The Hong Kong Observatory said on Friday that weather predictions carried out using artificial intelligence have surpassed conventional methods, and helped it to carry out early forecasts.

Speaking on an RTHK radio programme, acting senior scientific officer He Yuheng said a week prior to the downpours that lashed the SAR on Tuesday, some AI models had already forecast that there would be an east-to-west rainband affecting Guangdong.

Even though rainfall distributions predicted by AI might not be too accurate, He said, they have allowed the observatory to conduct predictions earlier to a certain extent and inform the public sooner about unstable weather conditions.

He said that when the observatory began using AI in mid-2023, the models only offered limited predictions for meteorological elements such as temperatures and humidity levels. Now it can directly predict rainfall without analysing other parameters.

AI models can even better predict the weather than conventional methods, he said.

"At least for now, if we talk about AI models' performance on the medium-range weather forecasts of large-scale meteorological patterns for the next 10 days, they have reached – or even slightly exceeded – the level of conventional methods," He said.

"For example, on the location and change in intensity for an anticyclone aloft and subtropical ridge of high pressure, AI is better."

However, He said AI models are unable to forecast the intensity of strong typhoons for now, while other AI-based predictions must be verified manually.

As for the coming days, He said an area of low pressure over the northeastern part of the South China Sea could bring showers to the SAR early next week.

The observatory expects another area of low pressure that has developed into a tropical cyclone north of Guam to move westward towards Taiwan.