Higher Climate Sensitivity Suggested By Multi-Objective Observational Constraints on Tropical Low Cloud Feedback

The cloud-radiative interactions of the tropical marine low-cloud feedback constitute one of the major sources of uncertainty in equilibrium climate sensitivity (ECS). Moreover, the magnitude of this feedback depends on the spatial pattern of tropical sea surface temperature (SST) which evolves during global warming. Specifically, a delayed warming in the tropical eastern Pacific allows low-cloud feedback to grow more positive in the long term.

In this study, we focus on the tropical Atlantic and Pacific which are the two largest marine stratocumulus regions. The observed interannual low-cloud variability in these two basins is used to constrain the long-term tropical shortwave cloud feedback (SWCF) simulated by the latest global climate models (GCMs).

Examining shortwave cloud radiative effect (SWCRE) anomalies as a function of potential cloud controlling factors, we find that a bivariate linear model with local sea surface temperature (SST) and air temperature at 700 mb (T₇₀₀) is effective. This simple linear model captures the dependence of marine low clouds on the tropical SST pattern, as the remote ascent area SST influences low clouds via the weak temperature gradient (WTG) in the tropical lower free troposphere. In the Pacific, the regression coefficients of this linear model agree well among observations, GCMs and the estimated cloud-top entrainment index (ECTEI). However, in the Atlantic, the sensitivity of SWCRE to local SST estimated from observations is higher than GCMs or the value predicted by the ECTEI theory. This discrepancy highlights the need of further research in understudied Atlantic stratocumulus.

Among GCMs, we notice a trade-off between the model-observation bias in the linear models for Atlantic and Pacific clouds. Therefore, a multi-objective approach based on Pareto optimality is introduced to combine the observational constraints from both basins. Using this approach, ~200,000 small model subensembles are systematically generated out of 28 GCMs, and a Pareto optimal set of these subensembles are identified according to their performance in both basins. The uncertainty of this optimal set is estimated from observational uncertainty, and all the subensembles are assigned probability weights based on the distance from the optimal set. We then update the posterior probability distribution of tropical SWCF using a Bayesian method. The multi-objective observational constraint of tropical SWCF reduces its inter-model spread and increases the median value by 71%. Negative tropical SWCF becomes very unlikely, while the probability of SWCF higher than 0.5 Wm^-2K^-1 increases by 4 times. With other feedback components unchanged, our tropical SWCF constraint suggests a higher ECS than the original model ensemble.