Search arXiv⌕ 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

SwAIther-Precip: Lead-Time-Aware Bias Correction Enables Kilometer-Scale Downscaling of Global AI Precipitation Forecasts over Switzerland

Skillful medium-range precipitation forecasting at kilometer scale remains challenging over complex terrain because precipitation arises from multiscale nonlinear processes that global models cannot explicitly resolve at affordable cost. Global AI weather models can produce skillful medium-range forecasts, but their native 0.25 degrees resolution limits direct use for local hazard applications. Statistical downscaling can help bridge this gap, yet existing approaches often struggle with state-dependent, and especially lead-time-dependent, biases in global forecasts. We introduce SwAIther-Precip, a lead-time-aware downscaling framework that converts coarse-resolution AIFS forecasts into probabilistic km-scale precipitation fields over Switzerland. First, a U-Net conditioned on lead time via feature-wise linear modulation deterministically corrects systematic biases at coarse resolution. This targeted correction enables a cheaper super-resolution stage conditioned only on corrected precipitation, allowing direct training on observations rather than on the full atmospheric state. A diffusion-based model then generates fine-scale spatial variability independently of lead time. Using AIFS forecasts and CombiPrecip radar-gauge observations, SwAIther-Precip reduces CRPS by 48% relative to raw AIFS. The generated fields reproduce observed spatial variability with spectral fidelity above 0.85 at large scales and 0.88 at small scales, corresponding to an effective resolution of approximately 4 km on a 1 km grid for lead times up to 5 days. Training across lead times further improves long-range performance, yielding a 13% CRPS reduction at 6 days relative to lead-time-specific models. These results show that explicitly correcting lead-time-dependent biases before generative super-resolution is key to efficient km-scale probabilistic downscaling of global AI precipitation forecasts.

physics.ao-ph↗
arXiv2026

Wind vector retrieval from miniaturized wave-enabled sea-surface drifters

Open-ocean in situ winds are sparse, limiting the calibration and validation of satellite wind products. We evaluate wind-vector retrieval from MELODI (Miniaturized Electronics Lagrangian Oceanographic DrIfter), a compact freely drifting GNSS/IMU wave buoy. Wind speed is inferred from the shape of the measured wave acceleration spectrum. A regularized regression with a gated nonlinear correction (Wind Inversion using Tikhonov Regularization, WITR), trained on scatterometer winds and evaluated by leave-one-buoy-out cross-validation against ASCAT MetOp-B/C and HY-2B/C, gives a root-mean-square difference (RMSD) of 0.91 m/s. Symbolic regression distills the trained model into compact spectrum-only and motion-enhanced equations suitable for onboard use; the reduced drag law gives 0.93-0.94 m/s. Wind direction is estimated, without any fitted reference-wind regression, from IMU-derived directional wave moments in the 0.60-0.90 Hz wind-sea band, with a circular mean absolute angular difference (MAAD) of 9.4 degrees against ASCAT. These differences are comparable to published scatterometer-moored-buoy comparisons and improve substantially on the traditional single-band Toba inversion. Validation is strongest from 4 to 14 m/s and sparse above approximately 16 m/s; stronger winds are assigned low confidence. The results show that small wave-sensing drifters, calibrated to the well-established scatterometer wind scale, can provide useful wind-vector observations for validating satellite wind products and for atmospheric and wave modeling; stronger-wind sampling remains a priority.

physics.ao-ph↗
arXiv2026

Scale-aware and spatially-localised optimisation of kilometre-scale MLWP models

Preserving fine-scale structure without degrading spatial accuracy is a central challenge for kilometre-scale machine-learning weather prediction (MLWP). We introduce a loss based on the dual-tree complex wavelet transform that constrains atmospheric fields jointly by scale, location, and orientation. We evaluate its performance using an experimental joint global--regional FastNet configuration, which produces \SI{1.5}{km} forecasts over the UK alongside $\sim$\SI{31}{km} global forecasts. Relative to both mean-squared error (MSE) and an FFT-based spectral objective, the wavelet loss substantially improves spectral fidelity. Compared with MSE training, it maintains similar pointwise performance, improves neighbourhood-based verification, and produces sharper precipitation fields with more realistic intensity distributions. These results demonstrate that spatially localised wavelet constraints can preserve multiscale atmospheric structure while largely avoiding the loss of pointwise accuracy associated with globally supported spectral objectives.

physics.ao-ph↗