Comparison of Short-Term Cloud Feedbacks at Top of the Atmosphere and the Surface in Observations and AMIP6 Models

We compared short-term cloud feedback, defined at the top of the atmosphere (TOA), the atmospheric column (ATM), and the surface (SFC), between observations and models participating in Atmospheric Model Intercomparison Project Phase 6 (AMIP6) for the period 2000-2014. The globally averaged net cloud feedbacks observed at TOA, ATM, and SFC are -0.06 +/- 0.63, -0.17 +/- 0.70, and 0.11 +/- 0.81 W m-2 K-1, respectively. While most models produced TOA cloud feedbacks that agreed with the observations within uncertainty ranges, the intermodel spread at SFC and within ATM was relatively larger. This demonstrates that models are diverse in how their TOA feedback is distributed between ATM and SFC. Because short-term cloud feedback is mainly driven by El Nino-Southern Oscillation (ENSO), the global-mean cloud feedback was further decomposed into components from the ENSO and non-ENSO regions. Results show that cloud feedback in these two regions tends to be inversely related. Compared to observations, almost all models overestimated the longwave cloud feedback in the ENSO region due to the overestimation of cloud amount changes for high-topped clouds. For these models, it is the offset between deviations in ENSO and non-ENSO regions that leads to the overall agreement of global mean with observations. Sensitivity tests show that the main conclusions still hold when alternative kernels are used in estimating cloud feedback. The large spread in cloud feedback among climate models contributes to the largest uncertainty in climate sensitivity. There is an urgent need to utilize observational data to validate cloud feedback in climate models. By using a combination of reanalysis data and satellite measurements for the period 2000-2014, results show the observed cloud feedback at TOA, ATM, and SFC is -0.06 +/- 0.63, -0.17 +/- 0.70, and 0.11 +/- 0.81 W m-2 K-1, respectively. Most models produced TOA cloud feedback that was consistent with observations within the uncertainty range, yet there was larger difference at the surface and within the atmosphere, causing some models to produce values outside the uncertainty range. Further study shows that even well-modeled TOA feedback is due to bias offsetting. The models tend to overestimate longwave cloud feedback in the El Nino-Southern Oscillation (ENSO) region due to an overestimation of high-topped cloud changes, which is partially offset by an underestimation in the non-ENSO region, resulting in a global mean that is close to the observed value. Decomposing cloud response and cloud feedback in this way highlights the deficiency in ENSO-related cloud modeling that would otherwise be hidden by the global average. The intermodel spread of the net short-term cloud feedback is larger at the surface and within the atmosphere than at the top of atmosphere An inverse relationship between cloud feedbacks in the El Nino-Southern Oscillation (ENSO) and non-ENSO regions is found in both observations and AMIP6 models Compared to observations, models tend to overestimate longwave cloud feedback in ENSO regions due to the overestimation of high-topped cloud