University of Montana and Oracle bring new precision to weather forecasting

Climate change is creating new and unforeseen weather conditions that demand new technology and creative planning. Traffic in the Arctic is actually increasing as the ice caps melt, bringing more tourists, shipping, and mining. With more traffic also comes a greater need for accurate weather prediction, particularly in one of the harshest environments in the world.

The Weather Research & Forecast (WRF) model is an advanced weather prediction model used for atmospheric prediction and research around the world. The model serves a wide range of uses, generating simulations based on actual and hypothetical weather conditions. Thanks to the contributions of the research community, WRF now reflects many of the latest advances in physics, numerics, and data assimilation.

Because WRF users tend to need high-performance computing to produce these forecasts, few organizations can use this valuable service today. The University of Montana's Autonomous Aerial Systems Office (AASO) is now working to create a WRF model that runs autonomously—first for the Arctic and then for the Pacific Northwest area. For this, the researchers needed access to a cloud computing solution capable of outstripping their on-premises supercomputers.

Oracle for Research offered access to powerful cloud computers that promised to significantly speed up the AASO’s research. Montana clustered 12x BM.HPC2.36 shapes with RDMA networking on Oracle Cloud to run the WRF software, pushing the boundaries on how fast the WRF model could run. Oracle for Research is also providing additional Cloud Architect expertise in HPC, processing, networking, model tuning, and containers to support the project.

UM is using Oracle Cloud Infrastructure (OCI) with Enterprise Database Service to develop and run this autonomous WRF. In preliminary tests, OCI ran the WRF model at high resolution successfully, significantly reducing compute time in comparison to the other HPC resources available. OCI lowers the time it takes to complete a high-resolution weather forecast to such an extent that these forecasts can be completed and shared during the same timeframe for which they are valid.

For example, the research team recently completed a 36-hour model run in about 5 hours with 36 computing cores on OCI. By comparison, the team used to take 15 to 18 hours for this same run using Montana Tech’s HPC, which has a comparable number of cores. As for cost, this compute totaled only $15.01, and storage cost $0.31 for the single run.

This autonomous WRF model has the potential to vastly improve civil engineering, public safety, and national security mission planning in the Arctic and beyond. Supplemental weather forecasts during periods of intermittent or disconnected data streams will also become possible with artificial intelligence (AI) preconditioners to support operational planning in the area.