Position paper on high performance computing needs in earth system prediction

Abstract

The United States experiences some of the most severe weather on Earth. Extreme weather or climate events - such as hurricanes, tornadoes, flooding, drought, and heat waves - can devastate communities and businesses, cause loss of life and property, and impact valuable infrastructure and natural resources. The number and severity of extreme weather and climate events in the U.S. has risen since 1980, and is projected to continue rising this century. Growing populations in vulnerable areas create increased risks. If current trends continue, damages from extreme weather and climate events could grow four-fold by 2050. Predictions and projections of weather and extreme events across time scales from weather to climate rely on sophisticated numerical models running on High Performance Computing (HPC) systems, which press the frontier of the Nation’s HPC capability. The Nation’s Earth system modeling community has a unique set of HPC requirements which differ from industry needs. Typically, HPC advances are measured using computational peak performance metrics that are ill-suited to Earth system modeling applications. We advocate for a shift in processor design to increase emphasis on memory bandwidth, so Earth system models run more efficiently and better serve the public need.