Estimating Climate Sensitivity: Report of a Workshop (2003)

Eugene Rasmusson, from the University of Maryland, provided a brief overview of an ongoing National Academies’ study Climate Change Feedbacks: Characterizing and Reducing Uncertainty. This study is being carried out by a panel of the Climate Research Committee and is sponsored by NOAA, the National Science Foundation (NSF), the Department of Energy (DOE), and the National Aeronautics and Space Administration (NASA). The specific tasks of the committee are described as follows.

He explained that the study was still undergoing peer review, and thus the committee’s findings and recommendations could not be discussed at the workshop. However, this topic is obviously of great relevance to those interested in quantifying climate sensitivity, and thus the workshop participants were urged to contact BASC if they are interested in being notified about the report release.

William Collins, from the National Center for Atmospheric Research, discussed cloud feedbacks. NCAR and GFDL models yield substantially different estimates for S_eq (2.2 °C for NCAR, 3.78-4.14 °C for GFDL). This disparity has provided a great learning opportunity, particularly with regard to understanding the models’ differences in cloud feedbacks.

NCAR and GFDL scientists ran a series of experiments in which the climate system response is decomposed into components associated with individual atmospheric fields (e.g., atmospheric moisture, temperature lapse rate, surface albedo, cloud amount) and then the climate feedback associated with each of those changes was quantified. In these experiments, they changed the global average ocean surface temperature and examined the resulting shortwave and longwave feedbacks (Coleman and McAvaney, 1997).

It was found that the NCAR and GFDL models have very similar clear-sky feedbacks (i.e., changes in atmospheric water vapor and temperature lapse rate), but there are big differences in the longwave feedback parameter. The NCAR model had a strong negative cloud feedback (two to three times larger than the GFDL model), which was driven primarily by shortwave forcing from increasing cloud cover and liquid water content in the lower atmosphere. Positive cloud feedbacks from infrared absorption in high clouds were weaker in the NCAR model than in the GFDL model.

Another class of NCAR model experiments was carried out to look at cloud feedbacks in a transient integration of a 1 percent annual increase in CO₂ (ranging from 2*CO₂ to 4*CO₂). They found a steady increase in the longwave trapping associated with increasing high clouds (a positive feedback), but the dominant feedback is still associated with the shortwave. Now the challenge is to understand this result and compare it to results from the GFDL model.

Observational studies may confirm the idea of negative forcing from increasing low cloud cover. In a global survey of cloud cover taken from shipboard observations, Norris et al. (1999) found an increase in global

average low cloud cover of 3.6 percent between 1952 and 1995. It is not clear what is driving this change; it could potentially be related to aerosol indirect effects.

It appears that the NCAR model may have a problem with predicting cloud optical properties in the upper troposphere and thus be missing additional negative feedbacks from tropical high clouds. Once this bias is corrected for, the model probably will end up with an even stronger net negative cloud feedback.

To address these uncertainties, we need a multifaceted approach to testing models against observations. Two strategies for future diagnosis of cloud parameterizations may involve (1) compositing observed response of low clouds to different meteorological conditions and (2) analysis of high cloud characteristics using Webb-Klein International Satellite Cloud Climatology Project (ISCCP cloud) simulations.⁵

Venkatchalam Ramaswamy discussed the results of a study to examine water vapor feedbacks in the Hadley and GFDL models, in comparison with available observations (Allan et al., 2002). Both models were input with the same reported sea surface temperatures and run over the specified period. The two models were quite consistent with regard to interannual changes in column water vapor and clear-sky outgoing longwave radiation. However, they appear to have given the same result for different reasons. The GFDL model has a stronger positive lapse rate feedback, while the Hadley model had a stronger water vapor feedback. So there was compensation between these two effects, yielding a similar net sensitivity. The differences between the GFDL and Hadley models have not been evaluated quantitatively. They are probably due primarily to different convection schemes, but possibly to different treatments of advection of water vapor. Regardless, it points to the fact that two models can give the same value for climate sensitivity, for different reasons.

