GenCast from Google DeepMind outperforms the European Centre for Medium-Range Weather Forecasts ENS system by 20% for day-to-day weather, extreme events, and hurricane paths. The model produces 15-day forecasts in eight minutes by using 40 years of historical weather data on a single Google Cloud TPU. Jua delivers spatial resolution of one kilometre square every five minutes. The model predicts over 15 parameters. Jua requires 1000 times less computing power than numerical weather models and provides 25 times higher spatial resolution.
GraphCast uses a graph neural network with 36.7 million parameters. It completes a 10-day forecast in less than 60 seconds using an encode-process-decode architecture. Pangu-Weather uses 64 million parameters and 13 pressure levels to model the three-dimensional volume of the atmosphere. Pangu-Weather uses Earth-specific positional encoding to learn different physical rules for different parts of the atmosphere. Aurora uses 1.3 billion parameters and 3D Swin Transformers to handle datasets with different spatial resolutions and pressure levels.
Can AI models eventually bypass the need for physics-based equations entirely?
Satellite and sensor networks
Tomorrow.io maintains 11 TMS satellites across six orbital planes to achieve hourly average global revisit rates. The 6U CubeSats use microwave wavelengths to penetrate clouds and observe atmospheric structures, such as moisture patterns and precipitation cores, that traditional visible and infrared satellites cannot capture. The TMS constellation uses 12 channels to sample temperature and humidity at different vertical levels throughout the atmosphere. Tomorrow.io generates Level 1 and Level 2 products, including resampled geolocated products and storm-top height estimates, using Argo Workflows and Kubernetes infrastructure.
Spire Global holds a 3.7 million dollar NOAA contract to deliver GNSS radio occultation data through December 1, 2026. This data provides vertical profiles of temperature, pressure, and humidity. Spire’s data includes temporal resolution enhancements to capture atmospheric evolution during periods that global RO coverage historically missed. Sorcerer uses high-altitude balloon airborne sensors to collect data to address the shortage of weather stations.
You should note that these commercial constellations supplement, rather than replace, government-operated satellites like NOAA’s ATMS or EUMETSAT’s MetOp series.
Industry-specific applications
| Company | Primary Industry Application | Specific Capability |
|---|---|---|
| Jua | Energy | Probabilistic short-term forecasts |
| Tomorrow.io | Aviation/Logistics | Weather-related delay reduction |
| Rainmaker Technology | Agriculture | Cloud-seeding via drones |
| Merqato | Supply Chain | Crop volume and price forecasting |
Jua provides the energy sector with its first global probabilistic short-term forecast, which helps improve profitability through higher resolution data and more frequent updates. JetBlue saves 3.7 million dollars annually by using Tomorrow.io’s data to reduce weather-related flight delays and cancellations. Rainmaker Technology uses weather-resistant drones to release silver iodide into storm clouds to induce rain or snow. Silurian AI creates models to simulate how weather impacts specific assets and infrastructure.
In Africa, Africlimate AI builds local capacity and develops solutions for climate-related challenges by improving local dataset availability. Sencrop uses machine learning with 35,000 weather stations to assist UK farmers with disease risk planning. Planette Analytics delivers long-range weather intelligence using 15 years of Earth science advancements. Capalo AI optimizes energy storage usage for profitability. Solarad AI provides an AI-powered platform for solar energy forecasting to reduce penalties for large-scale solar utility projects. Merqato uses AI-powered intelligence to forecast the volume and price of over 20 crops for the fruit and vegetable supply chain. Celest Science uses deep learning and generative models to provide seasonal climate forecasts and risk assessments. Ogre uses data science to shape the energy industry through utilities digitalization. Perceptive Space develops software to provide hyperlocal forecasts tailored to specific orbits using AI for space weather monitoring.
