Virtual Event: Annual Seminar 2020
ECMWF
This seminar reviewed recent advances in numerical methods for atmospheric and oceanic models that form part of modern weather prediction systems. There are several objectives that drive research and development in numerical methods, such as improving the accuracy and efficiency of discretisations and addressing long-standing biases; improving conservation properties; improving the coupling of the dynamical core with physical parametrizations and other Earth system components; and improving uncertainty estimates in ensemble predictions.
A closely related topic covered in this seminar was how to develop efficient time-critical, very high-resolution global solvers on massively parallel architectures, addressing demanding communication and algorithmic scalability problems. This need has stimulated research and development not only in new numerical methods but also in the adaptation of existing numerical algorithms to enable their efficient execution with heterogeneous high-performance computing architectures based on accelerators. Both are essential activities that allow global weather prediction centres to exploit new technologies for implementing demanding next-generation high-resolution forecasting systems.
Seminar aims
The seminar is part of ECMWF's educational programme and is aimed at early career scientists but also more established scientists that want to engage more with numerical methods for atmospheric and oceanic modelling.
Organising committee
Michail Diamantakis, Samuel Hatfield, Christian Kühnlein, Sarah-Jane Lock, Kristian Mogensen, Andreas Müller, Inna Polichtchouk
-
-
09:45
→
11:30
Moderator: Michail Diamantakis (ECMWF)
- 09:45
-
10:00
Towards a Digital Twin of the Earth System 45m
In this talk I will briefly review the development and state-of-the-art of the Integrated Forecasting System (IFS) model as well as steps taken to secure the future efficiency of IFS in view of emerging HPC architectures, and in view of increasing demand for complex coupled data assimilation and simulations of the hydrological and carbon cycle. Exploring the limits of IFS for simulations with an explicit representation of deep convection, we have recently completed the world’s first seasonal timescale global simulation (DJF 2019) of the Earth’s atmosphere with 1.4 km average grid-spacing using the power of Summit (No2, Top 500) as part of an INCITE award. Despite the significant cost, global simulations at resolutions of about 1 km have been advocated as a way forward commensurate with the challenges posed by climate change. The achieved simulation represents a milestone in atmospheric modelling and the resulting output will serve as a benchmark dataset for a number of scientific studies, including the support of future satellite mission planning. The achieved simulation may be seen as a prototype contribution to a future "digital twin" of our Earth. Results show that the hydrostatic NWP model configuration of the IFS performs well even at an average 1.4 km grid-spacing. This seems to challenge a common belief in dynamical meteorology that assumes that non-hydrostatic equations would be required at this level of resolution. The impact of non-hydrostatic effects is not known, but our simulation provides a baseline against which future non-hydrostatic simulations can be measured. In this sense, the performed simulation provides factual numbers of what it takes to reach the advocated goal of km-scale weather and climate predictions.
Speaker: Nils Wedi (ECMWF) -
10:45
GungHo: Designing a next generation dynamical core for weather & climate prediction 45m
To meet the challenges posed by future developments of supercomputers the Met Offices’ Unified Model for climate and weather prediction is being redesigned with the aim of achieving improved scalability and portability whilst remaining at least as accurate as the current model. The new model is named LFRic after Lewis Fry Richardson.
The dynamical core of this new model, Gungho, uses explicit finite-volume type discretizations for the transport of fields coupled with an iterated-implicit, mixed finite-element discretization for all other terms. This gives a stable & accurate method whilst maintaining many of the mimetic and conservation properties of the underlying equations on arbitrary quasi-uniform meshes. The Gungho dynamical core is then coupled to the existing sub-grid physical parameterizations used in the current Unified Model to form an atmospheric model suitable for weather and climate prediction.
To remain agnostic to future programming paradigms a separation of concerns approach, along with a novel automatic code generation (PSyclone) of the parallel layer, is used. This allows script based automatic optimisation of the entire model code and the ability to adapt the model to different parallelisation strategies
This talk details the formulation and design used in Gungho and demonstrates its performance across a variety of benchmarks
Speaker: Thomas Melvin (Met Office)
-
11:30
→
12:30
Lunch break 1h
-
12:30
→
14:00
Moderator: Christian Kühnlein (ECMWF)
-
12:30
Dynamics developments within the ALADIN NWP consortium 45m
We give an overview of the dynamics developments within the ALADIN NWP consortium that develops the ALADIN System. This system is used by the ALADIN partners to run limited-area model (LAM) configurations for their national needs. The applications run at resolutions of about 1 km. The code of the ALADIN System is shared with the global ARPEGE model of Météo-France and the IFS of the ECMWF.
