We have been alerted to an excellent paper, that further document substantive issues with the climate model simulations that are used to create multi-decadal climate projections as used by the IPCC. The article is
Lucarini, V., and F. Ragone (2011), Energetics of climate models: Net energy balance and meridional enthalpy
transport, Rev. Geophys., 49, RG1001, doi:10.1029/2009RG000323.
The abstract reads [with highlight added]
We analyze the publicly released outputs of the simulations performed by climate models (CMs) in preindustrial (PI) and Special Report on Emissions Scenarios A1B (SRESA1B) conditions. In the PI simulations, most CMs feature biases of the order of 1 W m−2 for the net global and the net atmospheric, oceanic, and land energy balances. This does not result from transient effects but depends on the imperfect closure of the energy cycle in the fluid components and on inconsistencies over land. Thus, the planetary emission temperature is underestimated, which may explain the CMs’ cold bias. In the PI scenario, CMs agree on the meridional atmospheric enthalpy transport’s peak location (around 40°N/S), while discrepancies of ∼20% exist on the intensity. Disagreements on the oceanic transport peaks’ location and intensity amount to ∼10° and ∼50%, respectively. In the SRESA1B runs, the atmospheric transport’s peak shifts poleward, and its intensity increases up to ∼10% in both hemispheres. In most CMs, the Northern Hemispheric oceanic transport decreases, and the peaks shift equatorward in both hemispheres. The Bjerknes compensation mechanism is active both on climatological and interannual time scales. The total meridional transport peaks around 35° in both hemispheres and scenarios, whereas disagreements on the intensity reach ∼20%. With increased CO2 concentration, the total transport increases up to ∼10%, thus contributing to polar amplification of global warming. Advances are needed for achieving a self‐consistent representation of climate as a nonequilibrium thermodynamical system. This is crucial for improving the CMs’ skill in representing past and future climate changes.
The conclusion contains this excerpts
These findings suggest that in spite of the great successes of climate modeling, serious efforts seem to be needed in terms of basic science for substantially improving CMs’ performances. This is of great relevance when devising strategies for the future direction to be taken in the development of numerical simulations and is presently of great interest in the context of the preparation of the Fifth Assessment Report of the IPCC. While Earth system science requires models of increasing degrees of complexity that are able to correctly simulate the chemical and biological cycles of our planet and their changes in a changing climate, we cannot expect to improve our understanding of and ability to describe the CS just by indefinitely adding and coupling new modules. Nor can the advocated “quantum leap” in climate simulations be obtained just by improving the computational capabilities [Shukla et al., 2009]. In particular, the definition of strategies for improving the closure of the energy cycle in the fluid components of the CS [Becker, 2003] seems to be a crucial step, as important as taking care to consistently represent phase transitions and heat fluxes over land. Some CMs already adopt the strategy of reinjecting into the system the kinetic energy lost to dissipation as a uniform heating term. While this helps in reducing the energy imbalance, it creates regional temperature biases since climate regions featuring almost inviscid, horizontal dynamics are overheated (e.g., the stratosphere), and it defines a spurious source of entropy production as heat is moved down the gradient of the temperature field [Johnson, 2000; Lucarini, 2009a].
The consideration of multimodel ensemble means has often been used as a way to partially circumvent the problem of dealing with CMs’ discrepancies [see, e.g., Held and Soden, 2006; Trenberth and Fasullo, 2009]. The rationale for this choice is that it is often found that such means provide enhanced skill with respect to randomly chosen specific CMs [Gleckler et al., 2008], even if this is far from always being the case [Lucarini et al., 2008]. Although various statistical techniques have been proposed in place of the usual arithmetic average [Min and Hense, 2006, 2007; Tebaldi and Knutti, 2007] with encouraging empirical results, it must be emphasized that there is no rigorous statistical ground for combining data from different CMs: outputs of different CMs are not samples of the same probability space, and no large number law of any sort applies [Lucarini, 2008, 2009b]. Thus, a detailed understanding of CMs’ uncertainties and discrepancies is of the greatest urgency.