Karen M. Shell     


Quantifying Feedbacks in Climate Models: Why do Models Project Different Future Climates, and Which is Right?

For a not-very-technical description of some of our work, see this overview.

The sensitivity of the Earth's climate to anthropogenic forcing is one of the most pressing concerns in the earth sciences. If the Earth's sensitivity is low, higher concentrations of greenhouse gases are allowed before a significant change in the climate occurs, while a high sensitivity implies much larger, and probably more detrimental, climate changes. Thus, the sensitivity is a fundamental quantity for predicting future climate change. However, computer models of the Earth's climate show a large range of sensitivities. For example, the IPCC Third Assessment Report gives a global climate sensitivity range of 2.0 to 5.1 degrees C for doubled of CO2 for these models (IPCC, 2001). Note the spread in the colors around each of the solid lines in the figure to the right (from the IPCC report). The solid lines correspond to the multi-model average temperature change for a given emission scenario (A2, A1B, etc.), while the shading indications the spread on temperature changes using all the models.

A fundamental goal of my work is to understand why different climate models have different responses to the same change in atmospheric constituents (like CO2). The large range in climate change estimates is mainly due to differences in modeled physical climate feedbacks (those that alter the radiation balance of the planet). Calculation of these feedbacks, while essential to understanding the climate sensitivity, is not straightforward. As part of the development of the "radiative kernel" technique (Shell et al., 2008; Soden et al., 2008), I calculated one of the three original kernels (based on the National Center for Atmospheric Research (NCAR) climate model). This technique is less computationally expensive than most other techniques, while providing fairly similar results. Thus, it can be used to study very large data sets. For example, in Sanderson et al. (2009), we studied thousands of model simulations. Additionally, the kernel technique provides information regarding whether individual feedbacks differences are due to differences in the radiative transfer calculations or in how the snow, water, clouds, etc. respond to a warmer climate.

Some of my work has focused on improving the usefulness of the kernel technique. For example, in Held and Shell (2012), we explored a different framework for separating the individual feedbacks (specifically the water vapor and lapse rate feedbacks), which better identifies the feedback differences contributing to the spread of models. The figure to the left shows the relative magnitudes of the different feedbacks, helping us understand the differences in sensitivity among models. Another issue is that the technique becomes less accurate when the climate is very different from the present-day climate. One of my graduate students (Jonko) has analyzed these errors using a simulation of climate with 8 times the present-day 2 concentration. Fortunately, we were able to slightly modify the technique to enable its use in this very different climate, while retaining its computational efficiency (Jonko et al., 2012a). With the less-accurate kernels, climate sensitivity appears to decrease with warming (for one model), whereas the more accurate technique indicates an increased sensitivity (Jonko et al, 2012b). Another way to improve the kernel technique can be found in Sanderson and Shell (2012), where we demonstrated improvements using 21st Century simulations of over a dozen models.

Much of this work has centered around the Coupled Model Intercomparison Project (CMIP). Modeling centers all over the world contribute standardized simulations for comparison. The CMIP3 provided results from two dozen models to the IPCC AR4, while CMIP5 is contributing to the upcoming IPCC AR5. Obviously, the kernel technique is very useful for dealing with this amount of data; I have provided some results for the climate sensitivity sections and will be reviewing the 2nd order draft coming out this fall. I've also worked with colleagues at NCAR to understand why their climate model has changed so much from the previous version; we found that most of these changes are due to differences in cloud behavior (Bitz et al., 2012; Gettelman et al., 2012).

Extending the kernel technique beyond climate models to observational data, co-authors and I made the first estimate of Northern Hemisphere cryosphere (ice and snow) feedback from satellite data (Flanner et al., 2011). We found that the CMIP3 models underestimate the feedback between 1979-2008 by about half. [See the image to the right of cryospheric forcing. Image by Robert Simmon, made with data provided by Mark Flanner, University of Michigan]. Motivated by this result, Eric Skyllingstad and I have begun using high-resolution Arctic sea ice data to evaluate melt pond processes in models. Melt ponds decrease the reflectivity of sea ice and modify the seasonal evolution of ice cover. We are working with NCAR collaborators to studying and improve melt pond behavior in NCAR's model. Of course, we ultimately want to know not just why the models differ, but which one, if any, is correct. For this, we need to compare models to observations. One example is Flanner et al. (2011), where we estimated the Northern Hemisphere cryosphere (ice and snow) feedback using a variety of satellite data. We found that the CMIP3 models underestimated the cryospheric feedback between 1979-2008 by over half of the observed values.