ALADIN can use a hydrostatic and a non-hydrostatic dynamical core. It is a spectral semi-implicit semi-Lagrangian model and shares the hydrostatic dynamics with the global model. We will explain how the LAM spectral methods are implemented to keep the model consistent with the global models. We will give an overview of its specificities, such as the horizontal diffusion, the vertical discretizations and the physics-dynamics coupling.
Alternatives to the spectral approach are currently being investigated. This is done in a twofold approach. First, the Atlas data structure framework of the ECMWF is being adapted to accommodate LAM geometries. This will facilitate the creation of an Atlas-based LAM model. Secondly, a gridpoint solver is being implemented within the codes of the existing ALADIN-NH dynamical core as an alternative to the spectral one.
Speaker: Piet Termonia (Royal Meteorological Institute) -
13:15
Towards a more stable formulation for the Non-hydrostatic ”constant-coefficient semi-implicit” dynamical core of AROME system 45m
Despite its proven robustness and accuracy for NWP purposes at kilometric scales, the so-called “constant-coefficient” linear approach for solving the system of Euler equations with a semi-implicit scheme may suffer from two serious weaknesses when severe high-resolution flows and steep orography are at stake. Actually, the inconsistency in the boundary treatment between the horizontally homogeneous implicit linear part and the non-linear explicit part of the model, and the absence of implicitly treated orographic metrics terms arising from the use of terrain-following coordinate, may jeopardise the stability of the system at high-resolution. A bespoke solution circumventing these issues without relaxing the constant-coefficient assumption is presented. This solution results in a new formulation of the vertical momentum prognostic equation, that leads to a substantial gain in term of stability for the resulting constant-coefficient semi-implicit dynamical core of AROME model.
Speaker: Fabrice Voitus (Météo-France)
-
12:30
-
14:00
→
14:30
Coffee break 30m
-
14:30
→
16:00
Moderator: Kristian Mogensen (ECMWF)
-
14:30
The Ocean Model ICON-O as part of the Earth System Model ICON-ESM 45m
In the area of ocean and climate modelling the anticipated end of Moore’s law has motivated efforts to formulate ocean and atmosphere models on unstructured grids. The ocean-sea-ice model ICON-O and the Earth System Model ICON-ESM of the Max-Planck Institute of Meteorology are particular examples of a new generation of ocean and earth system models. This talk provides a look behind the scenes of ICON-O’s structure-preserving design. We present ocean-stand alone as well as coupled simulations that demonstrate the capabilities of the model. We outline also the emerging scientific opportunities in ocean and climate modelling that ICON-O and ICON-ESM allow to explore and report on our first steps on these new research avenues.
Speaker: Peter Korn (MPI) -
15:15
Motivations, successes, and problems, with the Lagrangian-remap dynamical core in MOM6 45m
The sixth version of the Modular Ocean Model, MOM6, uses the vertical Lagrangian-remap method as the basis of its dynamical core. One of the driving motivations for this approach was to permit generalized vertical coordinates and reduce the numerical mixing associated with the use of Eulerian vertical coordinates. The inspiration for the approach can be traced back to the isopycnal layer models that have their own limitations but do have excellent preservation of water masses and can represent dense overflows with high fidelity. In several studies, many idealized, the role of vertical coordinates in controlling spurious heat uptake has been established with confidence, and the utility of both generalized vertical coordinates and the Lagrangian-remap method demonstrated to be advantageous in reducing numerical mixing. The spurious mixing problem was known to be most dire for eddy-permitting models in which high grid-scale energy exacerbates the numerical mixing. Despite this, the ¼ degree OM4 model (using MOM6) has a well controlled global heat uptake by virtue of using hybrid vertical coordinates, and we consider this a success of the Lagrangian-remap and general coordinate approaches. However, one aspect of the solution is particularly unsatisfactory: the solution has a shallow Atlantic meridional overturning circulation which is reminiscent of the earlier Eulerian vertical coordinate models. The reasons for this bias are unclear and the result is at odds with the rationale behind developing MOM6. We will review the motivations behind adopting general vertical coordinates, illustrate the methods and algorithms used in MOM6 and discuss the outstanding conundrum of the shallow overturning circulation.
