While meteorologists have long struggled with the capricious nature of weather forecasting, artificial intelligence is changing the game. Since mid-2023, the Hong Kong Observatory has been letting AI do what humans couldn't: predict rainfall directly without fussing over temperature and humidity initially. Pretty revolutionary stuff.

The results? Not too shabby. AI's medium-range forecasts now match or beat traditional numerical weather prediction models in terms of large-scale patterns. It's particularly good at tracking those pesky anticyclones and subtropical ridges that conventional models struggle with. Still can't nail typhoon intensity though. Some things remain mysterious, even to the machines.

AI forecasts rival or outperform traditional models, excelling with anticyclones but still stumped by typhoon intensity.

Here's where it gets interesting. AI actually predicted, a full week ahead, an east-to-west rainband that would soak Guangdong before record-breaking downpours hit. The exact spots that got drenched weren't perfectly mapped, but hey—timing and general trends were spot-on. That's better than most meteorologists could manage with their fancy degrees. The system's regular software updates ensure optimal security and performance in weather predictions.

These AI systems aren't lightweight operations. They're trained on a staggering 40 years of global weather data—about 25 trillion data points. Twenty-five trillion! The computational power required is enormous, but once trained, these systems spit out forecasts faster than traditional methods. The technology combines remote sensing data with traditional observations to enhance prediction accuracy.

The real-world impact has been significant. AI-enabled earlier warnings informed Hong Kong's second-longest black rainstorm warning during the August 2025 downpour. Tsim Sha Tsui recorded an unprecedented 14.1 inches of rainfall, marking the highest daily precipitation in August since records began in 1884. People had time to prepare. Lives were potentially saved.

Don't toss out the traditional weather infrastructure just yet, though. The Observatory still relies on conventional methods for emergency alerts. They're also candidly honest about AI's limitations when talking to the public. No overselling the magic.

Looking ahead, Hong Kong Observatory is doubling down on AI development. The focus? Better typhoon intensity prediction and improved spatial accuracy. Because let's face it—knowing exactly where that torrential downpour will hit matters when you're planning a beach day.