Second law of thermodynamics in modelling
Almut Gassmann (Institute of Atmospheric Physics)
Abstract
The subgrid-scale terms in numerical modeling must contribute to positive internal entropy production.
The second law of thermodynamics determines the direction of subgrid scale fluxes, but not their amount. The general task of parameterizations is thus to estimate the strength of the fluxes.
The poster presents where the usual approaches to subgrid-scale fluxes conform with the second law and where they do not.
A prominent example, where traditional approaches do not conform with the second law is the heat flux at stable stratification. It can be shown that such a formulation may lead to the generation of resolved gravity waves instead of the attenuation of gravity waves. The 2nd law conforming formulation leads to the expected attenuation of gravity waves.
Also other typical subgridscale flux directions do not exactly conform with the second law. This is for instance the turbulent moisture flux which is typically parameterized down the specific moisture gradient. The correct gradient approach depends on the partial pressures of dry air and water vapour. This formulation takes automatically into account that the lighter water molecules are mixed upward with a higher probability than dry air. The difference of this 2nd law conforming approach to the traditional approach is presumably small.
Understanding Physics-Dynamics coupling in a simple 2D model
Susannah Hearn (University of Exeter)
Abstract
How we include small-scale processes in large scale models is important in understanding and improving the accuracy of models. Challenges arise in weather and climate modelling when attempting to incorporate subgrid processes which a model cannot resolve, or diabatic sources, such as radiation, which involve extra terms in the governing equations (Gross et al, 2016). Often, these challenges involve the application of numerical methods and resulting errors. Simplified models can provide a method of analysing coupling strategies (Gross et al, 2016) and so the intention is to investigate coupling between convection, the boundary layer and large-scale dynamics in a simplified 2D model setting.
Numerical Analysis of Multi-Fluid Equations
William McIntyre (University of Reading)
Abstract
Shallow and deep convection have historically been statistically parameterized as they occur in far smaller horizontal scales relative to the resolutions of weather prediction models. Modern resolutions are too high to make some large-scale statistical assumptions and too coarse to model the convection outright meaning we currently find ourselves in the "grey zone" of convection modelling. A technique known as conditional averaging offers an alternative approach to model convection in the grey zone by treating the convective regions as a separate fluid which interacts with the remainder of the environment. However, this approach presents additional challenges in producing a stable and accurate numerical scheme. We present extensive analysis of the conditionally averaged (or multi-fluid) shallow water equations to inform the stable treatment of the multi-fluid fully compressible Euler equations for convection modelling.
The ECMWF coupled atmosphere-wave-ocean-sea ice modelling system
Kristian S. Mogensen (ECMWF)
Abstract
All operational forecasts issued by ECMWF are either currently based on a coupled atmosphere-wave-ocean-seaice model or will soon be based on this model in the very near future with the with HRES-Coupled (9km global) completing the coupling upgrade of the Integrated Forecasting System (IFS) in 2018.
The poster will describe the flow of information between the different component models and how we have implemented a single executable coupled model. The single executable is based on a single time step loop where we call the wave (WAM) and ocean+seaice (NEMO with OPA/LIM2) models within the atmospheric time stepping. In our current setup, we typically call the wave model every atmospheric time step and the ocean model every model hour. The input fields from both the atmosphere and the wave models to the ocean model are averaged over the hourly coupling interval whereas instantaneous ocean fields are passed back to the atmosphere/waves after the ocean model has been integrated. This means that after each coupling time step all models have the same valid time and the coupling sequence can start over.
Advantages and disadvantages with our approach from both a technical and scientific point of view will be discussed.
Towards a turbulent multi-fluid approach to modeling atmospheric convection
Daniel Shipley (University of Reading)
Abstract
A general framework for parameterising convection has recently been proposed, based on conditionally averaging the compressible Euler equations. This results in a set of equations for multiple fluids, which are coupled together by terms involving the transfer of momentum, entropy, moisture, and tracers between different partitions. The problem of parameterising convection then becomes the problem of parameterising these transfer terms. Traditional convection models such as mass flux-based schemes and eddy diffusivity-mass flux schemes can be seen as specific limits of this framework. However, the multi-fluid approach is much more general: for example allowing for a fully 3-dimensional treatment, and net mass transport due to convection. This makes it a good candidate for building parameterisations at current “convection-permitting” resolutions and beyond, where the underlying assumptions of traditional models break down.
We model the effects of convection-permitting resolution on two common test cases for convection: a “rising bubble” of buoyant air (e.g. Bryan & Fritsch (2002)), and Rayleigh-Bénard convection. Our simulations use the fully compressible Navier-Stokes equations in 2D, but do not include the effects of moisture. We simulate the effects of coarse resolution by introducing an artificial high viscosity. This magnifies diffusive effects, to the extent that buoyant instability can be suppressed. However, it is required in many atmospheric models to maintain stability. The coarse-resolution runs are compared to a high resolution “truth” which parameterises sub-grid scale turbulence via a Smagorinsky LES scheme. High viscosities at coarse resolution are found to suppress buoyant convection in both test cases. By parameterising the transition between fluids based on a buoyancy condition, we hope to recover the transition to convection even in the coarse resolution and high viscosity case. This would serve as a proof-of-concept for the multi-fluid approach to parameterising convection at convection-permitting resolutions.
