arXiv Science⌕ Search
Home/Archive/arXiv

arXiv · 2610.39334

SD3: The First Operationally-Proven, Fully Heterogeneous-Accelerated Global Kilometer-Scale Model with Inline Physics Architecture

Yuan Liang·Xiaomin Hu·Xi Chen·Jingsheng Zhu·Li Yang·Leisheng Li·Qian Song·Xu Wang·Hao Wang·Zhenzhen Wang·Huiyuan Li
View original source ↗Read source document ↗Export metadata ↓

Abstract

We present Super Dynamics on the Cube (SD3), the first operationally-proven, fully heterogeneous-accelerated global kilometer-scale model. Through a full-stack reconstruction in native CUDA C, SD3 introduces an "Inline Physics" architecture that deeply couples its non-hydrostatic cubed-sphere dynamical core with fast physical processes. This design maintains high physical consistency, ensures energy and mass conservation, and mitigates "Grey Zone" mismatches. Five years of continuous operational forecasting, together with validation against diverse observations, demonstrates the scientific fidelity of SD3. Recently, SD3 achieved an unprecedented 205.7 simulated days per day at global 1.5 km resolution using 24576 GPUs. SD3 establishes a hardware-agnostic, portable, and reproducible paradigm across diverse Exascale ecosystems. Beyond weather and climate research, SD3 offers a transformative paradigm for the HPC community and the next-generation Digital Twin Earth to establish a robust, high-performance blueprint for the future of Earth system modeling.

Explore related subjects

physics.ao-ph

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Yuan Liang, Xiaomin Hu, Xi Chen, Jingsheng Zhu, Li Yang, Leisheng Li, Qian Song, Xu Wang, Hao Wang, Zhenzhen Wang, Huiyuan Li. 2026-10-29. SD3: The First Operationally-Proven, Fully Heterogeneous-Accelerated Global Kilometer-Scale Model with Inline Physics Architecture. https://arxiv.org/abs/2610.39334

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

From Sea Surface Height to Velocity - A Machine Learning Approach

Inferring upper-ocean velocity from altimetric sea surface height (SSH) remains challenging below 50~km, scales now resolved by the Surface Water and Ocean Topography (SWOT) mission. Conventional methods rely on geostrophic or cyclogeostrophic balance. Here, we demonstrate that deep learning can predict the full velocity vector from SSH and environmental data and recover derivative-based diagnostics, including spectral energy flux. We trained a U-Net encoder--decoder on hourly fields from the pre-SWOT Cape Basin extraction of the LLC4320 ocean simulation, with 1.5-km grid spacing ($11.8$--$15.8^\circ$E, $41$--$45^\circ$S). Inputs were SSH, sea surface temperature (SST), and wind stress; targets were horizontal velocities at the surface and 50~m, where wind influence is weaker. At 50~m, the best model achieves a speed RMSE of $0.049\,\mathrm{m\,s^{-1}}$, 29% lower than the geostrophic approximation. Predicted relative vorticity has a correlation of 0.93, compared with 0.75 for the geostrophic approximation. The model also recovers divergence with a correlation of 0.66, whereas geostrophic velocity is nondivergent by definition. Tests of training strategies yield four findings: SSH gradients outperform raw SSH as inputs; wind stress is the most useful covariate for total velocity prediction; SST benefits vary seasonally; and including vorticity and divergence in the loss improves accuracy. These results offer a path for transferring high-fidelity simulation physics to SWOT applications, where kilometer-scale SSH snapshots contain superposed balanced and unbalanced signals.

physics.ao-ph↗
arXiv2026

Machine learning emulators of dynamic global vegetation models for fast gridded carbon budget predictions

The land carbon sink explains most of the variability of the atmospheric CO2 growth rate, but the Global Carbon Budget estimates it with dynamic global vegetation models (DGVMs) run once a year. Emulators could provide it sooner if they reproduce each model outside its training period. Here we show that convolutional neural network emulators of 19 of these models, driven by ERA5 climate and their own past predictions, reproduce the global land sink of these models and its spread over two years not used in training. Global monthly net biome production (NBP) anomalies were reproduced with a median R^2 of 0.81. The error of the multi-model mean annual NBP was at most 0.23 GtC yr^-1. At the grid-cell level, the median R^2 of monthly anomalies was 0.30, partly because of errors that accumulate in the emulated NBP history.

physics.ao-ph↗
arXiv2026

RadarFill: Observation-Constrained Gap Filling of Systematic Blind Zones in 3-D Weather Radar Reflectivity Mosaics

Ground-based weather radar mosaics contain systematic blind zones, especially at low altitudes, where missing reflectivity lacks corresponding complete observations for direct verification. We present RadarFill, an observation-constrained conditional diffusion framework for gap filling in three-dimensional radar reflectivity mosaics. RadarFill conditions on partially observed reflectivity and an observation mask, enforces available observations throughout reverse sampling, and preserves valid measurements in the final field. For training and controlled evaluation, structured synthetic missingness is designed from altitude-dependent missing-data statistics and the vertical nesting of blind zones in an operational 3-D radar mosaic product. Across 4,384 held-out test times and representative convective and typhoon cases, RadarFill recovers principal precipitation structures and moderate-to-strong echoes under extensive low-level missingness and generally outperforms conventional interpolation and a learning-based reference model. For real blind-zone evaluation, RadarFill model is applied directly to Fujian Province and compared with 90 matched volume scans from an independent phased-array radar. Within the principal 1-2 km low-level blind zones, RadarFill yields lower reflectivity errors than conventional interpolation, and the gap-filled precipitation structures show broad agreement with independent observations. In two controlled nowcasting experiments, improvements in RadarFill-initialized fields persist into subsequent forecasts, with the largest benefits at low altitudes and short lead times. These results provide complementary evidence from controlled synthetic missingness, independent radar observations, and downstream nowcasting, supporting the potential of observation-constrained conditional generation for gap filling in real 3-D radar blind zones.

physics.ao-ph↗