How NVIDIA Is Rewriting Weather and Climate Forecasting With Open AI

Accurate weather and climate forecasting underpins everything from disaster preparedness to renewable energy optimisation. Traditionally, these forecasts have relied on physics-based numerical models running on large, energy-intensive supercomputers. While scientifically robust, such systems are expensive, slow, and accessible only to a small number of national meteorological agencies.

Recent advances in artificial intelligence are reshaping this landscape. AI-driven models can ingest vast volumes of atmospheric data, generate high-resolution forecasts in minutes rather than hours, and operate with a fraction of the energy and cost of conventional systems. This shift is expanding access to advanced climate intelligence for scientists, businesses, and policymakers at a time when extreme weather risks are rising globally.

 

Introducing NVIDIA Earth-2 as an Open Climate Platform

 

At the center of this transition is NVIDIA, which has launched Earth-2, an open-access AI platform designed to support weather and climate prediction at scale. Earth-2 brings together models, frameworks, and inference tools that were previously fragmented across institutions, enabling users to generate everything from near-term forecasts to multi-week outlooks within a unified environment.

According to NVIDIA, Earth-2 is the first open, accelerated software stack that allows organisations to run, customise, and deploy production-grade weather AI on their own infrastructure. This approach lowers barriers to entry while encouraging collaboration across research, public agencies, and industry.

 

Accelerating Forecasting Across the Entire Weather Pipeline

 

Earth-2 replaces key bottlenecks in traditional forecasting by using GPU-based AI models to accelerate every stage of the workflow. Tasks such as data assimilation, short-term nowcasting, and medium-range prediction can now be completed in seconds rather than hours.

The platform is designed as an open ecosystem. Researchers and operational users can fine-tune models for regional conditions, integrate proprietary data, and deploy forecasts tailored to specific applications such as grid management, insurance risk modelling, or emergency response. This flexibility enables high-resolution, localised insights that were previously difficult to produce outside of national forecasting centres.

 

Read more: Greece Ranks Among Europe’s Top Renewable Power Leaders in Record-Breaking Year

 

A Portfolio of AI Models Covering the Atmosphere

 

Earth-2 incorporates several specialised models that address different forecasting horizons and use cases. The Medium Range system, built on NVIDIA’s Atlas architecture, delivers forecasts up to 15 days across more than 70 atmospheric variables and has demonstrated accuracy gains over leading open-source alternatives.

For short-term hazards, Earth-2 Nowcasting uses generative AI to deliver kilometre-scale predictions for storms and extreme weather within a six-hour window. These rapid forecasts are particularly valuable for operational decisions where minutes and local impacts matter.

The platform also includes Global Data Assimilation capabilities through a model called HealDA, which produces detailed atmospheric starting conditions that improve downstream forecast accuracy. Combined, these components form a fully AI-driven forecasting pipeline capable of matching or exceeding traditional approaches in many scenarios.

 

Early Adoption Across Energy, Insurance, and Public Services

 

Earth-2 is already being adopted by organisations across multiple sectors. National weather agencies are using its models to reduce computational demand while improving forecast skill. The Israel Meteorological Service reports significant reductions in compute time alongside stronger precipitation verification when using Earth-2 downscaling models.

In the energy sector, companies such as TotalEnergies, Eni, and GCL are applying AI-based forecasts to improve solar output prediction and manage short-term operational risks. Insurers and data providers, including AXA and S&P Global, are using generative models to simulate thousands of extreme-weather scenarios for climate risk assessment.

These applications highlight how AI forecasting is moving beyond academic research into operational decision-making across climate-exposed industries.

 

Explore OneStop ESG Marketplace: AI (Artificial Intelligence)

 

Sustainability Gains Beyond Forecast Accuracy

 

Beyond speed and precision, Earth-2 delivers meaningful sustainability benefits. AI models run on GPUs require far less energy than traditional supercomputing systems, significantly lowering the carbon footprint of large-scale forecasting. Faster processing also allows more frequent scenario analysis, supporting better planning for renewable integration, infrastructure resilience, and disaster response.

By making its models openly available through platforms such as GitHub and Hugging Face, NVIDIA is positioning Earth-2 as shared climate infrastructure rather than a closed commercial product. This openness encourages innovation while aligning with broader goals of transparency and scientific collaboration.

 

AI as Infrastructure for Climate Resilience

 

Earth-2 reflects a broader shift in how climate intelligence is produced and consumed. Rather than relying solely on centralised institutions, AI enables distributed, scalable forecasting that can be adapted to local needs while maintaining scientific rigor.

As climate risks intensify, tools that allow governments, companies, and researchers to anticipate impacts more accurately and act more quickly are becoming essential. By combining open science with advanced AI, NVIDIA is contributing to a new model of climate infrastructure, one where insight is faster, more accessible, and increasingly embedded in real-world decision-making.

In this context, Earth-2 is less about replacing traditional science and more about extending its reach. The platform illustrates how AI, when designed as open and collaborative infrastructure, can strengthen planetary resilience while reducing the environmental cost of understanding the climate itself.