2.4.5.3 Estimates of the Radiative Forcing from Observations and Constrained Models

It is difficult to obtain a best estimate of the cloud albedo RF from pre-industrial times to the present day based solely on observations. The satellite record is not long enough, and other long-term records do not provide the pre-industrial aerosol and cloud microphysical properties needed for such an assessment. Some studies have attempted to estimate the RF by incorporating empirical relationships derived from satellite observations. This approach is valid as long as the observations are robust, but problems still remain, particularly with the use of the aerosol optical depth as proxy for CCN (Feingold et al., 2003), droplet size and cloud optical depth from broken clouds (Marshak et al., 2006), and relative humidity effects (Kapustin et al., 2006) to discriminate between hydrated aerosols and cloud. Radiative forcing estimates constrained by satellite observations need to be considered with these caveats in mind.

By assuming a bimodal lognormal size distribution, Nakajima et al. (2001) determined the Ångstrom exponent from AVHRR data over the oceans (for a period of four months), together with cloud properties, optical thickness and effective radii. The nonlinear relationship between aerosol number concentration and cloud droplet concentration (N_d ≈ (N_a)^b) obtained is consistent with Twomey’s hypothesis; however, the parameter b is smaller than previous estimates (0.5 compared with 0.7 to 0.8; Kaufman et al., 1991), but larger than the 0.26 value obtained by Martin et al. (1994). Using this relationship, Nakajima et al. (2001) provided an estimate of the cloud albedo RF in the range between –0.7 and –1.7 W m^–2, with a global average of –1.3 W m^–2. Lohmann and Lesins (2002) used POLDER data to estimate aerosol index and cloud droplet radius; they then scaled the results of the simulations with the European Centre Hamburg (ECHAM4) model. The results show that changes in N_a lead to larger changes in N_d in the model than in observations, particularly over land, leading to an overestimate of the cloud albedo effect. The scaled values using the constraint from POLDER yield a global cloud albedo RF of –0.85 W m^–2, an almost 40% reduction from their previous estimate. Sekiguchi et al. (2003) presented results from the analysis of AVHRR data over the oceans, and of POLDER data over land and ocean. Assuming that the aerosol column number concentration increased by 30% from the pre-industrial era, they estimated the effect due to the aerosol influence on clouds as the difference between the forcing under present and pre-industrial conditions. They estimated a global effect due to the total aerosol influence on clouds (sum of cloud albedo and lifetime effects) to be between –0.6 and –1.2 W m^–2, somewhat lower than the Nakajima et al. (2001) ocean estimate. When the assumption is made that the liquid water content is constant, the cloud albedo RF estimated from AVHRR data is –0.64 ± 0.16 W m^–2 and the estimate using POLDER data is –0.37 ± 0.09 W m^–2. The results from these two studies are very sensitive to the magnitude of the increase in the aerosol concentration from pre-industrial to current conditions, and the spatial distributions.

Quaas and Boucher (2005) used the POLDER and MODIS data to evaluate the relationship between cloud properties and aerosol concentrations on a global scale in order to incorporate it in a GCM. They derived relationships corresponding to marine stratiform clouds and convective clouds over land that show a decreasing effective radius as the aerosol optical depth increases. These retrievals involve a variety of assumptions that introduce uncertainties in the relationships, in particular the fact that the retrievals for aerosol and cloud properties are not coincident and the assumption that the aerosol optical depth can be linked to the sub-cloud aerosol concentration. When these empirical parametrizations are included in a climate model, the simulated RF due to the cloud albedo effect is reduced by 50% from their baseline simulation. Quaas et al. (2005) also utilised satellite data to establish a relationship between cloud droplet number concentration and fine-mode aerosol optical depth, minimising the dependence on cloud liquid water content but including an adiabatic assumption that may not be realistic in many cases. This relationship is implemented in the ECHAM4 and Laboratoire de Météorologie Dynamique Zoom (LMDZ) climate models and the results indicate that the original parametrizations used in both models overestimated the magnitude of the cloud albedo effect. Even though both models show a consistent weakening of the RF, it should be noted that the original estimates of their respective RFs are very different (by almost a factor of two); the amount of the reduction was 37% in LMDZ and 81% in ECHAM4. Note that the two models have highly different spatial distributions of low clouds, simulated aerosol concentrations and anthropogenic fractions.

When only sulphate aerosols were considered, Dufresne et al. (2005) obtained a weaker cloud albedo RF. Their model used a relationship between aerosol mass concentration and cloud droplet number concentration, modified from that originally proposed by Boucher and Lohmann (1995) and adjusted to POLDER data. Their simulations give a factor of two weaker RF compared to the previous parametrization, but it is noted that the results are highly sensitive to the distribution of clouds over land.