Speaker: Alistair Adcroft (NOAA-GFDL)
-
14:30
-
16:00
→
16:30
Breakout Groups: End time is flexible
-
16:00
Breakout Group 1 - hosted by Nils Wedi, Thomas Melvin 30m
-
16:00
Breakout Group 2 - hosted by Fabrice Voitus, Piet Termonia 30m
-
16:00
Breakout Group 3 - hosted by Alistair Adcroft and Peter Korn 30m
-
16:00
-
09:45
→
11:30
-
-
09:00
→
10:30
Moderator: Michail Diamantakis (ECMWF)
-
09:00
Exploring a representation of model uncertainty in the IFS due to the transport scheme 45m
ECMWF ensemble forecasts include stochastic perturbations, which are designed to represent model uncertainties. Their inclusion significantly increases the skill of the ensemble forecast at all forecast ranges. Currently, the stochastic perturbations used operationally in the IFS represent model uncertainties that are attributed to the parametrisations of atmospheric physics processes. Recent efforts have focussed on exploring how to introduce stochastic perturbations to skillfully represent model uncertainties that arise within the dynamical core.
Following work to improve the convergence of the IFS transport scheme, a new stochastic perturbation scheme has been developed. The “STOCHDP” scheme introduces stochastic perturbations to the calculation of the departure point (DP) in the semi-Lagrangian method, motivated by Diamantakis & Magnusson (2016, MWR), who analysed the impact that the complexity of the flow field has on the rate of convergence of the iterative DP calculation.
This talk will provide an overview of stochastic model uncertainty representation in the IFS and an outline of the design and sensitivities of the new STOCHDP scheme.
Speaker: Sarah-Jane Lock (ECMWF) -
09:45
Compatible finite elements for numerical weather prediction 45m
Compatible finite element methods have two aspects that make them exciting for use in dynamical cores. The first is that they extend the linear wave propagation of C grid methods (steady geostrophic modes, lack of spurious pressure modes, absence of spurious inertial oscillation modes) to non-affine grids such as the cubed sphere, whilst allowing flexibility to alter pressure/velocity degree-of-freedom ratios (to eradicate spurious inertia-gravity wave oscillations) and increase the order of consistency on arbitrary grids. The second is that they have a natural framework for incorporating conservation of energy and enstrophy (or appropriate dissipation of these at the small scale, with correct energy transfers between kinetic and potential energy). I will survey both of these, and report recent work on bringing these methods closer to operational use, considering efficient implicit solvers, upwinding methods and coupling with moisture parameterisations.
Speaker: Colin Cotter (Imperial College)
-
09:00
-
10:30
→
11:00
Coffee break 30m
-
11:00
→
12:30
Moderator: Inna Polichtchouk (ECMWF)
-
11:00
Towards global weather forecasting with IFS-FVM 45m
Over the next decade, various aspects of ECMWF’s Integrated Forecasting System (IFS) need to be adapted for higher-resolution global forecasts called for by ECMWF’s long-term strategy. The current IFS dynamical core relies on the spectral-transform method to solve the hydrostatic primitive equations with a semi-implicit semi-Lagrangian integration scheme. ECMWF is continuing to develop this dynamical core, which also includes a nonhydrostatic option, to make it as computationally efficient as possible. Additionally, ECMWF develops alternative nonhydrostatic dynamical core formulations with complementary numerical and computational properties for high-resolution applications. The IFS-FVM employs finite-volume semi-implicit integration of the deep-atmosphere fully compressible equations. Various options of flux-form non-oscillatory Eulerian advection are combined with a 3D implicit time stepping of rhs dynamics forcings, all embedded in a horizontally- unstructured vertically-structured co-located discretisation framework. We present the ongoing IFS- FVM development towards realistic weather configurations at ECMWF. We further highlight some of the recent numerical and computational enhancements.
Speaker: Christian Kühnlein (ECMWF) -
11:45
The DYNAMICO dynamical core : theory, implementation and outlook 45mSpeaker: Thomas Dubos (LMD/IPSL, Ecole Polytechnique)
-
11:00
-
12:30
→
13:30
Lunch break 1h
-
13:30
→
14:30
Posters
-
14:30
→
16:00
Moderator: Michail Diamantakis (ECMWF)
-
14:30
On the improvement of operational downscaling in spectral wave models 45m
In the past significant progress has been made in terms of downscaling our wave forecasts towards our coasts. We present in this talk the work of the last decade in terms of operational downscaling using unstructured grids and implicit numerical schemes on triangular unstructured grids. Special focus will be numerical limiters and the error order of the discretization itself. We will look at various cases ranging from large ocean basins e.g. U.S. east Coast and very high resolution coastal application for the sake of the validation of the physics and numerics used in the models. The given talk is not limited to a certain model, the presented schemes have been already implemented in the common forecasting systems like WAM or WW3 and are actually in testing and further development.
