Sunday 02 March 2025
A new approach to predicting daily rainfall in Hong Kong has been developed, using a combination of machine learning and historical weather data. The system, which is now operational at the Hong Kong Observatory, uses an analogue forecast method that identifies similar weather patterns from the past and applies them to current conditions.
The idea behind the system is simple: by analyzing historical weather patterns, researchers can identify the most common types of rainfall events in Hong Kong and use those patterns to inform future forecasts. The system uses a type of machine learning called an autoencoder, which is trained on a large dataset of historical weather data. This allows it to learn the relationships between different weather variables, such as temperature, humidity, and wind direction.
When a forecast is needed, the system identifies the most similar patterns from the past and generates a forecast based on those patterns. The forecasts are then validated against actual rainfall observations, allowing the system to continually improve its accuracy over time.
The new system has been tested over a three-year period and has shown significant improvements in forecasting daily rainfall compared to traditional methods. It is particularly effective at predicting heavy rainfall events, which are a major concern for Hong Kong due to the city’s high population density and frequent flooding.
One of the key advantages of the system is its ability to handle complex weather patterns, such as tropical cyclones and typhoons. These events can be difficult to predict using traditional methods, but the analogue forecast approach allows the system to identify similar patterns from past events and apply them to current conditions.
The new system has already been put into operation at the Hong Kong Observatory, where it is being used to support daily weather forecasting. The team behind the research hopes that the system will help improve public safety and reduce the economic impacts of heavy rainfall events in the city.
In addition to its practical applications, the research also highlights the potential for machine learning approaches to be used in other areas of meteorology. The ability of the system to handle complex weather patterns and improve forecasting accuracy could have significant implications for the development of more advanced weather prediction models.
Cite this article: “Machine Learning System Improves Daily Rainfall Predictions in Hong Kong”, The Science Archive, 2025.
Machine Learning, Hong Kong, Rainfall, Forecasting, Weather Data, Analogue Forecast, Autoencoder, Tropical Cyclones, Typhoons, Meteorology
