DCMIP-2025:
Synopsis 

The Dynamical Core Model Intercomparison Project (DCMIP-2025) and Summer School will highlight the newest modeling techniques for global climate and weather models. It will be held at the National Center for Atmospheric Research (NCAR) in Boulder, CO, from June/2-6/2025 and will emphasize high-resolution non-hydrostatic modeling approaches and Machine Learning (ML) emulators as overarching themes. The objectives of the DCMIP-2025 Summer School are (1) to teach a group of about 30 multi-disciplinary students and postdocs how today’s and future atmospheric models and their dynamical core are or need to be built, (2) to shed light on the skill and realism of machine learning emulators for atmospheric fluid flows, and (3) to use idealized test cases to expose selected model design choices in simplified modeling frameworks. A particular focus of the idealized tests will explore the impact of topography on the circulation.
The summer school includes mornings lectures and hands-on modeling activities in the afternoon. The focus will lie on the three non-hydrostatic dynamical cores (Spectral Element, MPAS, FV3) that are available via NCAR’s Community Earth System Model (CESM) with its Community Atmosphere Model (CAM). The exploration of the ML emulators for General Circulation Models includes Google’s GraphCast and NVIDIA’s FourCastNet. DCMIP-2025 thereby continues the DCMIP-2008, DCMIP-2012 and DCMIP-2016 model intercomparison and summer school series. The DCMIP-2025 event is supported by the National Science Foundation (NSF), NOAA, and NCAR. The available funding will support the travel, lodging, food, and local transportation expenses of the selected participants.

This hands-on summer school is based on morning lectures and afternoon modeling projects with the CESM dynamical cores as well as selected ML-based GCM emulators like GraphCast and FourCastNet. We will form 8 small groups that will be guided by modeling mentors. In particular, we will form two groups each for the CAM dynamical cores. The latter are the Spectral Element (SE), FV3 (Finite Volume model on the cubed-sphere), and Model for Prediction Across Scales (MPAS) dynamical cores. In addition, we will form two groups that focus on ML emulators such as Google's GraphCast and NVIDIA's FourCastNet. All participants will utilize a shared Wiki online workspace (this Google site)
https://sites.google.com/umich.edu/dcmip-2025

to share information and discuss the modeling results before, during, and after the event and create a collaborative spirit.

The DCMIP-2025 draft schedule is (subject to change, all times are for the Mountain Time (MT) zone):

Sunday June/1/2025:
Travel day, all participants travel to Boulder, Colorado and will check-in at the hotel
(Embassy Suites, 2601 Canyon Blvd, Boulder, CO 80302) 

Monday June/2/2025

8am bus pickup from hotel
8:30am - 9:00 Welcome and logistics

9:00am - 10:00am Lecture 1: Overview of General Circulation Modeling, Scales, and Dynamical Cores

10:00am-10:30am coffee break

10:30am - 11:30am Lecture 2: Overview of Machine Learning Approaches

11:30am-12:30pm Lecture 3: Introduction to the Hands-on component of DCMIP and test cases

12:30pm-1:45pm lunch break

1:45pm-3:15pm hands-on session 1: Learn about the CAM workflow and the CESM simpler models framework

3:15pm-3:45pm coffee break

3:45pm-5:15pm hands-on session 2: Run a DCMIP test case

5:30pm bus pickup at the Mesa Lab

Tuesday June/3/2025

8am bus pickup from hotel
8:30am - 9:00 Short course: Visualization, Python

9:00am - 10:00am Lecture 1: Design of Dynamical Cores: Equation Sets and Computational Grids

10:00am-10:30am coffee break

10:30am - 11:30am Lecture 2: Design of Dynamical Cores: Horizontal Discretizations

11:30am-12:30pm Lecture 3: Design of ML emulators for GCMs

12:30pm-1:45pm lunch break

1:45pm-3:15pm hands-on session 1

3:15pm-3:45pm coffee break

3:45pm-5:15pm hands-on session 2

5:30pm bus pickup at the Mesa Lab

Wednesday June/4/2025

8am bus pickup from hotel
8:30am - 9:00 Short course: Programming paradigms, parallel computing concepts and performance