Speaker: Aron Roland (BGS IT&E) -
15:15
Sea ice rheology and numerical solvers for sea ice dynamics 45m
Numerical solvers should be efficient, robust and scalable. Solving the sea ice momentum equation is recognized to be a difficult problem. This stems from the fact that sea ice rheology, i.e. the relation between applied stresses and the resulting deformations, is very nonlinear and leads to a stiff system of equations. From a physical point of view, this nonlinearity manifests itself by the formation of narrow features such as sea ice leads. In this talk, the viscous-plastic (VP) rheology and the recently developed elasto-brittle (EB) formulation will be presented. The various numerical methods for solving the sea ice momentum equation with a VP formulation will be discussed. A new implicit approach based on Anderson acceleration will be introduced with some preliminary results presented. We will also briefly examine numerical challenges with the EB rheology. Finally, we will discuss the potential evolution of sea ice modeling with a focus on sea ice dynamics.
Speaker: Jean-Francois Lemieux (ECCC)
-
14:30
-
16:00
→
16:30
Breakout Groups: End time is flexible
-
16:00
Breakout Group 1 - hosted by Jean-Francois LeMieux, Aron Roland, Thomas Dubos 30m
-
16:00
Breakout Group 2 - hosted by Christian Kühnlein, Colin Cotter, Sarah-Jane Lock 30m
-
16:00
-
09:00
→
10:30
-
-
09:00
→
09:45
Moderator: Kristian Mogensen (ECMWF)
-
09:00
Multiresolution ocean modelling 45m
Unstructured meshes introduce a multiresolution aspect in ocean modeling. Their resolution can be adjusted to practical needs according to the observed eddy activity, the Rossby radius of deformation, or the interest in small-scale dynamics in a particular area. Although the utility of unstructured meshes is well recognized in coastal ocean modeling, global ocean models used in climate studies are predominantly formulated on structured meshes. Several recent model development efforts such as FESOM at the Alfred Wegener Institute, ICON at the Max-Planck Institute for Meteorology (both Germany) and MPAS-Ocean at the Los Alamos National Laboratory (USA) propose fully functional ocean circulation models working on unstructured meshes. During long time such models were considered to be more computationally expensive than their structured-grid counterparts. The advances in computer technology and numerical methods lead to the fact that over recent years these models become similarly computationally efficient, showing a comparable throughput and offering an excellent scalability on massively parallel machines. The three models above are also used as ocean components of respective climate models.
We discuss the discretizations used, with a particular focus on numerical modes of these discretizations and measures needed to suppress them. Based on FESOM,we illustrate the througput and scalability on different meshes, and give practical examples from global simulations with a regional focus showing the benefits of unstructured meshes. We conclude with the statement that ocean models formulated on unstructured meshes are mature enough to be used in climate studies.Speaker: Sergey Danilov (Alfred-Wegener-Institut)
-
09:00
-
09:45
→
10:30
Moderator: Christian Kühnlein (ECMWF)
-
09:45
Temperature discretizations for the IFS, horizontal to vertical resolution aspect ratio and their importance for accurate global weather predictions 45m
In a continuous effort to increase the horizontal resolution of NWP models, vertical resolution increases often lag behind. This results in an inconsistent horizontal-to-vertical resolution aspect ratio, which has detrimental consequences for ECMWF-IFS employing vertical finite element vertical discretization. In particular, global-mean temperature in the stratosphere unphysically cools when the horizontal resolution is increased without a concomitant increase in the vertical resolution. Such horizontal resolution sensitivity is undesirable for model development and/or for 4d-Var, where each minimization loop is performed at a different horizontal resolution. The unphysical cooling arises because at higher horizontal resolution smaller scale gravity waves are generated not only in the horizontal direction but also in the vertical. If the vertical resolution is not adequate, these gravity waves alias into a vertical grid-scale mode in the temperature field, leading to spurious thermal sources. Apart from an increase in the vertical resolution, alternative solutions to the unphysical global-mean cooling are presented.
Another source of discretization errors in the thermodynamic formulation of ECMWF-IFS is the use of temperature as a prognostic variable instead of potential temperature. In the potential temperature formulation, the thermodynamic variable is materially conserved whereas it is not in the temperature formulation. In the second part of this talk, a semi-implicit system using potential temperature as a prognostic variable is derived and the performance of ECMWF-IFS is compared to the temperature formulation. It is shown that the two formulations behave comparably under several scenarios.
Speaker: Inna Polichtchouk (ECMWF)
-
09:45
-
10:30
→
12:00
Posters
-
12:00
→
12:45
Moderator: Christian Kühnlein (ECMWF)
-
12:00
The ICON model: actual state and first steps towards a new dynamical core based on the Discontinuous Galerkin method 45m
The ICON model is in operational use at DWD since 2015 for global forecasts and since 2016 for a nest over Europe.
