On November 12, 1970, the Bhola Cyclone struck East Pakistan, generating winds of 130 mph and a surge of 35 feet, resulting in 300,000 to 500,000 fatalities, making it the deadliest tropical cyclone recorded. Had this occurred a decade later, advancements in weather forecasting could have mitigated some devastation. The 1970s saw a shift toward computer models, improving storm predictions. Today, AI is again reshaping this field, but concerns arise over its reliability in forecasting unprecedented weather events—termed the “gray swan” problem.
Researchers point out that while extreme weather events are becoming more common due to climate change, AI models primarily learn from historical data, which often excludes these rare occurrences. For instance, a recent study indicated that AI struggles to predict strong hurricanes when trained on datasets lacking such categories. Experts warn that AI could confidently predict ordinary weather while failing to warn of significant threats.
Despite these risks, AI models are being embraced for their speed and low cost compared to traditional methods. Their accuracy is improving rapidly; during the 2025 Atlantic hurricane season, Google’s model outperformed many traditional ones. AI is particularly beneficial in developing regions where resources for traditional forecasting are limited, as shown in initiatives providing farmers in India with advanced monsoon forecasts.
However, the rapid deployment of AI without adequate testing poses dangers, especially in climate-vulnerable areas. Calls have emerged for more rigorous testing frameworks to evaluate AI models for extreme weather predictions. Developing strategies to enhance AI’s capability to forecast rare events, while ensuring it works alongside physical models, is crucial. Experts stress that despite current limitations, the continued evolution of AI in meteorology is essential for adapting to a changing climate.
Source link
