Google launches “advanced weather forecasting AI” MetNet-3, claiming to predict super-traditional physical models

IT House reported on November 3 that Google Research and DeepMind collaborated to develop the latest weather model MetNet-3. This model is based on the previous MetNet and MetNet-2 and can predict global weather 24 hours in advance. High-resolution forecasts of conditions, including precipitation, surface temperature, wind speed, wind direction and felt temperature.

IT House found that Google mentioned that the MetNet-3 model has been implemented in the "Google Mobile Software" weather forecast on the mobile platform.

The MetNet-3 model can create "smooth and high-precision" forecasts with a spatial resolution of 1 to 4 kilometers and an analysis interval of 2 minutes. Experiments have proven that MetNet-3's forecasting capabilities surpass traditional physical weather Forecast models, such as the traditional physical basic model "NWP (Numerical Weather Prediction)" and "Rapid Refresh Model (HRRR)" are both surpassed by MetNet-3.

MetNet-3 is different from other machine learning methods based on traditional methods in predicting weather. The key point is to directly use atmospheric observation data for training and evaluation. The researchers noted that the advantage of direct observation is the higher density and resolution of the data. In addition, MetNet-3 not only inherits the previous MetNet model data, but also adds temperature and wind measurement data from weather stations to try to comprehensively predict weather conditions at various locations

Researchers mentioned that the key innovation of MetNet-3 is the use of a technology called densification to improve the accuracy and scope of weather forecasts.

In traditional physical basic models, weather forecasting usually requires two steps, namely data assimilation and simulation. Data assimilation refers to integrating actual observation data into the model, while simulation Predict the weather based on this data.

In MetNet-3, the densification technology that combines the two steps of "data assimilation" and "simulation" through neural networks can achieve faster and more direct weather predictions. This technology makes the model more efficient in acquiring and processing data, while using neural networks to improve the accuracy of weather forecasts. The MetNet-3 model is able to process each specific data stream individually, including contour information, satellite information, radar information, etc., to obtain more accurate and comprehensive weather forecasts

In addition, the MetNet-3 model uses "directly observed" data as learning samples to gain the advantage of high resolution based on space and time. Weather stations and ground radar stations provide location-specific measurements every few minutes with a resolution of 1 kilometer. In comparison, even the most advanced physical models in the world can only generate data with a resolution of 9 kilometers and provide hourly forecasts every 6 hours

MetNet-3 can effectively process and simulate the collected observation data at intervals as short as 2 minutes. Combining densification technology, Lead Time Conditioning technology and high-resolution direct observation methods, MetNet-3 can generate 24-hour forecasts with a time resolution of 2 minutes, providing users with more accurate and real-time weather forecast information.

In addition, compared to the weather information observed by weather stations, MetNet-3 also uses precipitation estimates collected from ground radar, so the range of learning data is wider, whether in terms of wind speed or precipitation, MetNet-3 The prediction results are much better than the most advanced physical models in the industry.

The main value of MetNet-3 is that it can accurately predict weather with machine learning technology in real time and provide weather forecast services on Google products. The model continuously creates complete and accurate forecasts based on the latest data that is continuously collected. Researchers mentioned that this is different from traditional physical reasoning systems and can better meet the unique needs of weather forecasting.

The above is the detailed content of Google launches “advanced weather forecasting AI” MetNet-3, claiming to predict super-traditional physical models. For more information, please follow other related articles on the PHP Chinese website!