Climate modelling is one of the most remarkable scientific achievements of the last century. The basic physics of greenhouse gas forcing was described in the 19th century. The first credible numerical climate models appeared in the 1960s. And their broad predictions — global warming proportional to CO2 concentration, polar amplification, shifting precipitation patterns — have been confirmed by observed data over decades.
Where the Models Were Right
The 1988 congressional testimony of NASA scientist James Hansen projected three scenarios of future warming based on different emissions trajectories. Reviewing those projections today, in light of actual observed temperatures, shows they were broadly accurate — within the range of natural variability. This is striking given the computing power available in 1988 compared to today.
More recent models have been similarly accurate on large-scale temperature trends, ocean heat uptake, and sea level rise from thermal expansion. The fundamental physics that drives these models is well-understood, and the models represent it well.
Where the Models Underestimated
The failures are concentrated in areas where the physics is more complex and where feedback effects dominate. Arctic sea ice loss has occurred faster than most models projected. Extreme precipitation events have intensified more rapidly than expected. Ice sheet dynamics — the behaviour of the Greenland and Antarctic ice sheets under warming conditions — remain poorly modelled because they involve processes that are difficult to parameterise at climate-model resolution.
The pattern is concerning: where models are most uncertain, the observed changes tend to be worse than the central projections. This is not universal — some projections have been too high. But the direction of surprise has more often been toward greater-than-expected change rather than less.
