Water vapour: feedback or forcing?
First some basics. Long-wave (or thermal) radiation is emitted from the surface of the planet and is largely absorbed in the atmosphere. Water vapour is the principle absorber of this radiation (and acknowledged as such by everybody). But exactly how important is it? In terms of mass, water vapour is much more prevalent (about 0.3% of atmospheric mass, compared to about 0.06% for CO2), and so is ~80% of all greenhouse gases by mass (~90% by volume). However, the radiative importance is less (since all molecules are not created equal). One way to quantify this is to take a radiation model and remove each long-wave absorber (principally the greenhouse gases, but also clouds and aerosols) and see what difference it makes to the amount of long-wave absorbed. This gives the minimum effect from each component. The complementary calculation, using only each particular absorber in turn, gives the maximum effect. Generally these will not be equal because of overlaps in the absorbing spectra (i.e. radiation at a particular frequency can either be absorbed by water vapour or CO2).
|Removed absorbers||Fraction LW||Rad. Forcing|
|H2O||64 (64, RC78)||-56|
|Clouds||84 (86, RC78)||-|
|CO2||91 (88, RC78)||-23|
|O3||97 (97, RC78)|
|All except H2O+Clouds||85||-|
|All except H2O||66 (60-70, IPCC90)||-|
|All except CO2||26 (25, IPCC90)||-|
|All except O3||7||-|
|All except Other GHG||8||-|
|Instant calculation, global mean, Jan. 1, 1979||RC78=Ramanathan and Coakley (1978)|
|'All' includes aerosols, O3 and other minor gases as additional absorbers.|
The table shows the instantaneous change in long-wave aborption when each component or combination of components is removed using the radiation code from the GISS GCM. (The source code is available for those who have the patience to get it to work). This isn't a perfect calculation but it's quick and easy and is close enough to the right answer for our purposes. (N.B. This is very similar to what was done by Ramanathan and Coakley (1978) using a single column model - their numbers are in the table for reference). Because of the overlaps, the combined changes are larger than the changes due to each individual component. Another calculation is the instantaneous radiative forcing at the tropopause, but that is complicated for clouds, O3 and Aerosols which have impacts on solar radiation as well as the long wave, so I only give that value for the 'pure' greenhouse gases.
The overlaps complicate things, but it's clear that water vapour is the single most important absorber (between 36% and 66% of the greenhouse effect), and together with clouds makes up between 66% and 85%. CO2 alone makes up between 9 and 26%, while the O3 and the other minor GHG absorbers consist of up to 7 and 8% of the effect, respectively. The remainders and uncertainties are associated with the overlaps which could be attributed in various ways that I'm not going to bother with here. Making some allowance (+/-5%) for the crudeness of my calculation, the maximum supportable number for the importance of water vapour alone is about 60-70% and for water plus clouds 80-90% of the present day greenhouse effect. (Of course, using the same approach, the maximum supportable number for CO2 is 20-30%, and since that adds up to more than 100%, there is a slight problem with such estimates!).
Since we are looking at the whole of the present-day greenhouse effect (around 33 C), it is not surprising that the radiative forcings are very large compared to those calculated for the changes in the forcing. The factor of ~2 greater importance for water vapour compared to CO2 is consistent with the first calculation.
So where does the oft quoted "98%" number come from? This proves to be a little difficult to track down. Richard Lindzen quoted it from the IPCC (1990) report in a 1991 QJRMS review* as being the effect of water vapour and stratiform clouds alone, with CO2 being less than 2%. However, after some fruitless searching I cannot find anything in the report to justify that (anyone?). The calculations here (and from other investigators) do not support such a large number and I find it particularly odd that Lindzen's estimate does not appear to allow for any overlap.
While water vapour is indeed the most important greenhouse gas, the issue that makes it a feedback (rather than a forcing) is the relatively short residence time for water in the atmosphere (around 10 days). To demonstrate how quickly water reacts, I did a GCM experiment where I removed all the water in the atmosphere and waited to see how quickly it would fill up again (through evaporation from the ocean) . The result is shown in the figure. It's not a very exciting graph because the atmosphere fills up very quickly. At Day 0 there is zero water, but after only 14 days, the water is back to 90% of its normal value, and after 50 days it's back to within 1%. That's less than 3 months. Compared to the residence time for perturbations to CO2 (decades to centuries) or CH4 (a decade), this is a really short time.
Only the stratosphere is dry enough and with a long enough residence time (a few years) for the small anthropogenic inputs to be important. In this case (and in this case only) those additions can be considered a forcing. Oxidation of anthropogenic methane (which is a major source of stratospheric water) and, conceviably, direct deposition of water from increases in aircraft in the lower stratosphere, can increase stratospheric water and since that gives a radiative forcing effect, they do appear on the forcings bar chart (under "H2O from CH4"). Some scientists have argued that changes to irrigation and other land use changes (which effect evaporation) are also direct forcings to water vapour amounts, but I think it's cleaner to think of that as an indirect water vapour response to the change.
When surface temperatures change (whether from CO2 or solar forcing or volcanos etc.), you can therefore expect water vapour to adjust quickly to reflect that. To first approximation, the water vapour adjusts to maintain constant relative humidity. It's important to point out that this is a result of the models, not a built-in assumption. Since approximately constant relative humidity implies an increase in specific humidity for an increase in air temperatures, the total amount of water vapour will increase adding to the greenhouse trapping of long-wave radiation. This is the famed 'water vapour feedback'. A closer look reveals that for a warming (in the GISS model at least) relative humidity increases slightly in the tropics, and decreases at mid latitudes.
How do we know that the magnitude of this feedback is correctly simulated? A good test case is the response to the Pinatubo eruption. This caused cooling for up to 3 years after the eruption - plenty of time for water vapour to equilibriate to the cooler sea surface temperatures. Thus if models can simulate the observed decrease of water vapour at this time, it would be a good sign that they are basically correct. A good paper that demonstrated this was Soden et al (2002) (and the accompanying comment by Tony DelGenio). They found that using the observed volcanic aerosols as forcing the model produced very similar cooling to that observed. Moreover, the water vapour in the total column and in the upper troposphere decreased in line with satellite observations, and helped to increase the cooling by about 60% - in line with projections for increasing greenhouse gases.
To be sure there are still some lingering uncertainties. Some recent data indicates that tropical upper tropopsheric water vapour does not quite keep up with constant relative humidity (Minschwaner and Dessler, 2004) (though they still found that the feedback was positive). Moist convection schemes in models are constantly being refined, and it's possible that newer schemes will change things . However, given the Pinatubo results, the models are probably getting the broader picture reasonably correct.
*R.S. Lindzen, 1991. Quart. J. Roy. Met. Soc., 117, pp. 651-652