year program to provide the state with an operational capability.

The system was originally developed to run using on-premise hardware, purchased in the early stages of the project. This hardware would serve as the development platform as well as to provide an Initial Operating Capability (IOC) during the first few years of the project. While the hardware proved adequate for running three fire simulations at once (a contract requirement), most of the time the system was relatively idle due to the seasonal aspect of wildfires. In addition, during peak fire season there was no way to add temporary computing capability to handle those peaks in a cost-effective manner.

During the last two years, work was undertaken to port the system to a cloud vendor, specifically Amazon Web Services (AWS). The system was split into always-running and on-demand components. Always-running components include a host to ingest High Resolution Rapid Refresh (HRRR) data for the WRF-Fire boundary conditions, and an ArcGIS server to provide fire behavior and weather products to users. The WRF-Fire component was modified to run using on-demand compute resources, provisioned as needed when a user requested a simulation. The new cloud-based configuration has many benefits, including better resource utilization, almost unlimited scalability during peak season, and easy upgrade to more powerful compute resources as AWS makes them available in production.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner