Google Research’s NeuralGCM Delivers Faster, More Accurate Atmospheric Simulations

July 22 (Portaltic/EP) –

NeuralGCM it’s a new atmospheric model open-access model developed by Google Research that combines physics-based modeling and artificial intelligence to simulate the Earth’s atmosphere more quickly and accurately than existing models.

NeuralGCM offers More accurate and faster simulations than current climate models that attempt to predict the consequences of the warming experienced by the Earth in different regions.

Compared with X-SHiELD, Google Research’s new model was 15 to 50 percent less likely to predict global temperatures and humidity for 2020, and it produced those predictions in eight minutes versus X-SHiELD’s 20 days, according to a paper in Nature. your official blog.

Likewise, it is available on GitHub, from where researchers can work with NeuralGCM, and it uses fewer computing resources since it can run on a single TPU compared to the 13,000 CPU supercomputers required by X-SHiELD.

NeuralGCM is trained with data from the ERA5 reanalysis, an open-access dataset containing a three-dimensional reconstruction of the Earth’s atmosphere over the past 80 years, produced by the European Centre for Medium-Range Weather Forecasts (ECMWF).