Global tech giants dive into AI weather forecasting business

[Image source: DALL·E]

The artificial intelligence (AI) weather forecasting industry is heating up, with big tech companies launching AI weather simulators and entering into preemptive competition.

There is an expectation that by integrating AI into traditional weather prediction systems that relied on high-cost supercomputers, prediction speeds will increase while costs will decrease.

Google LLC, Microsoft Corp, and Nvidia Corp., as well as Huawei Technologies Co. from China have all entered the fray.

Google Research on Monday published a paper introducing its weather forecast simulator NeuralGCM in Nature.

According to the paper, this new weather simulator combines machine learning with traditional forecasting methods based on complex equations used over the past 50 years to model atmospheric changes and provide forecasts (general circulation model).

However, NeuralGCM learns physics through AI from existing weather data. This approach supplements the drawbacks of the previous method, which was slow and costly, by employing AI for precise climate features like fog in regions where traditional models may not perform well.

According to Google, NeuralGCM is more than 3,500 times faster than X-SHiELD, a physics-based model. While X-SHiELD takes 20 days for a simulation, NeuralGCM can achieve it in 8 minutes.

There is also a significant reduction in computing costs. X-SHiELD requires a supercomputer equipped with over 13,000 central processing units (CPUs) while NeuralGCM only needs a computer with a single tensor processing unit (TPU), reducing computing costs by 100,000 times, according to Google.

Google also unveiled a weather forecasting AI model called GraphCast 8n November 2023, which garnered attention for its high performance at low cost compared to existing models.

While previous models such as GraphCast focused on short-term predictions, NeuralGCM, which was newly introduced by Google, emphasizes its capability for both long-term and climate predictions.

Researchers are developing features to predict hurricanes a year in advance via NeuralGCM, expecting it to aid in preparing for storms and building infrastructure.

Nvidia, an AI semiconductor company, also unveiled its weather forecasting AI in 2024. Earth-2 is a digital twin created based on Nvidia’s climate prediction technology, capable of simulating the Earth’s atmospheric environment and achieving speeds 1,000 times faster than conventional meteorological methods.

Taiwan’s Central Weather Administration recently adopted Earth-2, while Microsoft unveiled the foundation model ClimaX in January 2023.

China is leading among countries outside the United States. Huawei Cloud’s Pangu-Weather, introduced in Nature in July 2023, predicts 10,000 times faster than existing models. The company has made this model available to the public for free, providing weather forecasts for up to 10 days.

Google, Nvidia, Microsoft, and Huawei are all developing high-performance AI, as they virw the weather forecasting sector as one where AI can significantly increase efficiency.

Meteorological prediction has traditionally relied on high-performance supercomputers, which were costly and time-consuming to use due to their reliance on CPU-based sequential operations.

With the advent of AI using GPUs for calculations, it is expected that meteorological predictions can be made faster and more efficiently. But there are concerns among researchers about AI being a “black box,” as it is not clear what its predictions are based on.

According to the Korea Meteorological Institute, the meteorological forecasting market is expected to grow from $3.09 billion to $5.8 billion in 2027.