Source: Science and Technology Daily
Author: Liu Yan
In recent years, numerical weather forecasting methods have achieved great success in the fields of daily weather forecast, extreme disaster warning, climate change forecasting.However, as the computing power growth and the gradually complexity of the physical model, the bottlenecks of traditional numerical forecasting are becoming increasingly prominent, and new weather forecasting methods are urgently needed.Among many prediction methods, scientists have aimed at rapid development of artificial intelligence.
The United Nations Meteorological Organization and its partners announced on August 8 that in July 2023, it became the month with the highest average temperature in the world since human meteorological records.In addition to heat waves, some countries have encountered continuous rainstorms and floods.The extreme weather incidents brought by climate change have become a reality we have to face.
The sooner the accurate information of extreme weather, the more conducive to human response.Among many prediction methods, scientists have aimed at the rapid development of artificial intelligence (AI).Recently, scientists from China and the United States have published research results on top international academic journals, respectively, revealing the potential of artificial intelligence to assist in forecasting weather.
1 hour to 7 days forecast accuracy overdue weather forecast
In recent years, numerical forecasting methods have achieved great success in the fields of daily weather forecast, extreme disaster warning, climate change forecasting.However, as the computing power growth and the gradually complexity of the physical model, the bottlenecks of numerical forecasting are increasingly prominent, and researchers have begun to dig a new way to predict weather.
On July 6, the Research results of the Research and Development Team of Huawei Yunpan Gu Da Da Da Models -3D neural networks were used for precision and medium -term global weather forecast.AI model of forecasting method.
According to the relevant personnel of the Huawei Yunpan Model R & D team, the application of numerical forecasting methods in the fields of medium and long -term forecasting is the most widely used.In these fields, the accuracy of the existing AI forecast methods is still significantly lower than the numerical forecasting method, and it is restricted by problems such as lack of explanatory explanatory and inaccurate weather prediction.The main reason for the lack of accuracy of the AI forecast model is that the first is that the original AI forecast models were built based on the 2D neural network, and it cannot handle the uneven 3D weather data; the second is that the AI forecast method lacks mathematical physical mechanisms lacks mathematical physical mechanismsThe constraints will continue to accumulate iterative errors during iteration.
For this reason, the R & D team of Huawei Yunpan Gu Da Da Da Model creatively proposed the three -dimensional neural network that adapts to the earth coordinate system to process complex uneven 3D weather data, and use the domain aggregation strategy to reduce the number of forecast iterations when using hierarchicalization, therebyReduce iteration errors.Huo Houkun, the chairman of Huawei, said that in the field of meteorological forecasting, the Pangu Daping model has exceeded the prediction accuracy of 1 hour to 7 days, which has exceeded the prediction accuracy of some Meteorological centers in Europe and the United States within the same forecast time.
Meteorological Models have been showing their skills in extreme weather forecasts
The European Mid -term Weather Forecast Center (ECMWF) has always called on the global weather forecasting community to make more efforts, use the AI model as an additional part of its forecast system, and further explore the advantages and disadvantages of such models to help the weather forecastEssence
Chen Yunzheng, deputy director and researcher of the Institute of Computing Technology of the Chinese Academy of Sciences, pointed out that the focus of AI -based meteorological science research is to improve the seasonal prediction and long -distance space -to -model predictive ability to improve the time scale of more time.This achieves precise forecast and control of meteorological systems.
The director of the European Medium Weather Forecast Center Florence Habier in detail at the 19th World Meteorological Conference, the real -time operation inspection comparison of Huawei Yunpan Book of Metropolitania and the European Medium Weather Forecast Center was amazing.The prediction ability made the participants on the spot felt the huge energy of AI technology.
In the fields of agriculture, aviation, energy, disaster warning, accurate weather forecasting has significant social and economic value.However, due to factors such as the accuracy of meteorological observation, the complexity of the physical processes in the atmospheric system, the computing resources required for traditional numerical forecast methods are huge.According to the World Meteorological Organization data, the effectiveness of the global weather forecast can be improved by one day every 10 years, and data -driven AI forecast methods will be expected to quickly achieve high precision predictions at a lower calculation cost.
In 2020, the AI forecasting method is still far behind the numerical method in terms of accuracy. Today, the Pangu Meteorological Model has become the first AI model to exceed the numerical forecast method.Not only that, its prediction speed has increased by 10,000 times compared to the traditional numerical forecast, which can achieve the "second -level" global meteorological prediction. The results of the meteorological prediction include the potential, humidity, wind speed, temperature, sea level air pressure and many other information. TheseInformation is critical to predict the development of weather systems, storm trajectory, air quality and weather mode. It can be directly applied to multiple meteorological research segmented scenarios.
The European Medium -term Forecast and China National Meteorological Center and other institutions have verified the superiority of Pangu Meteorological Model in the measured measurement.
The comparative test reports of the Pangu Ancient Meteorological Model and the European numerical model announced by the European Medium Meteorological Center showed that the AI forecasting method represented by the Dipka Meteorological Model will break through in recent years.bottleneck.The Central Meteorological Observatory stated that Huawei Yunpan Gu Da Model had previously performed well in the path forecast of the typhoon "Mova", and has been applied to the forecast of the typhoon path of "Du Surui" this year.
is auxiliary or replaced the existing weather forecast system, it is unknown
As the Researcher Ma Zhuguo, a researcher at the Institute of Atmospheric Physics of the Chinese Academy of Sciences, the economic losses and personal safety risks caused by extreme weather and climate cannot be ignored.
In order to minimize the loss as possible, meteorological scientists have been striving to improve the accuracy of forecasting.Although the technical means adopted by meteorological forecasting are experiencing rapid iteration and progress, the medium- and long -term weather in the next few weeks or months is still facing many challenges with AI.
Ma Zhuguo pointed out that people do not know much about the process of climate change, so they have to assume assumptions when studying some climate phenomena, but the conclusions obtained are sometimes not very accurate, because the more accurate the model, the observations needed, the observations requiredThe more information.The development of new technologies is often difficult to break through its own limitations. At present, the most advanced AI technology is only to achieve the processing of "existing information and data that exists enough".Although the application of AI technology in the field of meteorology, it represents the huge improvement of its performance, but there are many unpredictable weather in the future. Once the data accuracy in the model is insufficient, the prediction results will cause errors.
When AI enters the application scenarios such as meteorological forecasting and atmospheric physics, it is essentially a more effective integration of big data such as strong computing power and smarter algorithms to improve the accuracy and efficiency of forecasting.At present, there are still many problems in the field of meteorological fields that need to be broken.
As Dr. Tian Qi, chief scientist in the field of Huawei Cloud's artificial intelligence, said: "Weather forecast is one of the most important scenes in the field of scientific computing and a very complicated system. At present, the main ability of Pangu Meteorology model is to predict the atmosphere of the atmosphere.The evolution of the state is to strengthen the existing forecast system. Our ultimate goal is to use Pan Gu Da's model to create the next generation of AI meteorological forecasting framework. "
Some people in the industry have pointed out that although the Pangu ancient weather model has opened up new forecasting ways, whether it can supplement or replace the existing weather forecast system, it also needs to further study and verify the team, as well as further experts in the field of weather forecasting.Evaluate.
In addition, the complicated meteorological laws, high resolution, and huge amount of data determine that the AI meteorological forecast requires the use of a highly calculated AI model.Therefore, to create a continuous iterative leading AI meteorological forecast model, a stable cloud environment and correspondingWork kits are essential.
