Nvidia presented a shi model to build a digital double

NVIDIA announced CBottle, a generative SI model that creates a “digital double” of land with detail at a kilometer level. Development position as a new stage in the use of artificial intelligence in climatic studies.

The model became part of the Earth-2 platform. It allows you to model climatic scenarios in the real time, such as the parameters, such as the ocean temperature, time of day and season. The training has been carried out on historical atmospheric data over the last 50 years.

Data: NVIDIA github reposition.

CBottle generates forecasts much faster and at the expense of lower energy costs than traditional climatic models. It is stated that it is able to adjust the bias of simulations, restore lost data, scale climate images and squeeze samples up to 3000 times without loss of accuracy.

“AI allows you to move from cumbersome models to adaptive systems available to a wide range of researchers,” Nvidia said.

The project partners have already become Max Planck Institute and Allen Institute for AI.

The model was tested within the international hackathon Global KM-Scale Hackathon with the participation of researchers from eight countries. AI2 called CBottle “a new tool for accurate modeling of extreme weather conditions – showers, droughts and fires.”

According to Nvidia, the model can be the basis for creating an interactive system of queries for climatic data – a digital twin of the planet suitable for practical use.

Recall that Google has created a Shi model that can predict hurricanes more precisely than traditional systems.

The gun