Ogmius Exchange Part I
A caution to policy makers: Climate models fail key tests for accuracy by Tom Chase
Should policy makers base decisions on the results of current climate models? I remain unconvinced for several reasons. In a soon to be published paper, Reichler and Kim, 2008, argue that climate models have steadily improved in the last decade in simulating climate behavior when compared to present day observations. While models are improving, to be convincing they must simulate pertinent physical processes accurately to within some objective measure of skill. One minimal definition of accuracy might be that errors in the simulation are smaller than the signal we are trying to detect.
The recent Intergovernmental Panel on Climate Change report (IPCC) (Randall et al., 2007) does not provide any assessment of model skill (ability to reproduce observations) in simulating the present day greenhouse effect (downward longwave radiation). It does, however, supply comparisons of model error for other components of the radiation simulation. Figure 1 (from Randall et al., 2007) compares root mean square errors averaged around a latitude band for all current generation IPCC models for outward solar radiation (top panel) and outward terrestrial radiation. We can take these figures as representative of the magnitude of errors in simulating the observed greenhouse effect. Model error varies by latitude from approximately 5W/m2 (energy per unit time per unit area) to nearly 40W/m2 in the solar simulation with an averaged model error of greater than 8W/m2 at nearly every latitude. The longwave simulations have average errors of 5W/m2 or more at most latitudes with individual models exceeding 30W/m2 at some locations. Wild (2005) indicates that current models still have systematic errors in simulating parts of the radiation budget and concludes that considerable uncertainties and inconsistencies remain in model calculations particularly as related to water vapor, the main greenhouse gas.
Appreciable greenhouse warming also depends on positive feedbacks in the hydrological cycle (clouds, water vapor and ice) and yet these are precisely the areas which cause most errors in present day simulations (Randall et al., 2007; Wild, 2005).
The magnitude of model error in simulating the present day is important because evidence exists that natural fluctuations in extreme events and average climate may still be larger than any human climate signal (e.g. Chase et al., 2006; Keenlyside et al., 2008) indicating that precision is necessary to separate any human effect. Additionally, Figure 2 shows a summary comparing various expected climate effects and indicates the estimated radiative forcing due to greenhouse gases is approximately 2-3 W/m2. Even if this perturbation were doubled, the errors in the radiation simulation in models calibrated for the present day are still much larger than the signal we are looking for.
Figure 2 also includes an assessment of the Level of Scientific Understanding (LOSU) for the processes discussed. The effects due to greenhouse gases are judged to have high levels of scientific understanding notwithstanding the simulated errors discussed above but this confidence deteriorates to low and medium levels meaning errors in simulating these are potentially much larger than those due to radiation.
Finally, we have shown in our model simulations (Lawrence and Chase, 2007) that the hydrological response to changes in land cover dominates the albedo effect in line with studies by e.g. Chase et al. (1996) and Feddema et al. (2005). As shown in Figure 2, this does not even consider this hydrological response.
Current generation climate models are calibrated to reproduce present day climate and yet are unable to simulate present day radiation balances, the fundamental physical process we are interested in, to the required degree of accuracy (errors are much larger than the several W/m2 signal we are looking for). Simulation of future climate is dependent on accurate simulation of feedbacks in the hydrological cycle which have proved elusive. Processes for which we have low to medium levels of scientific understanding cannot be simulated to a high degree of accuracy. In some cases the assumed physical mechanisms involved might be entirely wrong. Evidence exists that natural fluctuations may still dominate any human climate change. Finally, there have been relatively few predictions made by climate models which have been unambiguously shown to be a signature of greenhouse gas warming. Climate models can be used effectively as guides to possible physical outcomes which need to be independently verified by other means. As yet output from climate models remains mostly speculative and should be used with caution as a guide to policy decisions.
CIRES/University of Colorado
Continue to read Part II: Models can be useful tools for planning ahead by Kevin E. Trenberth