Global, reliable satellite data are available for at most a few decades, too short to directly observe the century-scale feedbacks that will be active at the end of the 21st Century. Recently, my students and I have been exploring the potential of satellite data to evaluate feedbacks in models. Unfortunately, we have not yet found a clean relationship between short-term and long-term model feedbacks (Dalton and Shell, 2012; Jonko and Shell, in preparation), though the seasonal cycle appears promising. For example, models with large seasonal cycles in water vapor tend to have larger water-vapor feedbacks, but the results depend on the specific models used in the analysis. A complicating factor is that the spatial distributions of the seasonal and long-term feedbacks differ. We are currently developing better metrics for comparing satellite and model data.


Dalton, Meghan M., and Shell, Karen M., 2012: Comparison of short-term and long-term radiative feedbacks and variability in 20th century global climate model simulations, submitted to J. Clim.. preprint

Shell, Karen M., 2012: Consistent Differences in Climate Feedbacks Between Atmosphere-Ocean GCMs and Atmospheric GCMs with Slab-Ocean Models, accepted pending minor revisions J. Clim. preprint and supplement

Jonko, Alexandra, Karen M. Shell, Benjamin M. Sanderson, and Gokhan Danabasoglu, 2012: Climate feedbacks in CCSM3 under changing CO2 forcing. Part II: Variation of climate feedbacks and sensitivity with forcing, J. Clim, in press. preprint

Sanderson, Benjamin M. and Karen M. Shell, 2012: Model-specific radiative kernels for calculating cloud and non-cloud climate feedbacks, J. Clim., doi: http://dx.doi.org/10.1175/JCLI-D-11-00726.1, in press. Early online release.

Jonko, Alexandra K., Karen M. Shell, Benjamin M. Sanderson, and Gokhan Danabasoglu, 2012: Climate feedbacks in CCSM3 under changing CO2 forcing. Part I: Adapting the linear radiative kernel technique to feedback calculations for a broad range of forcings, J. Clim., 25, 5260–5272. doi:10.1175/JCLI-D-11-00524.1

Bitz, C. M., K. M. Shell, P. R. Gent, D. Bailey, G. Danabasoglu, K. C. Armour, M. M. Holland, and J. T. Kiehl, 2012: Climate Sensitivity of the Community Climate System Model Version 4, J. Clim, 25, 3053-070, doi: 10.1175/JCLI-D-11-00290.1

Held, Isaac M., and Karen M. Shell, 2012: Using Relative Humidity as a State Variable in Climate Feedback Analysis, J. Clim., 25, 2578–2582. doi:JCLI-D-11-00721.1.

Gettelman, A., J. E. Kay, and K. M. Shell, 2012: The Evolution of Climate Sensitivity and Climate Feedbacks in the Community Atmosphere Model, J. Clim, 25, 1453–1469. doi: 10.1175/JCLI-D-11-00197.1.

Flanner, M. G., K. M. Shell, M. Barlage, D. K. Perovich, and M. A. Tschudi, 2011: Radiative forcing and albedo feedback from the Northern Hemisphere cryosphere between 1979 and 2008, Nature Geosci, http://dx.doi.org/10.1038/ngeo1062.

Sanderson, Benjamin M., Karen M. Shell, and William Ingram, 2009: Climate feedbacks determined using radiative kernels in a multi-thousand member ensemble of AOGCMs, Clim. Dyn., 10.1007/s00382-009-0661-1.

Shell, Karen M., Jeffrey T. Kiehl, and Christine A. Shields, 2008: Using the radiative kernel technique to calculate climate feedbacks in NCAR's Community Atmospheric Model, J. Clim., 21, 2269-2282. (*)

Soden, Brian J., Isaac M. Held, Robert Colman, Karen M. Shell, Jeffrey T. Kiehl, Christine A. Shields, 2008: Quantifying Climate Feedbacks using Radiative Kernels, J. Clim., 21, 3504-3520. (*)

Last modified: Fri Nov 2 11:08:25 PDT 2012