NeuralGCM
Efficient and accurate climate simulation model
- Combining traditional physics models with machine learning to improve simulation accuracy and efficiency
- Generate high-precision weather forecasts for 2-15 days
- Reproducing temperature data from the past 40 years with accuracy surpassing traditional models
- Learning the physical characteristics of small-scale events from existing meteorological data using neural networks
- Rewrite the numerical solver in JAX to achieve gradient based optimization adjustment
- Efficiently running on TPUs and GPUs, it has performance advantages compared to traditional models that mainly run on CPUs
- Provide open-source code and model weights for non-commercial use and further development by researchers
Product Details
NeuralGCM is a climate model developed by a Google research team. Compared to traditional physics based climate models, it combines machine learning techniques to improve the accuracy and efficiency of simulations. NeuralGCM is capable of generating weather forecasts for 2 to 15 days, which exceeds the accuracy of current gold standard physical models and is more accurate than traditional atmospheric models in reproducing temperature data from the past 40 years. Although NeuralGCM has not yet been constructed as a complete climate model, it marks an important step in developing more powerful and user-friendly climate models.
