Conveners: Katherine Evans and Matthew Norman, Oak Ridge National Laboratory
Substantial and ongoing development of multiscale Earth system models has led to a broad understanding of the water cycle and its variability from local to global scales. Improved predictions of hydrological behavior will require better resolution of multiple spatial scales, representation of missing physical processes and feedbacks, and quantified uncertainties in the simulations. However, these developments will dramatically increase model cost and complexity and will tax even the largest current supercomputers. Looking to the next generation of computing capability to provide the necessary throughput, power and cooling limitations are driving computer architecture trends toward increased parallelism at all levels, heterogeneous nodes that include accelerators like GPUs, and more complex memory and storage hierarchies. New generations of these systems differ significantly from their predecessors in terms of core layouts, network interfaces, and memory systems and will continue to evolve rapidly. For this session, we solicit submissions that focus on recent successes and challenges developing and analyzing models of the water cycle that target the next generation of supercomputers.