Using observations of cloud regimes to assess cloud radiative feedbacks in current GCMs

We have extended the cloud-clustering analysis of the CMIP5 models in Tsushima et al. (2012) to global warming experiments to investigate whether there is any relationship in the properties of cloud regimes between their present-day annual-mean climatologies and their responses to climate change. Data from two CMIP5 experiments, amip and amipFuture are used. We applied the method to the six models currently available with the required data. We take correlations between the annual mean climatology in the control climate and the future response for various cloud properties. Significant correlations are found between the inter-model spread in the control climate and that in the future response of the in-regime net cloud radiative effect (NCREO) of the stratocumulus regime and the frequency of occurrence of the frontal regime. The stratocumulus regime is too reflective in models in the control climate. In the future climate, models with larger biases show the NCREO becoming less negative whereas models with smaller biases in the NCREO show the NCREO becoming more negative. In all models, the frontal regime is too frequent in the control climate, and it becomes more frequent in the future. The larger the bias in the control climate, the larger the increase in the future. We use a linear regression line to read off the responses which have control values consistent with the observations. The observations suggest a more negative future NCREO response in the stratocumulus regime. They also suggest a smaller increase in the frequency of occurrence of the frontal regime. The NCRE response of each cloud regime in the future depends in part on the change in the frequency of occurrence and the change in the NCREO of the regime as well as their control values. Constrained estimates of the NCRE response can be made for the stratocumulus regime and the frontal regime, using the observed control value and the future response which is consistent with the observations. For the stratocumulus regime, the NCRE responses from the original model simulations tend to be positive or close to zero. The observationally constrained estimates suggest values which are less positive or close to zero. For the frontal regime, the original NCRE responses from the simulations are all negative. The observationally constrained responses are much closer to zero. In most models, the observationally constrained total NCRE responses from all regimes are more positive than the original simulations (+0.28Wm-2 in model mean). This is mainly due to the constraint on the frontal regime rather than the stratocumulus regime.