Ramaswamy’s GFDL colleagues evaluated an Atmospheric Model Intercomparison Project (AMIP)⁶ study of the ability of atmospheric models to simulate interannual variability in tropical hydrological cycle intensity (30 models were included in the study) (Boyle, 1998). The models gave a reasonable simulation of variations in temperature, water vapor, and outgoing longwave radiation, but they underpredicted precipitation variations by a factor of four. Although there are substantial uncertainties associated with the satellite observations themselves, this disparity suggests that the models may be missing some process that causes variations in tropical organized convection and evaporation. This problem must be reconciled in order to have confidence in projecting future changes in hydrological cycle intensity.

Michael Prather, from the University of California, Irvine, discussed feedbacks related to changes in atmospheric composition. The observational record of increasing trace gas concentrations makes it clear that current atmospheric composition is significantly different from the pre-industrial era (e.g., see greenhouse gas atmospheric concentration time series in Figure 4, pg. 30), and it is likely that future atmospheric composition will be much different than today. Thus, we must be careful about extrapolating from one era to another and calibrating models to paleo-era data records.

Paleo-records (such as the Vostok ice core records) make it clear that natural feedback processes can lead to large changes in atmospheric concentrations of CO₂, CH₄ (methane), N₂O (nitrous oxide), and other trace gases. In such a context, these changes can be viewed as climate feedbacks, rather than forcings. The relationships

between temperature and trace gas concentrations seem to vary substantially in different time periods and in different paleo-records. Thus, one cannot simply scale the trace gas concentrations with temperature.

Even today we still do not fully understand what is driving interannual-decadal variations in the growth rate of CO (carbon monoxide), CO₂ and CH₄, observed over the past few decades. It is particularly difficult to quantify and interpret observed trends and patterns in tropospheric ozone. These trace gas changes probably result from some combination of human-driven sources and sink changes, spotty data coverage or quality, and short time scale natural feedbacks including coupled chemical interactions among CH₄, CO, OH (hydroxyl radical), O₃, VOCs (volatile organic compounds) and other trace species.

IPCC/TAR examined future scenarios of changes in atmospheric composition and chemistry. In the high-emission scenarios, significant regional increases in surface ozone were projected. However, other studies project that the warmer, wetter atmosphere resulting from climate change will tend to increase the destruction of ozone in the troposphere (Fuglestvedt et al., 1995; Brasseur et al., 1998). Thus, the net impacts of ozone concentration are very difficult to project with confidence. Scientists are just starting to understand these complex climate-chemistry feedbacks (including biogenic feedbacks), and these processes are not yet included in the IPCC climate projections.

Another important set of feedbacks that deserves more attention involves anthropogenically driven changes in hydrology and freshwater transport. Today, most freshwater and riverine systems in the world are heavily managed. This leads to changes in river discharge rates, which affects biogeochemical cycling, sea surface temperature, and salinity and thus could even potentially affect North Atlantic deepwater formation processes.

Socci: Is the issue of methane hydrate releases a serious possibility?

Prather: I have not seen much convincing evidence for this being a major factor, although this doesn’t rule out the idea that it could be important.

Mahlman: What about feedbacks related to stratospheric ozone?

Prather: This is worth careful consideration because the patterns of stratospheric ozone depletion will change over the course of the next century. The atmospheric concentration of CFCs (the major driver of ozone depletion in the polar regions) will lessen, but there will be an increase in CH₄ and N₂O-related ozone depletion, which is more globally distributed and has different altitude profiles. We also need to consider the coupling of ozone depletion with climate change. Some studies claim that climate change (and the associated cooling of the stratosphere) will exacerbate stratospheric ozone losses, although the magnitude of this feedback is a matter of debate.

Prather: Another interesting global change to consider is that over the past 20 years, there has been a slow increase in lower stratospheric water vapor. This, by itself, results in another addition to the greenhouse effect. We don’t understand what is driving this trend, but it is presumably related to gradual warming and moistening of the tropical upper troposphere. Unfortunately, the data quality is inadequate for testing this.

