Extreme rainfall attribution hinges on warming structure

Climate models are commonly evaluated based on their ability to reproduce global-mean warming, yet their fidelity in representing the three-dimensional (3-D) structure of atmospheric warming—and its implications for extreme event attribution—has received far less attention. Pseudo-global-warming (PGW) approaches implicitly assume that imposed warming perturbations realistically capture the observed vertical and horizontal temperature changes, an assumption that is rarely tested.

Here, we identify systematic, global-scale biases in the vertical and horizontal structure of atmospheric warming in leading CMIP6 climate models relative to observationally constrained reanalysis data. Despite accurately reproducing the integrated magnitude of warming, these models distort its three-dimensional structure. Using a storyline attribution framework, we show that these structural biases propagate into substantially different estimates of extreme rainfall intensification.

We illustrate this effect through high-resolution attribution simulations of the October 2024 Valencia storm (Spain) using the Model for Prediction Across Scales (MPAS). When simulations are forced with an observationally constrained warming signal rather than a CMIP6-derived one, the attributed increase in extreme rainfall roughly triples, driven by enhanced low-level moistening, increased convective instability, and stronger vertical wind shear.  Our results further show that an additional warming of similar magnitude to that observed—plausible under future climate conditions—would lead to an increase in extreme rainfall of around 50%, highlighting the strongly nonlinear response of the system.

Together, our findings reveal a previously unrecognized structural source of uncertainty in attribution science and demonstrate that the 3-D structure of warming is a first-order control on extreme rainfall attribution, potentially leading to systematic underestimation of anthropogenic contributions in high-impact precipitation events.