9:00am - 10:00am Lecture 1: Design of Dynamical Cores: Vertical discretizations

10:00am-10:30am coffee break

10:30am - 11:30am Lecture 2: Design of Dynamical Cores: Dissipation properties

Dissipation properties

11:30am-12:30pm Lecture 3: Use of ML in Weather and Climate Modeling, scientific performance / forecast accuracy / testbeds

12:30pm-1:45pm lunch break

1:45pm-3:15pm hands-on session 1

3:15pm-3:45pm coffee break

3:45pm-5:15pm hands-on session 2

5:30pm bus pickup at the Mesa Lab

Thursday June/5/2025

8am bus pickup from hotel
8:30am - 9:00 Short course: Today’s and future HPC architectures

9:00am - 10:00am Lecture 1: Design of Dynamical Cores: Temporal discretizations

10:00am-10:30am coffee break

10:30am - 11:30am Lecture 2: Specifics on the design of FV3, MPAS, SE (?)

11:30am-12:30pm Lecture 3: Design of coupled weather and climate models 

12:30pm-1:45pm lunch break

1:45pm-3:15pm hands-on session 1

3:15pm-3:45pm coffee break

3:45pm-5:15pm hands-on session 2

5:30pm bus pickup at the Mesa Lab

Friday June/6/2025

8am bus pickup from hotel
8:30am - 9:00 Short course: Parallel performance of weather and climate models

9:00am - 10:00am Lecture 1: Overview of Physical Parameterizations and ‘Simple Physics’ ideas  

10:00am-10:30am coffee break

10:30am - 11:30am Lecture 2: Tuning of climate models

11:30am-12:30pm Lecture 3: Prospects & frontiers: High-resolution weather and climate modeling, ML trends

12:30pm-1:30pm lunch break

1:30m-3:00pm Finalize presentations, optional: short after-lunch hike behind the Mesa Lab

3:00pm-3:30pm coffee break

3:30pm-5:30pm Presentations Mesa Lab Main Seminar Room, about 15 minutes per group

5:45pm-8pm Farewell BBQ, Tree Plaza

8pm bus pickup at the Mesa Lab

9pm optional: meet downtown at a bar or brewery, e.g. Avanti (Food and Beverage Court)

Saturday June/7/2025

Travel day, participants check out of the hotel and travel back home

DCMIP-2025 will enable the participants to describe and judge selected design choices for the dynamical cores of General Circulation Models (GCMs). The participants will learn how to configure, modify, run, and analyze simple configurations of the NCAR Community Earth System Model (CESM) with a particular focus on its Community Atmosphere Model (CAM). This will be accomplished via the use of idealized test cases. Furthermore, basic Machine Learning (ML) ideas will be conveyed to enable simple benchmarks of selected ML emulators for GCMs.

DCMIP-2025 will shed light on the

• differences between hydrostatic/nonhydrostatic and shallow-atmosphere/deep-atmosphere equation sets, and the selection criteria for the dry equation sets

• advective and flux-form of the transport (advection) equation and desirable physical properties of numerical schemes for advection: monotonicity and positive-definiteness

• various horizontal grids such as latitude-longitude, cubed-sphere, icosahedral, hexagonal, variable-resolution grids

• horizontal and vertical grid staggering options like the so-called A, B, C, D, E, and Z grids, as well as the Lorenz and Charney-Philips grids

• choice of the vertical coordinate in GCMs and the treatment of topography

• general characteristics of numerical schemes for the spatial discretization: spectral transform, finite element, finite volume and finite difference schemes

• general characteristics of numerical schemes for the temporal discretization: implicit versus explicit, semi-Lagrangian, one-step methods, multi-step methods

• accuracy and stability of numerical methods

• typical diffusion mechanisms in GCMs

• idealized test cases for the dynamical cores of GCMs

• Machine Learning concepts and ML emulators for GCMs