Artificial intelligence is rapidly becoming a core tool in climate science, helping researchers forecast weather, monitor greenhouse gas emissions, and process vast streams of satellite data far faster than traditional methods allow. But as ten experts interviewed by Carbon Brief make clear, the technology brings significant challenges alongside its promise — including the risk of overreliance on unverified results, heavy energy consumption, and the potential to deepen existing inequalities between research institutions around the world.
Perhaps the most visible application so far is in weather prediction. AI models that learn from large volumes of historical weather data can now generate forecasts up to two weeks ahead while using a fraction of the computing power required by conventional numerical models. The European Centre for Medium-Range Weather Forecasts has already incorporated an AI forecast system into its official predictions alongside traditional models. Dr Doug McNeall, science lead in data-driven climate modelling at the UK Met Office, described the progress as significant, but cautioned that climate projection is fundamentally different from short-range forecasting. “To provide useful climate information, we need to make projections for future climates that lie outside the observational record,” he said, pointing to hybrid approaches — combining AI with physics-based models — as a promising path forward. Dr Amirpasha Mozaffari of the Barcelona Supercomputing Centre added a further caveat: AI forecast models tend to underperform on record-breaking extremes, “precisely the events that matter most for adaptation.”
Beyond forecasting, AI is being applied across a wide range of climate research tasks. Satellite observation has generated enormous volumes of geospatial data, and AI algorithms are now used to scan that imagery automatically — detecting illegal logging, tracking glacier retreat, monitoring land-use change, and pinpointing industrial methane leaks in near real-time using hyperspectral data from orbit. In sea ice research, machine learning tools are used throughout the modelling workflow, from classifying snow layers to predicting sea ice concentration and drift. An early deep-learning system called IceNet was shown to outperform leading physics-based models in seasonal Arctic forecasts, demonstrating that AI could compete with established scientific approaches in specialised domains.
Despite the momentum, experts identified a fundamental obstacle for climate — as opposed to weather — applications: the absence of observational data for the future. “We’re literally trying to predict the future. We don’t have observations of it,” said Prof William Collins of the University of East Anglia. That gap makes purely data-driven AI approaches harder to validate in climate contexts, where model outputs may describe conditions with no historical analogue. Researchers are increasingly looking at hybrid models that embed physical constraints to ensure forecasts remain grounded in scientific principles, rather than relying on pattern-matching alone.
The broader picture, as experts see it, is one of genuine transformation tempered by unresolved questions. AI is accelerating what climate scientists can do, making some forms of analysis faster, cheaper, and more accessible. But the field is still grappling with how to ensure that speed and accessibility do not come at the cost of scientific rigour — or widen the gap between well-resourced and under-resourced research communities.