Artificial intelligence (AI) models are quickly advancing within the field of weather forecasting. They synthesize historical weather observations and forecasts to come up with a current forecast based on what occurred in the past after similar conditions were present. As an alternative to current numerical weather forecast models, these AI versions are much faster and cheaper to produce.
However, one distinct challenge that these models face is how they might handle conditions in a changing climate. With meteorological records often broken, these models may not well handle forecasts when they are dealing with observations and events that are outside the historical record. If there is no similar observation or weather pattern for the AI model to look at in that record from the past, how will it know what to predict?
That’s the concern of various meteorologists, including several at Colorado State University, who say that extreme weather events may result in very erratic forecasts in the AI systems. Issues like these would need to be overcome before AI systems could replace traditional numerical forecast models, if they ever do. Human experience may instead remain valuable in determining which type to use based on the strength of each.