Shulou(Shulou.com)11/24 Report--

CTOnews.com, Nov. 3, Google Research and DeepMind have jointly developed the latest weather model MetNet-3, which is based on the previous MetNet and MetNet-2 and can predict global weather conditions with high resolution 24 hours in advance, including precipitation, surface temperature, wind speed, wind direction and somatosensory temperature.

CTOnews.com found that Google mentioned that the MetNet-3 model has been installed in the weather forecast of the "Google mobile software" on the mobile platform.

MetNet-3 model can create "smooth and high precision" prediction, the spatial resolution can reach 1 to 4 km, and take 2 minutes as the analysis interval. The experimental results show that the prediction ability of MetNet-3 surpasses the traditional physical weather forecasting model, such as the traditional physical basic model "NWP (Numerical Weather Prediction)" and "Fast Refresh Model (HRRR)" are both surpassed by MetNet-3.

MetNet-3 is different from other machine learning methods based on traditional methods in weather prediction, the key point is that MetNet-3 is trained and evaluated directly through atmospheric observation data. The researchers note that the advantage of direct observation lies in higher data density and resolution. In addition, in addition to inheriting data from previous MetNet models, MetNet-3 also learns temperature and wind measurements from weather stations to try to make omni-directional weather forecasts for all locations.

The key innovation of MetNet-3, the researchers note, is the use of a technology called Densification to improve the accuracy and range of weather forecasts.

In the traditional physical basic model, weather forecast usually needs to go through two steps, namely, data assimilation (Data Assimilation) and simulation (Simulation). Data assimilation refers to the integration of actual observed data into the model, and simulation is based on these data to predict the weather.

In MetNet-3, densification technology combines the two steps of "data assimilation" and "simulation" through neural network to achieve faster and more direct weather prediction, which will make the model more efficient in obtaining and processing data, and can also use neural network to improve the accuracy of weather forecast. The MetNet-3 model can separately deal with each specific data stream, including contour information, satellite information and radar information, so as to obtain a more accurate and comprehensive weather forecast.

In addition, the use of "direct observation" data as learning samples brings high resolution advantages based on space and time to the MetNet-3 model, and weather stations and ground radar stations can provide location-specific measurement data at a resolution of 1 km at frequencies every few minutes. By contrast, even the most advanced physical models in the world can only generate data with a resolution of 9km every six hours and provide hourly forecasts.

While MetNet-3 can effectively process and simulate the collected observation data in a time interval as short as 2 minutes, combined with densification technology, time in advance adjustment (Lead Time Conditioning) technology and high-resolution direct observation method, MetNet-3 can produce a 24-hour forecast with a time resolution of 2 minutes, providing users with more accurate and real-time weather forecast information.

In addition, compared with 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, and the prediction results of MetNet-3 are much better than the most advanced physical models in the industry in terms of wind speed and precipitation.

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

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.