TL;DR:
- China leverages AI technology to enhance weather forecasts, aiding in coping with extreme weather events.
- Shanghai Artificial Intelligence Laboratory’s Fengwu meteorological model proves superior in predicting typhoon paths.
- AI-driven weather models like Fengwu offer improved accuracy over traditional methods.
- Pangu Weather, another AI model, excels in accurate medium-range global weather forecasting.
- AI’s integration accelerates progress in weather prediction accuracy and opens doors to broader Earth science applications.
Main AI News:
In the midst of China’s sweeping artificial intelligence (AI) surge, a novel application has emerged within the realm of meteorology. The integration of AI technology has catalyzed a marked enhancement in the precision of weather forecasts, furnishing a more potent mechanism for mitigating the impact of weather calamities during a summer beleaguered by scorching heat waves, torrential downpours, and tempestuous typhoons.
Throughout July and August of the present year, the Shanghai Artificial Intelligence Laboratory embarked on a trial deployment of its innovative Fengwu meteorological model. This strategic maneuver sought to fortify the prediction and advisory capabilities in the face of impending typhoons, including the likes of Talim, Doksuri, and Khanun. The collaborative initiative encompassed concerted efforts between China’s National Meteorological Center and the Shanghai Meteorological Service.
Amidst the roster of this summer’s typhoons, Typhoon Doksuri stood as an exemplar of meteorological might. Remarkably, the 24-hour forecast error of the Fengwu model rested at a mere 38.7 kilometers—a notable improvement compared to the European Centre for Medium-Range Weather Forecasts (ECMWF), which exhibited an error of 54.11 kilometers, and the U.S. National Centers for Environmental Prediction (NCEP), characterized by a 54.98-kilometer discrepancy.
Similarly, during the recent occurrence of Typhoon Khanun, which swept across sections of northeastern China, the Fengwu model offered an impressive 48-hour prediction error of only 47.5 kilometers. In contrast, the ECMWF and NCEP models yielded respective errors of 54.5 and 63.8 kilometers.
Unveiled in April of this year, the Fengwu model was the result of a collective endeavor involving the Shanghai Artificial Intelligence Laboratory, multiple Chinese universities and research establishments, along with the Shanghai Central Meteorological Observatory. This pioneering model functions as a global medium-range weather forecasting tool, underpinned by a multi-modal and multi-task deep learning framework. Its efficacy materializes in the ability to predict core atmospheric variables with heightened resolution for over 10 days—distinctly surpassing traditional methodologies.
Distinguishing itself from the conventional physical models, which predominantly rely on supercomputers, the Fengwu model harnesses the prowess of a solitary graphics processing unit. This resource-efficient marvel generates high-precision global weather predictions for the ensuing 10 days within a succinct 30-second timeframe.
A luminary from the Shanghai Artificial Intelligence Laboratory affirms that the horizon of AI-driven weather prediction presents ample room for enhancement. Presently, accurate weather forecasting at the district level of a city is within reach; a future where such precision extends down to the street level remains a tantalizing prospect.
In concurrence with the Fengwu research collective, the augmentation engendered by AI in weather prediction proves formidable, yet the inherently capricious nature of weather remains a challenge for absolute precision. The symbiotic relationship between AI-driven models and conventional physical paradigms is anticipated to flourish, furnishing meticulous weather forecasts vital to sectors encompassing agriculture, forestry, animal husbandry, fishing, aviation, navigation, and public safety. Moreover, the reach of AI technology is poised to extend across diverse domains of Earth science, including environmental studies, astronomy, geology, and beyond, all in service of carbon neutrality, disaster mitigation, and energy security.
The time-tested realm of numerical weather prediction has yielded commendable outcomes in forecasting and disaster preparedness. However, the convergence of sluggish computational progress and the mounting intricacies of physical models has exacerbated the bottleneck in numerical prediction. The World Meteorological Organization contends that advancements in global medium-range weather forecasting transpire at a rate of merely one day per decade. This impasse beckons for AI-fueled forecasting methods underpinned by data-driven principles, poised to deliver precision without incurring exorbitant computational costs.
Enter Pangu Weather—an ingenious model forged in China, designed to unfailingly anticipate atmospheric conditions. Recent revelations from the Pangu Weather research unit at Huawei Cloud, published in the esteemed journal Nature this July under the title “Accurate medium-range global weather forecasting with 3D neural networks,” unveil a three-dimensional neural network calibrated to Earth’s coordinate system. This cutting-edge paradigm deftly processes intricate 3D meteorological data, adroitly minimizing the accumulation of errors via a hierarchical temporal aggregation strategy.
The forecasts rendered by the Pangu model encompass critical meteorological parameters—humidity, wind speed, temperature, and sea level pressure—integral to projecting weather system evolution, storm trajectories, air quality, and atmospheric patterns.
According to Huawei, the predictive efficacy of the Pangu model spans from one hour to seven days, surpassing the performance of select meteorological centers in Europe and the United States within the same temporal scope. A comparative assessment conducted between the Pangu model and the European numerical model from April to July, as documented by ECMWF, underscores the paradigm-shifting potential of AI-driven weather forecasting, poised to surmount the obstacle of sluggish progress in predictive accuracy.
For the curious and the concerned alike, the Pangu model is accessible online through the ECMWF website. This unparalleled resource affords global weather projections for the upcoming 10 days, accessible to worldwide meteorologists, ardent weather enthusiasts, and the general populace—completely free of charge.
The Central Meteorological Observatory of China avows its steadfast commitment to augmenting AI’s role in typhoon surveillance and prediction. A harmonious collaboration with academic institutions and research entities is set to redefine global typhoon monitoring, precise forecasting, and the delivery of premium-quality services. Through these synergistic endeavors, the union of AI and meteorology stands poised to weather the storm of uncertainty with unwavering precision and innovation.
Conclusion:
The intersection of artificial intelligence and meteorology in China signifies a transformative leap in weather forecasting precision. With AI-powered models like Fengwu and Pangu Weather delivering exceptional accuracy, the market can expect a paradigm shift in global weather prediction capabilities. Collaborative efforts among research institutions and technological innovators promise a future where accurate weather forecasts are not only possible but essential for various sectors and industries.
