The Bhola cyclone, which struck East Pakistan on November 12, 1970, resulted in an estimated death toll of between 300,000 and 500,000 individuals, making it the most lethal tropical storm recorded. With winds reaching up to 130 miles per hour and a storm surge of 35 feet, its impact was catastrophic. Today, experts reflect on the advancements in weather forecasting since that time, particularly the significant improvements made during the 1970s with the introduction of physics-based computer models.
Currently, the evolution of forecasting techniques is being challenged by the emergence of artificial intelligence, which some experts believe may struggle with predicting rare, unprecedented weather events, termed “gray swans.” These phenomena, while physically plausible, are infrequently represented in the datasets used to train AI models, raising concerns as climate change increases the occurrence of such extremes.
Researchers, including Pedram Hassanzadeh from the University of Chicago, have highlighted the limitations of AI in predicting rare weather events. Their studies indicate that when AI models are tasked with forecasting Category 5 hurricanes after excluding all major storms from their training, they fail to extrapolate accurately to unseen conditions. This could lead to misleading forecasts during critical weather events, as noted by Rose Yu from the University of California San Diego.