Quantifying Uncertainties in Climate System Properties with the Use of Recent Climate Observations
Tóm tắt
We derive joint probability density distributions for three key uncertain properties of the climate system, using an optimal fingerprinting approach to compare simulations of an intermediate complexity climate model with three distinct diagnostics of recent climate observations. On the basis of the marginal probability distributions, the 5 to 95% confidence intervals are 1.4 to 7.7 kelvin for climate sensitivity and −0.30 to −0.95 watt per square meter for the net aerosol forcing. The oceanic heat uptake is not well constrained, but ocean temperature observations do help to constrain climate sensitivity. The uncertainty in the net aerosol forcing is much smaller than the uncertainty range for the indirect aerosol forcing alone given in the Intergovernmental Panel on Climate Change Third Assessment Report.
Từ khóa
Tài liệu tham khảo
C. E. Forest M. R. Allen A. P. Sokolov P. H. Stone Clim. Dyn. in press.
Here we define climate sensitivity ( S ) as the equilibrium global-mean surface temperature change in response to a doubling of CO 2 concentration. We measure the rate of heat uptake by the deep ocean by an effective global diffusivity of heat anomalies ( K v ) into the ocean below the climatological mixed layer. For AOGCMs this effective diffusivity can be derived from transient climate change experiments and should not be confused with the model's sub–grid scale diffusion coefficient. S and K v are closely related to the climate sensitivity and transient climate response factors used in (36) to represent both the short- and long-term behavior of AOGCMs.
This analysis in contrast to ours used a simple energy balance model relied solely on surface temperatures and did not take into account the uncertain ocean heat uptake.
J. E. Penner et al. in Climate Change 2001 The Scientific Basis J. T. Houghton et al. Eds. (Cambridge Univ. Press Cambridge 2001) pp. 289–348.
V. Ramaswamy et al. in Climate Change 2001 The Scientific Basis J. T. Houghton et al. Eds. (Cambridge Univ. Press Cambridge 2001) pp. 349–416.
K. Hasselmann in Meteorology of Tropical Oceans D. B. Shawn Ed. (Royal Meteorological Society Bracknell UK 1979) pp. 251–259.
M. D. Webster et al. MIT Joint Program on the Science and Policy of Global Change Report 73 (Massachusetts Institute of Technology Cambridge MA 2001) pp. 1–3.
M. R. Allen et al. Surv. Geophys. in press.
r 2 = u T C N −1 u where u is the difference vector between the observation and model data ( y obs − y model ) and C N −1 is the inverse of the climate noise covariance matrix as estimated from segments of AOGCM control simulations. Because only limited length control segments are available a truncated representation of C N −1 must be used based on a projection onto the κ leading eigenvectors of the estimated C N (21). We use κ = 14 for both the surface and upper-air temperature diagnostics but similar results are obtained for a range of truncations (5). As described in (5) Δ r 2 ∼ mF m n where F m n is the F statistic with m and n degrees of freedom m is the number of constrained model properties and n is the degrees of freedom in estimated climate noise. