A key factor in determining future global warming is the reaction of the clouds to rising temperatures: if cloud cover diminishes or clouds reflect less solar radiation they can exacerbate the effect of CO2: by the same token, they can moderate the warming if they become more extensive or more reflective in the future. Since climate models are nor particularly good at simulating clouds, this is the most important single source of uncertainty.
The following figure, taken from one of my favorite papers,Stephens 2005, shows the simulated changes in cloud cover in two simulations driven by the same scenario of increases in atmospheric concentrations of greenhouse gas.
(Figure caption taken directly from Stephens paper) The response of a number of present-day climate models forced by a 1% yr−1 increase of CO2. Shown is the difference of the 20-yr average of the simulation with fixed and increasing CO2. The averages are over years 1961–80; corresponding broadly to the time of a CO2 doubling. To the right are the changes to low clouds averaged over this same period for two models that fall on either end of the projected warming range (courtesy of B. Soden).
The model that displays the larger temperature increase is the one that simulates a stronger reduction of cloud cover; the model with the lowest temperature increase simulations an increase of cloud cover. Between these two models, the rest of the models show a fairly reasonably inverse relationship between the simulated temperature increase and cloud cover. In a recent paper, Trenberth and Fasullo also point to the the important role that cloud cover changes play for the amplitude of warming simulated by the end of the 21st century in the IPCC climate models:
While there is a large increase in the greenhouse effect from increasing greenhouse gases and water vapor (as a feedback), this is offset to a large degree by a decreasing greenhouse effect from reducing cloud cover and increasing radiative emissions from higher temperatures. Instead the main warming from an energy budget standpoint comes from increases in absorbed solar radiation that stem directly from the decreasing cloud amounts
According to Trenberth and Fasullo, cloud cover tends to decrease in most IPCC models as surface temperature rises. Clouds themselves not only back-scatter solar radiation but are responsible for a large greenhouse effect. They contain water vapor and liquid and frozen water droplets of varying sizes, and therefore they also absorb and emit infrared radiation. These changes offset somewhat the increasing greenhouse effect caused by rising concentrations of CO2 and water vapor in the atmosphere. But, at the same time, the amount of solar radiation reflected back to space by clouds also diminishes, and so they contribute to the simulated warming in the 21st century. It has to be underlined that in the absence of cloud changes in the simulations
warming would still occur and, while it may not be quite as large, it would be manifested somewhat differently owing to the change in solar radiation at the surface
Clouds can modulate the warming, but not offset it.
Clouds are a key factor in modulating anthropogenic warming. How good are climate models are simulating cloud formation and evolution? Unfortunately, not that good, for several reasons. One is that the resolution of climate models is too coarse to properly simulated clouds. Currently, the mesh size in a typical climate models is about 150-200 km, whereas clouds processes play out in much smaller scales of a few hundredths of meters or a few kilometers. This limitation forces climate modellers to try to represent cloud properties, formation and evolution by semi-empirical equations that might not be realistic enough. Also, there are many different types of clouds, as everyone can observe. Each type of cloud has different radiative properties. For instance, ice clouds are quite transparent to solar radiations but are very good absorbers of infrared radiation. Their therefore strongly contribute to the greenhouse effect. boundary layer tropical clouds, on the other hand, mostly reflect solar radiation and have a net cooling effect. Observations of cloud properties on a global basis are difficult to obtain, although nowadays a large effort is being made by remote sensing from satellites and observing aircraft. Stephens points us to an important mismatch between the properties of clouds simulated by climate models and those observed by remote sensing. This particular mismatch is related the so called optical depth, a measure of the the amount of solar radiation that is reflected by a a cloud. Optical depth is influenced by the amount of water contained in a cloud (more water means large optical depth) and by the droplet size (smaller droplets cause a large optical depth). Wetter clouds have in general larger optical depths and thus reflect more solar radiation. In a warmer world clouds are expected to become wetter and thus more more reflective of solar radiation. What Trenberth and Fasullo find is that, despite this increase reflectivity, cloud cover itself diminishes in enough amount for the total affect of cloud changes to reinforce the warming.
Stephens, in a short newsletter article, argues that the changes in cloud reflectivity are too small in climate models and thus the negative feedback of individual clouds is underestimated. This can be understood by looking at the following figure, displaying the relationship between optical depth (for us here roughly equivalent to cloud wetness) and reflectivity.
This relationship is non-linear, meaning that it tends to saturate at higher optical depths: if the optical depth is high, in the range of 40 for instance, the reflectivity cannot increase very much.;If the optical depth is rather in the range of 10, the reflectivity still has some leeway to increase if the optical depth increases as temperature rises. What Stephens finds is that climate models tend to produce clouds in the upper ranges of optical depth, whereas observed clouds are in the lower ranges. These differences cannot be explained by measurement uncertainties and seem to be real model biases. If this is true, the consequence would be that climate models would be underestimating the negative feedback under warming of a typical cloud.
This finding seems to be supported by more comprehensive cloud modelling studies. To by-pass the semi-empirical representations of clouds in climate models, some groups are starting to include full-fledged cloud-submodels embedded in the grid-cells of global climate models - the so called super-parametrization This of course increases enormously the computer requirements. There are, to my knowledge, very few super-parametrization studies. I think the following paragraph taken from one of the most recent ones (Wyant et al is a fair summary of their results so far (GCM=global circulation model):
The climate sensitivity of an atmospheric GCM that uses a cloud-resolving model as a convective superparameterization is analyzed by comparing simulations with specified climatological sea surface temperature (SST) and with the SST increased by 2 K. The model has weaker climate sensitivity than most GCMs, but comparable climate sensitivity to recent aqua-planet simulations of a global cloud-resolving model.
Other groups have tried to establish the sign and magnitude of the cloud feedback by analyzing the co-covariations of temperature and cloud cover, for instance Lindzen and Choi or Clement et al. The target here is to see whether cloud cover increases or decreases in years when temperatures have been higher than normal. The difficulty is that the measured temperature is also affected by the cloud cover itself, so that it is not easy to establish that the changes in cloud cover have been caused by the changes in temperature, or viceversa. More interestingly would be to establish whether periods in the past, as the Little Ice Age, which were colder because the sun was weaker and the CO2 concentrations lower, were also more cloudy (or less cloudy). As you can imagine, the challenge of reconstructing cloud cover in the past centuries or millennia is daunting, but there are hints that some biological proxies are indeed capable of archiving information about past solar radiation at the surface. This could help, hopefully, to reduce the present uncertainties.