Extreme rainfall attribution distorted by structural warming biases in climate models

While the performance of climate models in simulating the magnitude of global warming has been extensively assessed, their fidelity in representing the three-dimensional (3-D) structure of warming, and how this affects extreme event attribution, remains poorly understood. Pseudo-global-warming experiments implicitly assume that imposed anthropogenic warming perturbations realistically capture the observed vertical and horizontal distribution of atmospheric temperature change. However, this assumption is rarely evaluated explicitly.

We diagnose 3-D structural warming discrepancies by comparing a representative set of six CMIP6 climate models against ERA5 temperature trends over 1940–2024. We show that widely used models exhibit systematic vertical and horizontal warming biases, typically over-amplifying warming in the mid-to-upper troposphere while damping the response near the surface, particularly across Northern Hemisphere mid-latitudes. We further show that these structural biases propagate into substantially different estimates of extreme rainfall intensification.

Using an ensemble of 81 high-resolution MPAS simulations within a storyline attribution framework, we analyze the October 2024 Valencia flood-producing storm as a high-impact case study. The diagnosed anthropogenic rainfall signal is highly sensitive to the 3-D structure of the imposed warming: CMIP6-based counterfactual experiments yield weak reductions in extreme rainfall (~10%), whereas observation-constrained warming profiles produce a stronger and more significant anthropogenic contribution (~30%). This amplification arises from enhanced low-level moistening and increased convective instability, together with dynamically consistent upper-level flow strengthening. The results confirm that 3-D warming structure is a first-order control on extreme-rainfall attribution, and that persistent model-structural errors can lead to a systematic underestimation of attribution signals in mid-latitude, high-impact precipitation extremes.