Huawei Cloud, in collaboration with the meteorological bureau of the Shenzen municipality, has launched the first regional of its kind called Zhiji. The name “Zhiji” is a combination of characters evoking AI technology, Chinese culture, and good weather.
Highlights:
- Zhiji is the new AI weather prediction model, developed by the team behind Huawei’s Pangu-Weather model.
- It can generate 5-day forecasts for Shenzhen and its neighbouring regions with a spatial resolution of 3 km.
- The model showed 10,000 times faster prediction over traditional weather prediction models.
Meet HuawaZhiji AI Weather Forecaster
Zhiji leverages Huawei Cloud’s Pangu-weather model and was pre-trained on high-quality regional datasets. The Pangu weather model was announced in July of last year and was the first AI prediction model to demonstrate a much higher precision than traditional models.
Zhiji model allows short-term forecasts spanning a 5-day period to be created for an accurate 3 km range. This is a significantly improved scope compared to traditional weather prediction models that operate in a spatial perimeter of 25 km. The forecasts cover a wide range of meteorological elements, including temperature, precipitation, and wind speed.
It allows for a 10,000x improvement in prediction speed, reducing global weather prediction time to mere seconds.
During its trial run last month, Zhiji demonstrated high accuracy in predicting multiple cold-temperature periods. Going forward, the team plans to further enhance the model and refine its ability to provide accurate precipitation forecasts.
What were the Challenges faced by Traditional Models?
There was a widely held assumption that AI forecasting is inferior to traditional numerical forecasting methods. With the rapid development of computing power over the past 30 years, the accuracy of numerical weather forecasts has improved dramatically, providing extreme disaster warnings and climate change predictions.
However, the method remains relatively time-consuming. To improve prediction speeds, researchers have been exploring using deep learning methods.
Still, the precision of AI-based forecasting for medium and long-term forecasts has remained inferior to numerical forecasts. AI has been mostly unable to predict extreme and unusual weather such as typhoons.
AI models for weather prediction have been attractive for their minimal computing time but have suffered in prediction for two reasons.
Firstly, the existing meteorological models are based on 2D neural networks and can’t handle uneven 3D meteorological data for accurate predictions.
Second, the medium-range forecast can develop cumulative forecast errors when the model is called too many times. Due to this, previous predictions influence the current prediction causing errors.
How is Pangu-weather better?
During its trial period, Pangu-weather demonstrated its higher precision compared to traditional methods with a huge increase in speed and in accuracy. The model can accurately predict in seconds, fine-grained meteorological features including humidity, wind speed, temperature, and sea level pressure.
The Zhiji model uses a 3D Earth-Specific Transformer (3DEST) architecture to process complex non-uniform 3D meteorological data. Using a hierarchical, temporal, aggregation strategy, the model was trained for different forecast intervals using 1-hour, 3-hour, 6-hour, and 24-hour intervals.
This resulted in a minimization of the number of iterations for predicting a meteorological condition at a specific time and a reduction in erroneous forecasts.
To train the model for specific time intervals, the researchers trained 100 epochs (cycles) using hourly samples of weather data from 1979-2021. Each of the sub-models that resulted required 16 days of training on 192 V100 graphics cards.
Pangu-Weather Model can now complete 24-hour global weather forecasts in just 1.4 seconds on a V100 graphics card, a 10,000-time improvement compared with the traditional numerical prediction.
In August, Huawei Cloud’s Pangu-Weather Model was made available publicly on the website of the European Center for Medium-Range Weather Forecasts (ECMWF)
Huawei Cloud also announced a potential collaboration with the Thai meteorological department in December 2023 to develop a Pangu-Weather model for monsoon prediction in Thailand.
The monsoon season in Southern China is approaching, and Huawei Cloud with the Meteorological Bureau of Shenzhen Municipality plan to work together to further verify and comprehensively evaluate Zhiji during this season, as well as continue to enhance the model and use it to provide useful information for weather forecasters.
Conclusion
With such massive improvements in both prediction time and prediction accuracy, the impact of Zhiji model is undeniable. With the results from the preliminary testing of Zhiji, it won’t be too long before the technology is adopted in areas with volatile weather and cyclone and typhoon-prone regions across the world