Actually, it runs additionally in a convection-permitting setup (called ICON-D2) in a parallel routine to replace COSMO-D2 probably in Q4/2020. ICON uses horizontally a triangle grid, extracted by multiple subdivisions of icosahedral triangles. The spatial discretization is a mixed finite-volume, finite-difference approach. Newer developments in the dynamical core are a deep atmosphere variant.During the next years, it is planned to develop an alternative dynamical core based on the Discontinuous Galerkin (DG) approach. The DG method allows the conservation of each prognostic variable and to achieve a higher order approximation. Beyond this, it promises to run faster on massively parallel architectures due to a more compact data transfer and due to a higher computational intensity.On the other hand, relatively small time steps might be a potential obstacle for its operational use.
The current work towards this goal consists in the development of a 2D toy model. The solution of the shallow-water equations on the sphere is done via local coordinates on each triangle and a transformation of fluxes between their edges. To get rid of too small time-steps for flat grid cells, a horizontally explicit-vertically implicit (HEVI) approach for the solution of the Euler equations in a 2D slice model is chosen.
Speaker: Michael Baldauf (DWD)
-
12:00
-
12:45
→
13:45
Lunch break 1h
-
13:45
→
16:30
Moderator: Michail Diamantakis (ECMWF)
-
13:45
Mixed-Precision Arithmetic in Earth-System Modelling 45m
Earth-System models traditionally use double-precision, 64 bit floating-point numbers to perform arithmetic. According to orthodoxy, we must use such a relatively high level of precision in order to minimise the potential impact of rounding errors on the physical fidelity of the model. However, given the inherently imperfect formulation of our models, and the computational benefits of lower precision arithmetic, we must question this orthodoxy. At ECMWF, a single-precision, 32 bit variant of the atmospheric model IFS has been undergoing rigorous testing in preparation for operations for around 5 years. The single-precision simulations have been found to have effectively the same forecast skill as the double-precision simulations while finishing in 40% less time, thanks to the memory and cache benefits of single-precision numbers. Following these positive results, other modelling groups are now also considering single-precision as a way to accelerate their simulations.
In this talk I will present the rationale behind the move to lower-precision floating-point arithmetic and up-to-date results from the single-precision atmosphere at ECMWF. I will also present the first results from running ECMWF's chosen ocean model, the community model NEMO, with single-precision. Finally I will discuss the feasibility of even lower levels of precision, like half-precision, which are now becoming available through GPU-based and ARM architectures.
Speaker: Sam Hatfield (ECMWF) -
14:30
The GFDL Finite-Volume Cubed-Sphere Dynamical Core: Design and Prospects for Global and Unified Modeling 45m
The GFDL Finite-Volume Cubed-Sphere Dynamical Core, or FV3, is designed to be an accurate, efficient, and adaptable dynamical core useful for a variety of weather and climate applications. In this talk, I discuss how FV3 is designed, implemented, and used. I recapitulate the history of FV3's development from S-J Lin's original finite-volume advection scheme, to the vertically-Lagrangian FV dynamical core, and to the present-day all-scale nonhydrostatic FV3 core. The motivations and philosophy behind the discretizations of FV3 are discussed. I conclude with a discussion of FV3's moist thermodynamics, integrated physics, and prospects for unified modeling with FV3-based models.
Speaker: Lucas Harris (NOAA-GFDL) -
15:15
Coffee break 30m
-
15:45
Global Nonhydrostatic Atmospheric Modeling using Spherical Centroidal Voronoi Meshes 45m
During the last two decades global nonydrostatic atmospheric models have been developed that use alternative spherical tilings to latitude-longitude grids. The Model for Prediction Across Scales (MPAS) is one such model, and it uses a spherical centroidal Voronoi tessellation (SCVT) for its horizontal mesh. A C-grid staggering of the prognostic variables is used within MPAS so as to maximize the resolution capabilities in the nonhydrostatic solver for the strongly divergent motions characterizing convection. We will discuss the advances that were necessary to enable the C-grid staggering on the SCVT within a finite-volume formulation for the atmospheric solver, and we will also consider some of the advantages and disadvantages of employing this approach in a modeling system. Forecast examples and idealized test case results illustrate the effectiveness of the model for convection-permitting resolution. We will conclude with some thoughts on what constitute good test cases for global nonhydrostatic atmospheric solvers.
