Ensemble High-Resolution Weather Forecasting on AMD Instinct GPU Accelerators

2026-03-06 · Source: AMD ROCm Blogs · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Emerging Technologies & Innovation · Depth: Advanced, extended

Summary

AMD's blog post details how generative AI, specifically diffusion models, enhances high-resolution weather forecasting by managing inherent atmospheric uncertainty through ensemble predictions. The article focuses on StormCast, an NVIDIA convection-allowing model, demonstrating its ability to generate multiple slightly varied atmospheric states (an ensemble) to quantify forecast uncertainty. It explains how diffusion models transform a simple noise distribution into a complex target data distribution, forming the basis for StormCast's ensemble generation. The post also describes how to derive enhanced predictions from these ensembles using methods like the Probability Matched Mean (PMM), which counteracts the smoothing effect of traditional ensemble averaging. Practical instructions and Python scripts are provided for running and visualizing StormCast ensemble forecasts on AMD Instinct GPUs, such as the MI300X, using Earth2studio.

Key takeaway

For AI Engineers developing weather prediction systems, understanding and implementing ensemble forecasting with generative AI is crucial for managing inherent atmospheric uncertainty. You should consider integrating diffusion models, like those in StormCast, to generate probabilistic ensembles rather than relying solely on deterministic models. Utilize methods such as the Probability Matched Mean (PMM) to ensure your forecasts accurately represent the full range of possible outcomes, including extreme values, providing more comprehensive and reliable predictions.

Key insights

Diffusion models enable robust ensemble weather forecasting by generating diverse atmospheric states to quantify uncertainty.

Principles

• Uncertainty is inherent in physical observations.
• Ensembling quantifies forecast uncertainty.
• Diffusion models transform noise into target distributions.

Method

StormCast uses a deterministic model for a mean state, then a diffusion model to sample a residual. Adding this residual to the mean state creates an ensemble member, with different residuals leading to ensemble divergence.

In practice

• Use `run.ensemble` in Earth2studio for ensemble forecasts.
• Employ Probability Matched Mean (PMM) to preserve extreme values.
• Visualize ensemble outputs with `plot-stormcast.py`.

Topics

• Ensemble Weather Forecasting
• Diffusion Models
• StormCast
• AMD Instinct GPUs
• Probability Matched Mean

Code references

• NVIDIA/earth2studio

Best for: Machine Learning Engineer, AI Engineer, MLOps Engineer

Related on AIssential

• Pointwise is Pointless? A Multimodal Ablation Study for Precipitation Nowcasting with Graph Neural Networks — Multimodal GNNs improve precipitation nowcasting, but point observations require careful integration to impact dense spatial fields effectively.
• Rigorous uncertainty quantification of probabilistic AI weather forecasts with conformal prediction — Online conformal prediction rigorously calibrates AI weather forecasts, guaranteeing statistical coverage without performance loss.
• Utilizing AMD Instinct GPU Accelerators for Weather and Precipitation Forecasting with NeuralGCM — NeuralGCM combines GCMs with ML for stable, accurate, and physically consistent weather and precipitation forecasts.
• How Earth-2 Delivers Global AI Weather Forecasts on Demand — AI models can simulate atmospheric physics directly from observational data, outperforming traditional methods.
• EO-WM: A Physically Informed World Model for Probabilistic Earth Observation Forecasting — EO-WM uses physically informed diffusion transformers to improve probabilistic Earth Observation forecasting by better modeling weather impacts.
• Protecting cities with AI-driven flash flood forecasting — AI-driven flash flood forecasting using news data significantly enhances urban climate resilience globally.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AMD ROCm Blogs.