Efficient and Scalable Time-Stepping Algorithms and Reduced-Order Modeling for Ocean System Simulation
Ocean circulation is affected by processes occurring over a wide range of spatial and temporal scales, from sub-kilometer scale features near coasts to global-scale currents such as the Gulf Stream and time scales ranging from daily responses to wind and heating to century-scale mixing and deep-ocean circulation. Ocean models must faithfully represent these phenomena for accurate simulations using earth system models. Simulating these disparate spatial and temporal scales require computations that, even when using the latest computational algorithms implemented on the latest high-performance computers, are not efficient enough to produce reliable predictions of ocean behavior over long time periods. Over the last few years, the team assembled for the the new effort has developed new algorithms that have the promise to be substantially more efficient compared to current algorithms used in ocean models while also preserving their fidelity. However, the new approaches have so far only been constructed and tested in simplified physical settings. To fulfill the promise of these new techniques, the team will first extend and thoroughly test the new algorithms in realistic ocean settings. Parallel to this task is the important task of integrating the new approaches into the ocean component (MPAS-Ocean) of the Energy Exascale Earth System Model (E3SM) under development under the leadership of the Los Alamos National Laboratory.
The new algorithms will improve other aspects of ocean simulation as well, especially for exploring how tracers (e.g. biochemical and other agents) are transported by ocean currents. Again, multiple time scales make the accurate and efficient simulation of tracer transport a difficult endeavor. As was the case for ocean dynamics, in recent years, the team has made progress in this setting. Building on that progress, the new effort will address the further development of new, more efficient, algorithms for tracer transport that will be integrated into MPAS-Ocean and evaluated in realistic ocean settings.
A higher-risk endeavor is the development of reduced-order models (ROMs) for ocean modeling. ROMs are very inexpensive models that are informed and constructed using several runs of an expensive ocean model. Their relative simplicity and lower cost permit a vastly greater number of simulations. As a result, ROMs can prove useful for gathering accurate statistical information about future ocean behavior, studying the sensitivity of such behavior when inputs are varied, and other time-consuming applications. Although ROMs have proven to be very successful in many other settings, relatively little has been done specifically for ocean modeling. Thus, another task is to develop a ROM capability for the ocean, again building on some very preliminary studies the team has made in the recent past.
The aforementioned tasks require the invention and thorough testing of new algorithms and a study of their mathematical properties (e.g., stability, convergence, conservation, scalable implementation on HPCs) that are important to their successful use in ocean modeling. It also requires the full integration of those algorithms that are proven to be effective into the E3SM. The interdisciplinary team assembled is very well suited for successfully completing the tasks.