A new AI tool from Google promises to improve climate forecasting: Developed by Google Research, NeuralCGM uses physics-based modeling and AI to create fast and accurate simulations of Earth's atmosphere.
Traditional climate models work by dividing the Earth into large pixels, making it difficult to accurately predict small-scale conditions like clouds, turbulence, and convection. Combining physics-based simulation with AI appears to solve this problem. In 2020, NeuralCGM predicted annual temperature and humidity with 15-50% more accuracy and much faster than the non-AI X-SHiELD. Neural CGM produced forecasts in 8 minutes that X-SHiELD took 20 days to produce.
Google claims that NeuralCGM can run on a single AI chip, or TPU (Tensor Processing Unit), but some high-resolution atmospheric models require expensive supercomputers and thousands of chips called CPUs (Central Processing Units). That means the model could be used by laptops and researchers around the world. Greater efficiency could also reduce the model's energy consumption.
The news comes weeks after Google's own environmental report concluded that its use of energy-hungry AI has caused emissions to increase by nearly 50 percent over the past five years.
topic
Google Innovation
