# GFDL Events & Seminars

### Upcoming GFDL events & seminars

**July 27, 2016: Global/Regional Non-hydrostatic Numerical Weather Prediction Model Using Semi-implicit and Semi-Lagrangian Method: progress and challenges**

Xueshun Shen (Center for Numerical Weather Prediction, China Meteorological Administration)

Since 2001, the China Meteorological Administration (CMA) began to develop the new generation NWP system. This system consists of a unified global/regional numerical model and a variational data assimilation system, which is called as the GRAPES (Global and Regional Assimilation & Prediction System). The governing equations of GRAPES are the fully compressible non-hydrostatic ones with shallow atmosphere approximation, which makes the model suitable for use at very high resolutions (such as grid size ). The GRAPES model was developed to be used for all production models in NWP, i.e., so called unified NWP model. The prognostic variables include 3-dimensional wind components, potential air temperature, non-dimensional pressure (The Exner function) derived from mass conservation and mixing ratio of water species. To solve the equations numerically, the spherical coordinate in the horizontal and terrain-following height coordinate in the vertical are utilized. To optimize balance in the model and to eliminate computational modes, C-grid staggering in the horizontal and Charney-Phillips staggering in the vertical were chosen. It should be noted that Charney-Phillips staggering presents a problem when used in the boundary layer mixing as care is needed to define the mixing coefficients appropriately as momentum and temperature are at different levels. For the time stepping scheme, the off-centered two-time-level SISL discretization is applied to each of the prognostic equations. The only distinction between scalar and vector quantities is that the departure and arrival point components and their directions are discretized together to include the change of vector direction from departure to arrival point with the help of a mid-point and a rotation matrix. The use of semi-implicit time scheme results in a 3-dimensional elliptic equation for solving the non-dimensional pressure. This is solved iteratively using the generalized conjugate residual technique (GCR) with appropriate preconditioning. For the spatial discretization, the equations of GRAPES model are evaluated on an Arakawa-C grid in the horizontal and a Charney-Phillips grid in the vertical as mentioned above. Standard second-order accurate finite differencing is applied to all terms. To apply conservation to tracers and non-negative quantities (monotonicity) under semi-Lagrangian advection, a conservative semi-Lagrangian scheme based on the piece-wise rational function is used (Shen, Wang and Xiao, 2011). However, the algorithm cannot be used to conserve the dry air mass. The solution procedure of GRAPES model is rather straightforward. For the dynamics part of GRAPES, 3-dimensional elliptic equation of non-dimensional pressure is solved at first. Then, the other prognostic variables are evaluated through substituting the solved non-dimensional pressure into the individual discretized equations. After finishing the integration of dynamics, the physics are computed by using the simple parallel-splitting method except for the microphysics. Before calling the microphysics, all the prognostic variables are updated to include the time changes of variables due to dynamics, radiation, land surface process, PBL and vertical diffusion as well as cumulus convection. The computation of microphysics is the final step of GRAPES model integration within one time-stepping loop. 2. Main applications of GRAPES model The successful development of GRAPES model and data assimilation system symbolizes the big change of CMA NWP system from technological dependence on the imported NWP into the self-development. Since 2006, a regional forecast system-GRAPES_Meso has been implemented for operation in National Meteorological Center (NMC) for meso-scale forecast over China, in Shanghai Typhoon Institute for typhoon track forecast, in Guangzhou Tropical Oceanographic and Meteorological Institute for monsoon and heavy rainfall forecast in succession. At present, the highest resolution of GRAPES_Meso is 3km. Since 2007, a pre-operational version of global forecast system-GRAPES_GFS (GRAPES global model: 0.5°x0.5°L36+ GRAPES global 3DVAR:1.125°x1.125°) has been established for global medium-range (0～10days) weather forecast. From June of 2016, GRAPES_GFS was updated with the horizontal resolution 0.25°x0.25°and 60 levels in the vertical, and became operation officially. And, a rapid update cycle system-GRAPES_RUC, based on GRAPES_Meso system, has been implemented for the very short range severe weather forecast (0～1day). 3. Future GRAPES model development Until present, applications of SISL scheme to the fully compressible equations have been proved the effectiveness. But, for the future high-resolution applications on massively parallel computers ( or more processors) as well as for the foreseeable seamless prediction from weather to climate, there exit great challenges for the future GRAPES model . As with many grid-point models using implicit time-stepping, the non-locality of 3-dimensional elliptic equation solver becomes a critical issue for the efficiency and scalability. Any non-local process will limit the speed of the model as will data transposes. The anisotropy of the latitude-longitude grid results in a large number of solver iterations, particularly at high resolutions. More uniform grids on the sphere and locally intensive algorithm are necessary. Traditional semi-Lagrangian schemes lose any conservation properties due to the use of point-wise interpolation. Although it is argued that mass conservation may not be so critical for weather forecast, but mass conservation should be revisited for the future earth system prediction, for example, the seamless weather-climate modeling system, chemical weather forecast, etc. In the current GRAPES model, conservation of moist species is guaranteed by using conservative SL scalar advection. Dry air mass and any other conservative properties will be the key issue for further development of GRAPES, either by developing the conservative Lagrangian algorithm or through revising the governing equations as well as the temporal-spatial discretization method. From 2012, we started to develop the new dynamical core based on the multi-moment constrained finite volume method. The basic consideration is to design a new core which has high accuracy, high scalability, better conservation properties and grid flexibility. Up to now, fundamental researches have been finished, including various tests in global shallow water and 2-D slice non-hydrostatic frameworks. Presentation on this new dynamics will be given during the visiting period.

