Making 1+1 Bigger Than 2
Modern computer models for weather, climate, and earth systems contain numerous modules that simulate the complex workings of distinct physical and dynamical phenomena, such as ocean currents, atmospheric winds, clouds, and river flows. These individual modules are typically developed by separate groups of researchers and then connected to form a comprehensive model system. The two-step approach can lead to inaccurate results in current model simulations. An international team of researchers, including scientists from the U.S. Department of Energy’s Pacific Northwest National Laboratory, reviewed the state of the science in module coupling and identified possible ways to address some key challenges.
Most model development work typically focuses on individual modules, resulting in a research gap in the simulation of weather and climate, which are the sum of many processes. This review paper demonstrates the pervasiveness of coupling problems in current models and highlights recent progress in module coupling. The paper also provides illustrative examples of coupling issues, such as insufficient frequency of information exchange between modules and double counting of processes by multiple modules. Such issues could, for example, strongly affect how much rain is predicted in a simulation of storms. The coupling issues must be overcome for each module to realize its full potential and improve the overall predictive skill of the entire model system. This is needed for the improvement of existing modules as well as in the design of new modules or model systems.
The compartmentalization of model codes and development teams is natural and also necessary to manage the endeavor of modeling complex weather, climate, and Earth systems, which are influenced by many processes. As a result, three types of issues can occur: First, the artificial distinction between physical phenomena based on whether they can be spatially resolved or not can lead to double counting or undercounting of processes. Second, infrequent information exchange between interacting modules can result in biases, or offsets from observations, in the balance between them. Third, inconsistent approximations employed by different modules can lead to unintended violation of basic physical principles. The topic of coupling typically receives little attention during model development, and more evidence suggests that this gap in research is becoming a bottleneck in further improving the corresponding models.
Aimed at increasing awareness of such issues, this review describes the symptoms of coupling problems using examples from existing literature and discusses their root causes and possible methods for analyzing the problems. The paper reflects the findings of many research highlights from an international biennial workshop series on physics-dynamics coupling. Since the first workshop in 2014, this small but evolving event series has gained attention from international weather, climate and Earth system model development groups and research sponsors to join forward-thinking discussions and address the ubiquitous coupling problems. Topics discussed in the workshop series and the paper include: conceptual inconsistencies in model equations; numerical methods for solving the equations with computers; sensitivities of model results to choices made for the numerical calculations and strategies for analyzing such sensitivities; and methods for making the computations more efficient on supercomputers.