Constraining aerosol–cloud radiative forcing using present-day observations

The anthropogenic perturbation to cloud droplet number concentration (ΔlnN_d) can be derived from the N_d-to-aerosol sensitivity in the present day (PD) (ß_PD), and the anthropogenic perturbation to aerosol from the pre-industrial (PI) to PD. A key assumption in this process is that the PD aerosol-N_d relationship is indicative of the actual sensitivity of N_d to anthropogenic perturbation to aerosol, i.e., the PI-to-PD sensitivity (ß_PI-PD). This assumption holds true only when using the cloud condensation nuclei at cloud base (CCN_b) as the CCN_b-N_d relationship is not dependent on aerosol regime. However, due to the difficulty in obtaining CCN_b at a large scale, in practice one has to use proxies for the CCN_b, which makes the above assumption less likely to hold. By combining multiple satellite observations, reanalysis, and AeroCom simulations, this study evaluates the performances of all existing proxies, and then constrain the radiative forcing from aerosol–cloud interactions (RF_aci) by selecting ‘good’ proxies.

To assess whether a proxy-N_d relationship is aerosol-regime dependent, we propose a 'hemispheric contrast' approach, using the Northern and Southern Hemispheres to mimic the PD and PI aerosol environments, respectively. Under the same meteorological background, the hemispheric contrast in N_d at a certain aerosol amount serves as a measure of the aerosol-regime dependency. The results show that aerosol optical depth (AOD) exhibits the strongest dependency, followed sequentially by aerosol index (AI), sulfate burden (SO4_C), surface sulfate mass (SO4_S) and CCN burden (CCN_c), and finally surface CCN (CCN_S).

We further calculate the biases in RFaci caused by using ß_PD instead of ß_PI-PD in an ideal model world, based on the AeroCom model outputs. The results suggest that CCN_S has the smallest bias (+3%), followed by AI, SO4_S and CCNc with positive biases of ~+25%. The AOD and SO4_C show the largest biases, with values of –60% and +80%, respectively.  Assuming CCNs would give the true RF_aci, the biases in observation-based RF_aci can be thus inferred, which turn out to be of similar magnitude to those in the model world. This gives us the confidence that true RF_aci is likely around –0.68 W m-2 (ocean only).