Time: 12:00 pm - 1:30 pm

Location: Smagorinsky Seminar Room**August 2, 2016: Spencer Hill Final Public Oral**

Spencer Hill Final Public Oral

Spencer Hill Final Public Oral

Time: 10:00 am - 12:00 pm

Location: Smagorinsky Seminar Room**August 5, 2016: Summer Intern Presentations**

Summer Intern Presentations

Summer Intern Presentations Bridge: 877-918-1365/4351085

Time: 10:00 am - 2:00 pm

Location: Smagorinsky Seminar Room**August 12, 2016: Jeff Strong Final Oral Presentation**

Jeff Strong Final Oral Presentation

Jeff Strong Final Oral Presentation

Time: 10:00 am - 12:00 pm

Location: Smagorinsky Seminar Room**September 8, 2016: TBD**

Yoshi Wada (Columbia University)

TBD

Time: 2:00 pm - 3:30 pm

Location: Smagorinsky Seminar Room**September 21, 2016: Louise Nuijens of MIT**

Louise Nuijens of MIT

Louise Nuijens of MIT Wed lunch seminar. Title and abstract forwarded in Sept.

Time: 12:00 pm - 1:30 pm

Location: Smagorinsky Seminar Room**September 23, 2016: TBD**

Jaya Khanna FPO

TBD

Time: 10:00 am - 12:00 pm

Location: Smagorinsky Seminar Room**October 6, 2016: TBD**

Annalisa Bracco (Georgia Tech)

TBD

Time: 2:00 pm - 3:30 pm

Location: Smagorinsky Seminar Room**October 12, 2016: CPMIP: Measurements of Real Computational Performance of Earth System Models**

CPMIP: Measurements of Real Computational Performance of Earth System Models

Traditional metrics of computational efficiency such as performance counters and scaling curves do not tell us enough about real sustained performance from climate models on different machines. They also do not provide a satisfactory basis for comparative information across models. We introduce a set of metrics that can be used for the study of computational performance of climate (and Earth System) models. These measures do not require specialized software or specific hardware counters, and should be accessible to anyone. They are independent of platform, and underlying parallel programming models. We show how these metrics can be used to measure actually attained performance of Earth system models on different machines, and identify the most fruitful areas of research and development for performance engineering. We present results for these measures for a diverse suite of models from several modeling centres, and propose to use these measures as a basis for a CPMIP, a computational performance MIP.

Time: 12:00 pm - 1:30 pm

Location: Smagorinsky Seminar Room**October 13, 2016: TBD**

Ming Xue (University of Oklahoma)

TBD

Time: 2:00 pm - 3:30 pm

Location: Smagorinsky Seminar Room**October 17, 2016: TBD**

George Kiladis

TBD

Time: 2:00 pm - 3:00 pm

Location: Smagorinsky Seminar Room**October 20, 2016: TBD**

Sergey Kravtsov (University of Wisconsin-Milwaukee)

TBD

Time: 2:00 pm - 3:30 pm

Location: Smagorinsky Seminar Room**November 16, 2016: TBD**

Brandon Reichl (GFDL)

TBD

Time: 12:00 pm - 1:30 pm

Location: Smagorinsky Seminar Room