The Energy Exascale Earth System Model (E3SM), like many other global system models, consists of modular components that are each responsible for a different subset of processes in the global system. Typically, the system state is advanced in time by sequential splitting: the state is advanced by a single process at a time and then passed to the next process in a predetermined sequence. While sequential splitting is very straightforward to implement with modular components, as one component simply passes the updated state to the next component, the associated splitting error is highly dependent on the ordering of the processes in the sequence. Various other splitting methods have been considered in the atmosphere community, including parallel splitting where all components operate in parallel on the same state and then combine all the resulting states into a single updated state. The presence of subcycling/substepping in one or more components bring the opportunity for even more approaches, including multirate infinitesimal step (MIS) methods and “dribbling” methods that distribute the action/tendency of one process across the subcycles/substeps of another process.
This work introduces an error analysis framework for identifying the splitting error independent of the other temporal discretization errors. Such a framework allows for obtaining the splitting error even when the temporal discretization of the individual components is too complex, or perhaps not documented enough, for standard truncation error analysis. This work uses the framework to evaluate sequential splitting, parallel splitting, “dribbling”, and MIS methods applied to a two-process system with subcycling/substepping. The results are interpreted in the context of coupling the cloud processes component of the E3SM atmospheric model (namely, CLUBB and MG2) with the rest of the E3SM atmospheric model components (dynamics, radiation, etc.). This work will compare the costs and benefits of the various methods against the current sequential splitting approach.
Prepared by LLNL under Contract DE-AC52-07NA2. LLNL-ABS-825270.