Parallel Numerical Methods for Model Coupling in Nutrient Cycle Simulations

We present a new approach to ecosystem nutrient cycle simulations. Nutrient conversion and fluxes between ecosystem compartments are driven by highly complex biogeochemical and hydrological processes. Nutrient cycles in soils supply vegetation and crop growth, affect groundwater and surface water eutrophication, and release greenhouse gases into the atmosphere. Simulating nutrient cycles is regarded a grand challenge due to the multi-scale properties and involved multiphysics of the considered ecosystems. Common approaches are restricted on several levels, often suffering from limited temporal and spatial extent, resolution or accuracy, model simplifications, or inability to use high performance computing effectively. In order to cope with their inherent complexity, we consider nutrient cycle simulations as a multiphysics problem by means of a coupling of dedicated biogeochemical and hydrological models. We formulate the model coupling problem in an abstract multiphysics setup to manage the complexity. We propose a new variant of operator splitting schemes for the time propagation of the coupled models. Our scheme employs local model propagators such that accuracy is maintained on the global level. Furthermore, the scheme features an inherent parallelism on the coupling level which is independent of possible parallelizations of the models. We present advances in a software technique which facilitates the model coupling on high performance computing platforms. Our development allows for dynamically changing the parallel configuration of models and their data distribution during runtime. We validate our approach to nutrient cycle simulations by means of numerical experiments using a greenhouse gas emission scenario and a nitrate leaching scenario. The results show the benefits of the proposed scheme in terms of an improved coverage of indirect and local effects. Furthermore, we present performance tests showing the superior parallel efficiency of our methodology over common approaches, which is due to its parallelism with respect to the coupling strategy.