Recent Land-Ocean Warming Contrast Underestimation Linked to Historical Radiative Feedbacks

Abstract: Models and observations show that surface air over land warms more than over the ocean under greenhouse forcing. However, state-of-the-art CMIP6 models underestimate the magnitude of the warming contrast in recent decades, a bias that has received relatively little attention. Using four observational datasets (HadCRUT5, GISTEMP v4, Berkeley Earth, and NOAA GlobalTemp) and ERA5 reanalysis, we estimate the observed land-ocean amplification factor Φ (defined as the ratio of land and ocean surface air temperature anomalies) to be 1.84 ± 0.05 for 1981-2025. This compares with a CMIP6 multi-model mean ratio of 1.62 ± 0.12 (mean ± intermodel standard deviation) over the same period. Only 3 out of 50 CMIP6 models exceed the mean observational estimate during 1981-2025.

We investigate the origin of this underestimation. We show that the amplification factor Φ in CMIP6 models is anticorrelated with the ratio of the effective radiative feedback over land (λL) and ocean (λO), and that a substantial fraction of the intermodel spread in CMIP6 Φ can be explained by the λL/λO ratio. Physically, a lower value of λL/λO indicates a radiative configuration that favors a larger land-ocean warming contrast, because of less stabilizing feedbacks over land than over the ocean. This relationship provides a framework to compare observations and models in terms of their values of λL/λO. Observations exhibit a relatively low value of λL/λO, which is related to the well-known observed La Niña-like warming pattern that favors stabilizing feedback over the ocean (large |λO|). Most CMIP6 models, on the other hand, do not capture the La Niña-like warming pattern in recent decades, and have a relatively higher value of λL/λO because of a less stabilizing feedback over the ocean. Through the anticorrelation between Φ and λL/λO, we link the model bias in λO (related to the pattern effect) with the underestimation of the observed Φ. Our results suggest that the CMIP6 underestimation of the land-ocean warming contrast may reflect broader challenges in simulating the observed La Niña-like warming pattern.