A typhoon monitoring and forecasting system based on artificial intelligence established by NMC
On May 22 and 23, the maximum winds near the center of this year's 2nd Typhoon Mawar strengthened from 38 meters per second (typhoon scale) to 60 meters per second (super typhoon scale) within 24 hours. The National Meteorological Centre (NMC) harnessed artificial intelligence (AI) rapid enhancement identification technology, and has achieved tendency forecast 12 hours in advance. Meanwhile, Huawei Cloud Pangu Model also performed well in the prediction of Mawar's track, predicting its turning track in the eastern waters of Taiwan Island five days in advance.
With the emergence of AI technology, the deep learning method combined with powerful data processing and characterization of complex structure features boasts a broad application prospect against the backdrop of meteorological big data. According to Mr. QIAN Qifeng, Deputy Director General of Typhoon and Marine Weather Forecast Center of NMC, NMC,in collaboration with scientific research institutes and universities, has carried out a series of AI exploration in typhoon monitoring and forecasting. It has established such technologies as typhoon vortex recognition, intelligent typhoon intensity identification and typhoon rapid enhancement identification, which play advantages in processing nonlinear and massive data, helping forecasters in increasing prediction accuracy.
In view of the advantages of AI in image recognition, the application of AI technology in typhoon objective positioning and intensity identification has attracted much attention.
In 2019, NMC and Beijing University of Posts and Telecommunications proposed an end-to-end visual intelligent typhoon intensity identification model. The model extracted satellite cloud image data to analyze the characteristics related to typhoon intensity, and then constructed a classification model and a retrieval model based on similarity to obtain decision-making results. Finally, by integrating the identification results of the two models, the typhoon intensity, confidence and reference cloud map are obtained. By analyzing and learning a raft of samples by machines, the deep learning method can implicitly extract the complex features of deep abstraction in images, and is increasingly applied in typhoon intensity estimation.
Based on the deep learning model of spatial-temporal correlation in the field of AI, NMC proposed an automatic and objective identification technology for the rapid typhoon enhancement trend by labeling and learning the key information in satellite cloud data of typhoons over the Northwest Pacific and South China Sea from 2005 to 2018 and introducing life cycle indication. A typhoon rapid enhancement identification model integrating spatial-temporal series features has been established.
According to Mr. ZHOU Guanbo, chief forecaster of Typhoon and Marine Weather Forecast Center, NMC has built a typhoon monitoring and forecasting system based on AI by establishing a typhoon vortex identification model, an intelligent typhoon intensity identification model and a typhoon rapid enhancement identification model. It provides significant technical support and reference products for ramping up the intellectual level of typhoon monitoring and forecasting operation in China and rapidly expanding the global multi-sea tropical cyclone operation.
NMC and Typhoon and Marine Weather Forecast Center will continue to intensify the application of AI in the field of typhoon monitoring and forecasting, and focus on the explanatory feature of AI. They will team up closely with universities, research institutes and other scientific research forces to further promote the deep integration of AI technology in typhoon monitoring, forecasting and services, furnishing innovative technical support for precise monitoring, accurate forecasting and tailored services of global services.
Is this page useful?Yes No Report an issue on this page
Thank you. If you have 2 minutes, we would benefit from additional feedback (link opens in a new window).