AMIP

AMIP is an experiment protocol in which an atmospheric model is run with observed sea surface temperatures and sea ice prescribed at the lower boundary, rather than coupled to an ocean model. It began as the Atmospheric Model Intercomparison Project and the name now refers to the configuration as much as the project.

The purpose is attribution of error. A coupled model that gets tropical rainfall wrong may have an atmospheric problem, an oceanic problem, or a coupling problem, and the three are hard to separate. Prescribing the observed ocean removes two of the three. If the atmosphere still gets rainfall wrong with the correct SSTs, the atmospheric model is responsible. This makes AMIP a diagnostic configuration rather than a predictive one, and its value is almost entirely in that separation.

The setup is also the natural one for comparing atmospheric models across institutions, since every model sees an identical lower boundary and differences in output are attributable to the models rather than to their oceans.

The limitation that matters

Prescribing SST removes the ocean’s ability to respond, and in the tropics that changes the physics rather than merely constraining it.

Over much of the tropical ocean the atmosphere drives the SST as much as the reverse: wind speed controls evaporative cooling, cloud cover controls insolation. Prescribing SST breaks that feedback, and the resulting relationship between SST and rainfall differs systematically from the coupled case. AMIP runs are known to misrepresent the covariance between the two, and to handle intraseasonal variability such as the MJO poorly for this reason. It is not a subtle effect and it is specifically worst in the tropics — the region a Galapagos-focused study cares about.

So AMIP answers “can the atmospheric model produce the right response to a known ocean state” and not “does the atmosphere behave correctly”, and the gap between those questions is largest exactly where I work.

Relation to regional modelling

The logic carries directly to regional climate modelling, where prescribed boundary conditions play the same role: dynamical downscaling driven by ERA5 is AMIP’s reasoning applied at the lateral boundaries instead of the surface. The same trade applies — constraining the driving fields isolates the model’s own behaviour, and forbids it from feeding back on what drives it.

See also: regional climate modeling, sea surface temperature, ERA5, and dynamical downscaling.