Careful Mathematical Choices Help More Rapidly Reduce Errors in Simulations
Numerical weather and climate predictions unfailingly contain errors that can be traced to the various finite resolutions of the underlying numerical models. At higher temporal resolutions, a numerical model recalculates the properties of the atmosphere more often and provides more accurate results by better capturing the detailed features of rapidly evolving phenomena. Earlier studies revealed that small-scale physical phenomena, such as the formation of clouds and rain, act as bottlenecks for achieving higher error reduction rates as a global atmospheric model’s temporal resolution increases. This study shows that improving the mathematical assumptions about individual small-scale phenomena and carefully choosing methods for numerically coupling different atmospheric processes can help improve the internal consistency of the simulations, which in turn can help produce higher error reduction rates.
By better reducing errors, the numerical models will converge at more accurate answers faster. This allows for more efficient use of computing resources as model resolution continues to increase. Improved physical realism also provides a more solid basis for the eventual weather and climate predictions made with the results of numerical models. This makes the predictions, ranging from near-term local rain events to regional and global-scale temperature changes over the next few decades, more accurate.
Assessing error reduction with respect to model resolution is routinely performed for atmospheric processes resolved by numerical weather and climate models. However, it is rarely done for the small-scale and unresolvable, yet impactful processes, such as the formation of clouds and rain. Earlier studies revealed these small-scale processes can be the primary culprits of slow error reduction, but why remained unclear. In this study, researchers conducted simulations using an idealized configuration created to facilitate the investigation, and a more sophisticated configuration, similar to those used for actual applications, of a state-of-the-art global climate model. The results indicate that overly simplified assumptions about small-scale processes, and simplistic choices made during the assembly of a complete model, can lead to model behaviors that are not only physically invalid but also mathematically problematic (e.g., singular and/or discontinuous). Addressing these issues at their roots should lead to a model with better internal consistency and higher numerical accuracy as the temporal resolution of a simulation is refined.