Why weather skill needs a separate climate argument

Weather and climate share physical systems, but their evaluation questions differ. A forecast asks about the evolution of a particular state over a specified horizon. A climate claim may concern distributions, rare events, or responses to changing external conditions. Success on the first task leaves additional questions for the second.

GraphCast is a useful reference for examining learned weather prediction. Reading such work carefully means retaining the forecast horizon, variables, reference data, and comparison methods. A headline about forecasting skill becomes scientifically useful when those details remain attached to it. Removing them invites an unsupported expansion of the claim.

For climate use, the forcing conditions deserve explicit treatment. Does the model represent the variables that change between scenarios, and has its response to those changes been evaluated? A model trained to continue observed weather states may face a different problem when asked to characterize conditions outside that historical range.

Time scale also changes the test. A sequence of locally reasonable predictions may accumulate drift or settle into an unrealistic distribution. Evaluation should examine the statistics relevant to the proposed claim, including seasonal patterns and tails where appropriate. Average error alone cannot establish that every important feature of a distribution is preserved.

Reference data add another layer. Observations, reanalyses, and numerical simulations each have a role, but agreement with one should be named precisely. A comparison can support performance relative to a particular reference without resolving all uncertainty about the physical system. This matters especially when the same data lineage influences training and evaluation.

The sensible bridge from weather to climate is a new validation argument. Specify the climate quantity, the conditions under which it is estimated, and the tests that could expose a misleading answer. A capable weather model may contribute to that work. Its contribution becomes clearer when the additional claim is evaluated on its own terms.