Joyce Penner, from the University of Michigan, discussed feedbacks and uncertainties related to aerosol radiative forcing.

Determining climate sensitivity by fitting the observational temperature record requires specifying radiative forcing from aerosols. Most climate models currently consider only forcing from fossil fuel sulfate aerosols. Ultimately, however, one has to consider several other types of atmospheric aerosols, including organic carbon and black carbon from fossil fuels, smoke from biomass burning, and fossil fuel nitrate and associated ammonium. These each have different time histories (Figure 5, pg. 31) and different spatial distributions, thus making it very difficult to determine a net sensitivity and uncertainty range. So how can we assign unbiased estimates of uncertainty associated with aerosol forcing?

In IPCC/TAR (Chapter 5) a “bottom-up” approach was used to estimate uncertainty associated with direct forcing from fossil fuel and biomass-burning aerosols. A simple box model based on what we know about

critical aerosol parameters such as aerosol concentration, size distribution, and composition was employed. This approach yielded the following estimates:

• Fossil fuel direct: -0.6 Wm⁻² (-0.1 to -1 Wm⁻²) Main source of uncertainty: upscatter fraction, burden (with emissions), mass scattering efficiency.

• Biomass smoke direct: -0.3 Wm⁻² (-0.1 to -0.5 Wm⁻²) Main source of uncertainty: single scattering albedo, upscatter fraction, burden (with emissions).

Estimating the indirect effect of aerosols and associated uncertainties poses a far more complex problem, as it involves at least three different processes:

1. The “first indirect effect” (Twomey effect, or cloud albedo effect): changes to cloud albedo associated with changes in droplet number concentration.

2. The “second indirect effect” (Albrecht effect, or cloud lifetime effect): changes to cloud albedo and cloud fraction associated with changes in precipitation efficiency

3. The “semidirect effect” (Hansen effect): rises in local temperature causing changes to cloud amount due to absorbing aerosols (i.e., soot) within the atmospheric column.

The bottom-up estimate of uncertainty in Twomey (the first) indirect forcing from fossil fuel aerosols was -1.4 Wm⁻² (0 to -2.8 Wm⁻²). The main uncertainties in this estimate were cloud height, liquid water content, and the relationship between aerosol number concentration and cloud droplet number concentration. The uncertainty also depends on how the effects of aerosols on cloud droplets are parameterized. For sulfates, it depends on how much is formed through aqueous versus homogeneous chemical means and what the natural background aerosol is in the model.

Uncertainties associated with the second indirect effect were not estimated. In fact, some scientists argue that this should be considered a response (and part of the climate sensitivity) rather than a forcing. Rotstayn and Penner (2001) tested whether the second indirect effect should or should not be considered a forcing. Their approach was to run the model twice, calculating temperature change with only the first indirect effect, and then with both the first and the second effects. They found that if the second indirect effect is not included as a forcing, the climate sensitivity changes by a factor of two, but if the second indirect effect is included as a forcing, the climate sensitivity factor remains nearly constant (i.e., within 20 percent of that from 2*CO₂)

All such estimates of uncertainty are only as good as our theoretical and observational understanding, and some potentially important aerosol forcing effects have not yet been considered. This includes positive forcing from aerosol semidirect effects and from changes in ice concentration (cirrus clouds), which may lead to a substantial positive forcing. Likewise, recent studies have emphasized that biomass-burning aerosols have very different spatial and seasonal patterns than fossil fuel aerosols and cannot be ignored in future climate modeling studies.

The summary points are the following:

• Bottom up estimates of uncertainty must be developed to provide unbiased estimates of spatial distribution of uncertainty in forcing.

• Efforts should be made to extend the uncertainty estimates to temporally varying estimates of both direct forcing and first indirect forcing.

• Substantial variations apparently exist between different model estimates of the effects of black carbon and organic matter.

• There is conflicting evidence regarding the importance of the second indirect effect based on combined observation and modeling studies, but including this factor could change results substantially.

