Convectively Coupled Equatorial Waves dominate tropical weather on 2-30 day timescales, driving episodes of heavy precipitation across the equatorial belt. Climate models must represent them to capture tropical variability. The standard approach: parameterize convection — replace the turbulent details of individual thunderstorms with a smooth statistical scheme that removes vertical instability without resolving the convective plumes themselves.
McKinnon-Gray and colleagues compared two Met Office simulations at identical km-scale resolution. One parameterized convection. One resolved it explicitly. The difference was not just accuracy. It was energy.
The explicitly convective simulation produced approximately 50% more upscale energy transfer in the upper troposphere. The waves were more coherent — phase-speed variability only 80% above observations, versus 166% for the parameterized version. And within the waves themselves, the correlation between interscale energy transfer and equatorial wave winds was stronger.
The mechanism: convection is not just a source of heating. It is a pump. Deep convective plumes generate resolved-scale circulations that transfer kinetic energy from small scales (4-8 times smaller than the wave wavelength) to the wave itself. Active deep convection within a particular sector of the wave phase drives this upscale transfer. The convective plume doesn't merely respond to the wave — it feeds it.
Parameterization removes vertical instability. That is its job. But by smoothing over the resolved-scale circulation that accompanies each plume, it also removes the pump. The energy that should cascade upward into tropical waves instead dissipates at the parameterization scale. The scheme is not a neutral approximation of convection. It is a selective filter that passes the heating while blocking the mechanical forcing.
The tropical atmosphere looks different in parameterized models not because the models lack resolution but because they lack a mechanism. The energy budget is not a rounding error that better tuning could fix. It is a structural consequence of replacing a process with its statistical summary. The summary captures the mean effect but discards the correlations that drive the transfer.
Parameterization is an energy sink disguised as an approximation.