Train your AI weather models with physics-based simulations.
Weatherwise specializes in simulating past and future weather events at a 400-meter resolution worldwide. Our team of meteorologists and engineers provides reliable data to help you train your AI models on realistic weather scenarios.
Easy to configure
Configure boundary conditions between GFS (forecast and hindcast) and IFS (hindcast), define the geographical domain, spatial resolution (10km, 2km, 400 meters) and output variables.
Fast to compute
Our objective is to help you get access to physics-model output in a few hours. So most of our efforts are dedicated to optimise how our weather model runs on parallel computing resources.
Streamline data access
Track computing status in real-time on a dedicated dashboard. Once the simulation is done running, you can visualise its outputs online and download NetCDF data archive.
The leading physics-based platform to train AI weather models.
global scale
GFS
State of the art weather models Global Forecast System (GFS) and Integrated Forecast System (IFS) data serve as the large-scale boundary condition input, establishing the initial and boundary conditions for the Meso-NH model.
synoptic scale
Meso-NH
Meso-NH is the research limited-area model used to simulate small-scale meteorological phenomena such as turbulence, convection, microphysics, cloud formation, radiation, wildfire propagation, and dust aerosols.
fine scale
Weatherwise
On-demand weather forecast capable of resolving km-scale processes, down to 400-meter resolution globally.
A focus on performance and experience.
Peer-reviewed model
Meso-NH is the research model we use to simulate meteorological events—it is referenced in 1,000+ publications and its physics module is used in the French operational weather model.
High-performance computing
Running a 24 hours simulation at 400 meters resolution on a 120km * 120km domain takes 3 hours to complete. For benchmark we can produce a 1 year simulation with this setup in less than a week.
On-demand
Our user-friendly interface ensures a smooth setup experience, from selecting the domain area, to the list of output variables you wish to analyse to train your AI models.