Severe weather events, particularly heavy rainfall, are becoming increasingly common. Having reliable assessments of location and timing for such occurrences is crucial for saving lives and property. Researchers have developed a new AI method that converts low-resolution global weather data into high-resolution precipitation maps.
Researchers utilize historical weather model data, which describe global precipitation at hourly intervals with a 24-kilometer resolution. Their generative AI model, spateGEN-ERA5, was trained with this data. It learned from high-resolution regional radar measurements how precipitation patterns and extreme events correlate across scales.
This AI model generates multiple physically plausible, high-resolution precipitation maps, with details visible at 2-kilometer and 10-minute resolution. It also provides statistical uncertainty for the modelled predictions. Tests in the US and Australia have confirmed its broad applicability for different climates.
This AI tool presents new opportunities for assessing regional climate risks. It is particularly beneficial for vulnerable regions lacking detailed weather observations. The potential for more reliable assessments of heavy rainfall and flood threats and could aid in emergency disaster control and result in long-term preventive measures.