Speaker: Bill Skamarock (NCAR)
-
13:45
-
16:30
→
17:00
Breakout Groups: End time is flexible
-
16:30
Breakout group 1 - hosted by Lucas Harris, Michael Baldauf, Sergey Danilov 30m
-
16:30
Breakout group 2 - hosted by Inna Polichtchouk, Sam Hatfield, Bill Skamarock 30m
-
16:30
-
09:00
→
09:45
-
-
10:30
→
12:15
Moderator: Michail Diamantakis (ECMWF)
-
10:30
Transport schemes for weather and climate models 45m
In this talk "transport" means advection. A google scholar search for "numerical advection schemes" gives 140,000 results. And those will mostly be in the atmosphere and ocean modelling community because advection is called convection in mathematics and engineering. Why so many? It is probably because none of them work very well. There is a lot that we demand from our advection schemes:
1. Accuracy (high order and good accuracy even at coarse resolution).
2. Computational efficiency.
3. Monotonicity - spurious oscillations or unbounded values should not be generated.
4. Conservation - the advected quantity should not be artificially created or destroyed.
5. Stable for long time steps (often the time step is limited by high wind speeds or small grid boxes).
6. Multi-tracer efficiency - it would be nice if advecting 100 tracers were less expensive than 100 times advecting one tracer).
7. Maintaining correlation between tracers - imagine if you were advecting temperature and water vapour. If the warm air were advected too quickly then you could end up with high water content and cold air that could lead to spurious precipitation.
There are many classes of advection scheme and some of these overlap. For example Eulerian, semi-Lagrangian, method of lines, finite volume, finite element. And there are different ways to achieve monotonicity and long time steps. This talk will describe some of these approaches and their advantages and disadvantages. Implicit time stepping for advection has been largely ignored in atmospheric modelling due to its high cost and large errors for long time steps. It may be time to reconsider these assumptions in order to achieve long stable time steps and conservation.Speaker: Hilary Weller (University of Reading) -
11:15
Coffee break 15m
-
11:30
Physics-dynamics aspects of the AROME model and its coupling with NEMO and WW3 ocean/wave model 45m
AROME is a Limited Area Model (LAM) designed for mesoscale "Convection Permitting" applications.
It is the result of the coupling between the ALADIN dynamical core and the MésoNH Physics.As the IFS, the Aladin DynCore is semi-Lagrangian with a spectral semi-implicit solver and hybrid hydrostatic pressure levels (more than 50% of the code of the dynamics is shared with the IFS). The non-hydrostatic (NH) option has been developed by the ALADIN consortium and is now shared with the IFS.
The Arome dynamics and data assimilation benefits from IFS developments (and the other way around).The mesoscale convection permitting physics of AROME is shared with the Research community model MésoNH: TKE scheme, shallow convection, 1 moment Bulk microphysics, radiation.
The coupling with the surface is driven by the SURFEX platform and the description of the land covers comes from ECOCLIMAP. Four different types of surfaces are treated separately by SURFEX: land, town, lake and sea. On sea tiles, SURFEX allows the coupling with a 1D ocean mixed layer (OML) scheme or a 3D Ocean Models and a wave model.
The Arome physics and surface interactions benefit from scientific and technical developments of a large University and Research community (and the other way around).AROME-Western Europe is operational from 2008. It currently runs at a horizontal resolution of 1.3km and 90 levels in the vertical. The initial condition of AROME-Western Europe comes from a 3DVAR data assimilation cycle and the lateral boundary conditions come from ARPEGE global forecasts. An ensemble with 16 members at a resolution of 2.5km documents the predictability of the mesoscale forecast over the France domain .
AROME is also running operationnaly over 5 overseas domains at a resolution of 2.5 km. The AROME-overseas run as a downscaling at a convection permitting resolution of the HRES IFS. As all oversea domains may be hit by tropical cyclones, the coupling with a prognostic 1D OML is activated. The initial conditions of the OML come from MECRATOR-OCEAN products. So far, the downscaling approach gives better results than the 3DVAR data assimilation in tropical regions covered with little observations. The development of an ensemble for AROME-Overseas is work in progress.
In order to improve the forecast of tropical cyclones but also of medicanes and polar lows, the AROME community is working on the ocean-atmosphere interactions.
- improve surface flux parametrisation for strong winds: coupling with a wave model, sea spays.
- improve of the feedbacks between the atmosphere and the ocean: coupling with the 3D ocean model NEMO.A research effort is made in parallel to improve the representation of clouds.