• Feedbacks due to the effects of climate change on aerosol sources (e.g., impacts on dust, sea salt) have not been included in the climate models.

Andronova: If biomass burning aerosols were included in simple climate models, then aerosol radiative forcing patterns would become more symmetric (since biomass burning aerosols are not concentrated in the northern midlatitudes like the fossil fuel aerosols). The result would be lower inferred values for

sensitivity, but remains uncertain due to the still unquantified net radiative effects of carbonaceous aerosols.

Mahlman: The high indirect forcing estimates are thought to be physically implausible because they would suggest that the climate of the past 40 years should have been cooling.

Schlesinger: Agreed, the inferred large negative aerosol radiative forcing is hard to reconcile with temperature observations. If total aerosol radiative forcing is as large as -2.8 Wm⁻², this implies that sensitivity must be huge.

Penner: Models are probably overestimating the large negative forcing, but this might be offset by positive forcing from ice clouds, which is not yet included in any model estimates. Modeling studies that fit the observational record may give a reasonably accurate net forcing, but if they are not capturing the spatial and temporal pattern of forcing, you will not have an accurate representation of the true forcing or the capability to predict the future.

Ramaswamy: Models indicate that a large percentage of aerosol radiative forcing occurs over the oceans. Do these models include interactions between sea salt and pollution aerosols?

Penner: Only one model includes this effect, and the results are difficult to understand.

Mahlman: Another huge uncertainty is the role of large-scale atmospheric dynamical phenomena that produce widespread thick clouds. These dynamical effects could overwhelm the indirect effects on cloud physics.

Jonathan Gregory discussed feedbacks related to sea ice and oceanic processes. Sea ice covers only 5 percent of the earth’s surface but may have a disproportionately large influence on global climate sensitivity, as the main driver of the large polar amplification of climate change (i.e., the large projected warming at high latitudes). Reduced sea ice affects sensitivity in two distinct ways: (1) the local warming effect, which occurs primarily in winter when the ice insulates the ocean from the atmosphere, and (2) the albedo effect, which occurs only in the summer when there is sunlight.

Northern hemisphere sea ice cover has been declining over the past few decades (~2.5 percent per decade in real annual-average sea-ice cover). Projections of how arctic sea ice will be affected by future climate change vary greatly from model to model. There are many different aspects of parameterizing sea ice in models that could be responsible for these differences. In the Hadley Centre’s third-generation coupled ocean-atmosphere GCM (HadCM3), sea ice cover declines at 13 percent per Kelvin of global warming, and in some “high-end” projections, summer sea ice disappears entirely by the end of the century.

In all models there seems to be a fairly linear relationship between rise in global average temperature and reduction in the aerial extent of sea ice cover. This relationship is scenario independent, which implies that sea ice responds rapidly to climate warming and thus provides a useful rapid feedback diagnostic. There is also a scenario-independent relationship between declining volume and declining area, but this relationship is not linear. The volume initially declines about twice as fast as the area, but the volume-area relationship tends to flatten out as the ice decreases.

The reason oceans are important for climate change projections is because of the long time scales required to reach equilibrium and the possibility of nonlinear responses (e.g., disruption of thermohaline circulation or other rapid, irreversible changes). On shorter time scales, the important questions are related to oceanic heat uptake, a process that effectively mitigates global warming by taking up some of the heat that would otherwise increase atmospheric temperatures. Oceanic heat uptake effectively sets the rate of climate change. There is substantial variation among models in heat uptake efficiency. Many processes involved in heat uptake that have to be parameterized in models are not known with certainty. In scenarios in which the radiative forcing is increasing with time, it is found that the heat flux into the ocean is roughly proportional to global average temperature change and therefore looks like a negative feedback term in sensitivity. However, this linear relationship holds only in the nearer time scales; as equilibrium is approached, the heat uptake goes to zero, but the temperature increase does not.

Broccoli: One complication of modeling ice albedo effects is that during the summer, sea ice is often masked by clouds, which effectively reduces the albedo feedback.