Global-Regional Integrated Forecast System (GRIST)

Online Tutorial

• Model Version: 1.2
• Latest Code Version: A25.7.1, Fortran

---

✨ Features

• Global-Regional Integrated Dynamical Framework
• Two Integrated Physics Suites: PhysW / AMIPW_Physics; PhysC / AMIPC_Physics
• Simplified and Idealized Models: SCM, SWM, 3D Dry/Moist Cores
• Extensive Testing across multiple configurations:
• Custom Diagnostic Templates for various applications (see the GRIST tutorial for details).

---

General Usage

0. A Tutorial on Running GRIST on a Laptop from scratch

   • https://github.com/GRIST-Dev/GRIST-RunOnUrLapTop

1. Modify Environment Setup

   • Edit build.sh in the "build" directory on your local machine.
   • Choose a model type (e.g., amipw or amipc).
   • Run:

     ./build.sh amipw  # or ./build.sh amipc

2. Compilation

   • After successful compilation, the following files are generated:
     • ParGRIST_amipw.exe / ParGRIST_amipc.exe (Model executable)
     • partition.exe
   • Place the model executable in the "run" directory for each case.

3. Running the Model

   • Prepare namelist files and necessary input data.
   • Ensure the required runtime environment is set up.
   • Submit the job for execution.

4. Quick start

   • To quickly familiarize one with running the model, it is highly recommended to utilize the assets provided in this paper: https://gmd.copernicus.org/articles/17/6301/2024/gmd-17-6301-2024.html. These assets include the frozen model code, compilation and runtime configurations, and the input data required to reproduce each case.

---

GRIST Kernel

The GRIST_kernel is the minimum codebase supporting all GRIST v1.0 functions, including results from published papers (bit regression is ensured for identical setups and environments).

• GRIST_kernel can be easily extended with GRIST_increm (add-on modules) with minimal effort.

Highlight

An AI-Enhanced 1km-Resolution Seamless Global Weather and Climate Model to Achieve Year-Scale Simulation Speed using 34 Million Cores. Proceedings of the 30th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programm (https://dl.acm.org/doi/10.1145/3710848.3710893):

1. Model Architecture

Figure 1. The architecture of our AI-enhanced GRIST model. programming model, which is further extended to the Sun- way architecture (section 3.3). A mixed-precision dynamical core (Fig. 3) has been developed to reduce the computa- tional load, which maintains the stability and accuracy of the original double-precision code based on a careful itera- tive development procedure and a hierarchy of tests. This mixed-precision dynamical core is detailed in section 3.4.

2. 1km Global Storm Resolving Modeling (GSRM)

Figure 2. Super Typhoon Doksuri and the “23.7” extreme rainfall event over North China. The left panel shows the mean rainfall rate (mm/day) during UTC00, 29th-UTC00,30th, July, 2023, for CMPA, G11L60 and G12L30, the right panel shows the cloud top temperature (unit: K).

3. Computational Performance on New Sunway with 34M cores

Figure 3 A summary of recent high-resolution weather and climate modeling efforts on supercomputers: a continuous journey towards affordable global storm-resolving modeling.

Conclusion

The 1 km GSRM Nonhydrostatic Atmosphere simulation requires solving 11 Prognostic equations for 3D Variables, while accounting for sound waves propagating at speeds of 340–350 m/s across the entire Earth Sphere (~500 million edges, at one layer), under realistic surface topography and physics-dynamics coupling. These factors collectively make it an extremely challenging computational task.

By leveraging AI-trained physics suites, a generalized programming model, and mixed-precision computing, global weather and climate simulations are making significant strides toward higher accuracy, efficiency, and integration. These advancements provide more powerful tools for extreme weather prediction and deeper insights into climate change. This work serves as an initial step toward further exploration and innovation in this promising direction.

---

📌 References

A comprehensive description of the development and evaluation of the dynamical core framework is given in Zhang et al. (2019), (2020) and Zhang et al. (2024). The two baseline physics suites, PhysW and PhysC, specifically tailored for GRIST, are described and evaluated based on single column modeling (Li et al. 2023).

The full-model studies involving GRIST-PhysW, GRIST-PhysC, and their AMIP simulations have been discussed in Zhang et al. (2021) and Li et al. (2022), focusing on model performance and climate simulation analysis. Fu et al. (2024) (2025) further contributed to the intercomparison of the long-term state of two model climates and the simulation of the Madden-Julian Oscillation (MJO).

The model has found applications across several domains, including global variable-resolution modeling, global storm-resolving modeling, and climate change modeling (e.g., Zhou et al. 2020; Zhang et al. 2022; Sun et al. 2024). A Chinese-language brief introduction to the model framework was provided by Wang et al. (2024). Some background of the model development research can be found in Yu et al. (2019) and Zhang et al. (2023). More published references can be found at Online Tutorial.

---