- One moment versus 2 moments microphysics
- Aerosol/microphysics interactions (also involves coupling with a wave model)Speaker: Sylvie Malardel (Météo-France/LAcy)
-
10:30
-
12:15
→
13:30
Lunch break 1h 15m
-
13:30
→
16:00
Moderator: Andreas Müller (ECMWF)
-
13:30
Seamless integration of hydrostatic, soundproof and fully compressible equations 45m
When written in conservation form for mass, momentum, and density-weighted potential temperature, and with Exner pressure in the momentum equation, the pseudoincompressible model and the hydrostatic model only differ from the full compressible equations by some additive terms. This structural proximity can be faithfully transferred to a numerical discretization providing seamless access to all three analytical models.
This lecture will discuss the theoretical background of the models from the perspective of asymptotic analysis and it will introduce the numerical scheme proposed recently by Benacchio and the author (MWR, 147, 2019).
The semi-implicit second-order scheme discretizes the rotating compressible equations by evolving full variables, and, optionally, with two auxiliary fields that facilitate the construction of an implicit pressure equation. Time steps are constrained by the advection speed only as a result. Borrowing ideas on forward-in-time differencing, the algorithm reframes the authors’ previously proposed schemes into a sequence of implicit midpoint step, advection step, and implicit trapezoidal step.
The proposed scheme generalizes the authors’ acoustics-balanced initialization strategy to also cover the hydrostatic case in the framework of an all-scale blended multimodel solver.
Speaker: Rupert Klein (Freie Universität Berlin) -
14:15
Modelling geophysical flows with nonoscillatory forward-in-time methods 45m
The advance of massively parallel computing in the nineteen nineties and beyond encouraged finer grid intervals in numerical weather-prediction models. This has improved resolution of weather systems and enhanced the accuracy of forecasts, while setting the trend for development of unified all-scale Earth-System models. This lecture illustrates this trend with a review of a versatile nonoscillatory forward-in-time (NFT) approach proven effective in simulations of a broad range of geophysical flows and, especially, in simulations of atmospheric flows from small-scale dynamics to global circulations and climate. The outlined approach exploits the synergy of the MPDATA methods for the simulation of fluid flows based on the sign-preserving properties of upstream differencing and optional finite-difference or finite-volume discretizations of spatial differential operators comprising PDEs of geophysical fluid dynamics. The lecture consolidates the concepts leading to a family of generalized nonhydrostatic NFT flow solvers that include soundproof PDEs of incompressible Boussinesq, anelastic and pseudo-incompressible systems, common in large-eddy simulation of small-and meso-scale dynamics, as well as all-scale compressible Euler equations. Such a framework naturally extends predictive skills of large-eddy simulation to the global atmosphere and oceans, providing a bottom-up alternative to the reverse approach pursued in the weather-prediction models. Theoretical considerations are substantiated by calculations attesting to the versatility and efficacy of the NFT approach. Some prospective developments are also discussed.
Speaker: Dr Piotr Smolarkiewicz (National Center for Atmospheric Research) -
15:00
Coffee break 15m
-
15:15
Some Interesting Results with Element-based Galerkin Nonhydrostatic Atmospheric Models 45m
In this talk, I will present some recent results that we have obtained with our element-based Galerkin nonhydrostatic atmospheric models. The NUMA model is the computational engine inside of the U.S. Navy’s NEPTUNE weather forecast system while CLIMA-atmos is the dynamical core inside of the ClimateMachine climate modeling system. Since the last time I presented at ECMWF (back in 2018) we have made progress in both of these systems in terms of: (1) numerics, (2) process studies, and (3) computational performance. I will discuss each of these 3 topics which includes advances in our time integration strategies, analysis of the effective resolution of local high-order methods (including numerous turbulence closures) and high-altitude simulations, and scaling on GPU computers.
Speaker: Francis Giraldo (NPS)
-
13:30
-
16:00
→
16:30
Breakout Groups: End time is flexible
-
16:00
Breakout Group 1 - hosted by Piotr Smolarkiewicz, Rupert Klein 30m
-
16:00
Breakout Group 2 - hosted by Francis Giraldo, Hilary Weller, Sylvie Malardel 30m
-
16:00
-
10:30
→
12:15
-
-
09:00
→
10:30
Moderator: Inna Polichtchouk (ECMWF)
-
09:00
Development of the spectral-based dynamical core of the JMA operational global model 45m
Since 1988, Japan Meteorological Agency (JMA) has been operating a global spectral hydrostatic primitive equations model named Global Spectral Model (GSM). In its first implementation, GSM employed spherical harmonics transform on quadratic Gaussian grid, an Eulerian advection scheme, and a semi-implicit time integration method. These methods were chosen to take advantages of the spectral method such as accurate calculation of horizontal gradients and ease of solving a Helmholtz equation in spectral space.