In future work we will investigate the effects of moisture on the above experiments, particularly the effect of latent heat flux on buoyancy production. Beyond that we plan to move to more realistic test cases, such as models based on the Barbados Oceanographic and Meteorological Experiment.
Towards a better understanding of coupling strategies by use of aprotypical 1D ocean-ice interaction
Anusha Sunkisala (University of Hamburg)
Abstract
In earth system models (ESM) a number of sub-components, like atmosphere, ocean, terrestrial and cryospheric systems are coupled with each other. Most often these sub-systems communicate with each other by exchanging boundary values or fluxes at their interfaces. In many cases the communication intervals are relatively large due to the different time scales of the corresponding sub-components or simply due to computational cost considerations.
A rigorous mathematical analysis of the sensitivity of such coupling strategies on the overall system state (the solution of the system) is very difficult and to the authors knowledge has not yet been performed. In order to gain quantitative information on the effect of different coupling strategies, in particular coupling time scales and quantities, and convergence properties - a prototypical one-dimensional model of heat transfer between ice and ocean has been employed to test different coupling strategies. This ”toy problem” can be solved semi-analytically for both phases simultaneously such that a system solution is available. We have considered two separate domains for each phase and implemented two different coupling strategies (1)by adding the fluxes in each domain and (2) by adding and subtracting the average fluxes in separate domains.These two cases are compared with the system solution and they show significant differences. As a preliminary conclusion, our study suggests to investigate currently used coupling strategies with more rigor and in particular for long term paleo-climate simulations.
Choices for the physics-dynamics interface in ALADIN System for operational forecast
Martina Tudor (Croatian Meteorological and Hydrological Service)
Abstract
The operational forecast is computed on a LAM domain using ALADIN System. There are few options for physics dynamics coupling and interface. The consequences these choices have on the operational forecast of different weather conditions will be shown on several cases of moderate and extreme weather conditions.
Recent activity of improving physics-dynamics coupling in the JMA Global Spectral Model
Masashi Ujiie (Japan Meteorological Agency)
Abstract
The Japan Meteorological Agency (JMA) has operated the operational Global Spectral Model (GSM) since 1988. Although a number of improvements to the model in resolution, dynamical core (Eulerian to Semi-Lagrangian), and parameterization schemes have been made for over 30 years, little attention has been paid to its structure of time-integration in context of physics-dynamics or physics-physics coupling. However, our recent investigation has revealed problems in the its time integration and has made us recognize importance of the coupling. For instance, larger time stepping, associated with introduction of a Semi-Lagrangian scheme, has highlighted risks of numerical noises and undesirable time-step dependency in the model physics. These risks can prevent us from introducing more complex physics parameterizations, such as refinement of convection, boundary layer and of orographic drag schemes which contain fast processes.
For future sustainable model development, we attempt to refine the time-integration structure, such as (i) exploring implications of how good (in stability and consistency) our current physics-dynamics coupling is from simplified linear analysis, (ii) reconsideration of coupling “fast physics” with dynamics/other physics, (iii) testing a joint implicit solver for boundary layer and orographic drag. We will show and discuss our recent activities in the poster.
Common Community Physics Package (CCPP): a bridge between community physics advancements and NWP models
Man Zhang (NOAA Earth System Research)
Abstract
The Global Model Test Bed (GMTB) has been tasked by the U.S. NOAA Next Generation Global Prediction System (NGGPS) program to develop a Common Community Physics Package (CCPP), which aims at lowering the bar for community scientists to conduct tests and contribute innovations to be considered for operational implementation.
Once it is fully developed, the CCPP will contain a variety of physical parameterizations, grouped in suites, all fully documented and supported. As a starting point, the CCPP currently contains the physics suite used in the operational Global Forecast System (GFS) model implemented in 2017. Going forward, both operational and next-generation parameterizations will be available to support distributed development.
The first public release of the CCPP is scheduled for the spring of 2018 to run with the GMTB Single-Column Model. A subsequent release, planned for the Summer of 2018, will contain the capability to run the CCPP with NOAA's Finite-Volume on a Cubed-Sphere (FV3) Dynamical Core Global Forecast System (FV3GFS).
Employing a "plug and play" philosophy, the CCPP features well-defined interfaces for physical parameterizations, emphasizing portability and modularity, with clear separation between physical parameterizations and the dynamical core through an Interoperable Physics Driver (IPD), which permits using the CCPP with a variety of dynamical cores. The CCPP supports running the parameterizations in user-defined order, grouping the parameterizations in a single step or subdividing them in various steps with other processes (such as dynamics) in between, and subcycling (running some parameterizations with shorter time steps than others). As such, the CCPP is a useful tool for experimentation in physics-dynamics coupling.