In atmospheric modeling, superparameterization (SP) has gained popularity as a technique to improve cloud and convection representations in large-scale models by coupling them locally to cloud-resolving models. We show how the different representations of cloud water in the local and the global models in SP lead to a suppression of cloud advection and ultimately to a systematic underrepresentation of the cloud amount in the large-scale model. We demonstrate this phenomenon in a regional SP experiment with the global model OpenIFS coupled to the local model Dutch Atmospheric Large Eddy Simulation, as well as in an idealized setup, where the large-scale model is replaced by a simple advection scheme. As a starting point for mitigating the problem of suppressed cloud advection, we propose a scheme where the spatial variability of the local model's total water content is enhanced in order to match the global model's cloud condensate amount. The proposed scheme enhances the cloud condensate amount in the test cases, however a large discrepancy remains, caused by rapid dissipation of the clouds added by the proposed scheme.@en

In this study, we investigate uncertainties in a large eddy simulation of the atmosphere, employing modern uncertainty quantification methods that have hardly been used yet in this context. When analysing the uncertainty of model results, one can distinguish between uncertainty related to physical parameters whose values are not exactly known, and uncertainty related to modelling choices such as the selection of numerical discretization methods, of the spatial domain size and resolution, and the use of different model formulations. While the former kind is commonly studied e.g. with forward uncertainty propagation, we explore the use of such techniques to also assess the latter kind. From a climate modelling perspective, uncertainties in the convective response and cloud formation are of particular interest, since these affect the cloud-climate feedback, one of the dominant sources of uncertainty in current climate models. Therefore we analyse the DALES model in the RICO case, a well-studied convection benchmark. We use the VECMA toolkit for uncertainty propagation, assessing uncertainties stemming from physical parameters as well as from modelling choices. We find substantial uncertainties due to small random initial state perturbations, and that the choice of advection scheme is the most influential of the modelling choices we assessed. This article is part of the theme issue 'Reliability and reproducibility in computational science: implementing verification, validation and uncertainty quantification in silico'.

@en

In order to eliminate climate uncertainty w.r.t. cloud and convection parametrizations, superpramaterization (SP) [1] has emerged as one of the possible ways forward. We have implemented (regional) superparametrization of the ECMWF weather model OpenIFS [2] by cloud-resolving, three-dimensional large-eddy simulations. This setup, described in [3], contains a two-way coupling between a global meteorological model that resolves large-scale dynamics, with many local instances of the Dutch Atmospheric Large Eddy Simulation (DALES) [4], resolving cloud and boundary layer physics. The model is currently prohibitively expensive to run over climate or even seasonal time scales, and a global SP requires the allocation of millions of cores. In this paper, we study the performance and scaling behavior of the LES models and the coupling code and present our implemented optimizations. We mimic the observed load imbalance with a simple performance model and present strategies to improve hardware utilization in order to assess the feasibility of a world-covering superparametrization. We conclude that (quasi-)dynamical load-balancing can significantly reduce the runtime for such large-scale systems with wide variability in LES time-stepping speeds.

@en

We present a Python interface for the Dutch Atmospheric Large Eddy Simulation (DALES), an existing Fortran code for high-resolution, turbulence-resolving simulation of atmospheric physics. The interface is based on an infrastructure for remote and parallel function calls and makes it possible to use and control the DALES weather simulations from a Python context. The interface is designed within the OMUSE framework, and allows the user to set up and control the simulation, apply perturbations and forcings, collect and analyse data in real time without exposing the user to the details of setting up and running the parallel Fortran DALES code. Another significant possibility is coupling the DALES simulation to other models, for example larger scale numerical weather prediction (NWP) models that can supply realistic lateral boundary conditions. Finally, the Python interface to DALES can serve as an educational tool for exploring weather dynamics, which we demonstrate with an example Jupyter notebook.

@en

Here, we describe our efforts to create a multi-scale and multi-physics framework that can be retargeted across different disciplines. Currently we have implemented our approach in the astrophysical domain, for which we developed AMUSE (github.com/amusecode/amuse ), and generalized this to the oceanographic and climate sciences, which led to the development of OMUSE (bitbucket.org/omuse ). The objective of this paper is to document the design choices that led to the successful implementation of these frameworks as well as the future challenges in applying this approach to other domains.@en

As a computationally attractive alternative for global large eddy simulations (LESs), we investigate the possibility of using comprehensive three-dimensional LESs as a superparameterization that can replace all traditional parameterizations of atmospheric processes that are currently used in global models. We present the technical design for a replacement of the parameterization for clouds, convection, and turbulence of the global atmospheric model of the European Centre for Medium-Range Weather Forecasts by the Dutch Atmospheric Large Eddy Simulation model. The model coupling consists of bidirectional data exchange between the global model and the high-resolution LES models embedded within the columns of the global model. Our setup allows for selective superparameterization, that is, for applying superparameterization in local regions selected by the user, while keeping the standard parameterization of the global model intact outside this region. Computationally, this setup can result in major geographic load imbalance, because of the large difference in computational load between superparameterized and nonsuperparameterized model columns. To resolve this issue, we use a modular design where the local and global models are kept as distinct model codes and organize the model coupling such that all the local models run in parallel, separate from the global model. First simulation results, employing this design, demonstrate the potential of our approach.

@en