Keeping in pace with the increasing computing power, GSM has steadily increased its resolution, but this was not achieved with increase in computing capacity alone. A number of improvements to numerical methods have been incorporated, keeping the advantages of the spectral method, to harness the evolving computer architecture. With successive introduction of a semi-Lagrangian advection scheme on linear grid, reduced spectral transform and speedup of spectral transforms on massively parallel computers, GSM has evolved to accommodate horizontal resolution as high as 20 km (in operation) and 13 km (planned in the next few years).
The increase of horizontal resolution on the linear grid has led us to revisit the issue of spectral blocking which has been recognized early in the history of atmospheric modelling. To resolve this issue, we developed a new discretization of the pressure gradient terms which ensures their rotation-free by exploiting characteristics of the spectral discretization (Ujiie and Hotta 2019). Our recent experience also highlights the importance of numerics, particularly mimetic discretization, at high-resolution modelling.
Preparing for further resolution increase and possibility of non-hydrostatic global modelling, research is ongoing on a new nestable grid system that allows for both accurate spectral transforms and multi-grid methods (Hotta and Ujiie 2018), which will also foster gradual transition toward spectral-grid hybrid modelling.
The talk will present current status and future outlook of the global spectral model at JMA.Speaker: Masashi Ujiie (JMA) -
09:45
Development of a scalable high-order conservative nonhydrostatic model by using multi-moment finite volume method 45m
In this lecture we will present a novel scalable high-order conservative nonhydrostatic multi-moment finite volume dynamical core. The new dynamical core is based on so-called multi-moment constrained finite volume (MCV) method, which is well-balanced among solution quality (accuracy and robustness), algorithmic simplicity, computational efficiency and flexibility for model configuration. Rigorous numerical conservation is guaranteed by a constraint of finite volume formulation in flux form. The resulting MCV models have been verified with widely used benchmark tests. The numerical results show that the present MCV models have solution quality competitive to other exiting high order models. Parallelization of the MCV shallow water model on cubed sphere grid reveals its suitability for large scale parallel processing with desirable scalability
Speaker: Xingliang Li (CMA)
-
09:00
-
10:30
→
10:45
Coffee break 15m
-
10:45
→
12:15
Moderator: Michail Diamantakis (ECMWF)
-
10:45
Next-Generation Time Integration for Weather and Climate Simulations 45m
Running simulations on high-performance computers faces new challenges due to e.g. the stagnating or even decreasing per-core speed. This poses new restrictions and therefore challenges on solving PDEs within a particular time frame. Here, disruptive mathematical reformulations which e.g. exploit additional degrees of parallelism also in the time dimension gained increasing interest over the last two decades.
This talk will cover two examples of the current cutting edge research on parallel-in-time integration methods in the context of weather and climate simulations:
* Parallel-in-time rational approximation of exponential integrators (REXI) based on Terry's (T-REXI) and Cauchy Contour (CI-REXI).
* Multi-level time integration of spectral deferred correction (ML-SDC) & Parallel Full Approximation Scheme in Space and Time (PFASST)
These methods are realized and studied with numerics similar to the ones used by the European Centre for Medium-Range Weather Forecasts (ECMWF). Our results motivate further investigation for operational weather/climate systems in order to cope with the hardware imposed restrictions of future super computer architectures.
(I gratefully acknowledge contributions and more from Francois Hamon, Terry S. Haut, Richard Loft, Michael L. Minion, Nathanaël Schaeffer)Speaker: Martin Schreiber (Technical University of Munich) -
11:30
Discontinuous Galerkin methods for Numerical Weather Prediction 45m
In this lecture discontinuous Galerkin (DG) methods for numerical weather prediction (NWP) will be described. Both strengths and limitations of such methods applied to atmospheric modeling will be discussed and possible ways to improve their efficiency reviewed.
A particular strategy based on combining a high order DG discretization with p-adaptivity techniques and efficient semi-Lagrangian and semi-implicit time integrators will be presented in the context of simplified NWP model equations. An outlook towards the parallel implementation of such numerical formulation in the case of a dynamical core prototype will also be discussed.Speaker: Giovanni Tumolo (ECMWF)
-
10:45
-
12:15
→
12:45
Breakout Groups: End time is flexible
-
12:15
Breakout Group 1 - hosted by Masashi Ujiie, Xingliang Li 30m
-
12:15
Breakout Group 2 - hosted by Giovanni Tumolo, Martin Schreiber 30m
-
12:15
-
09:00
→
10:30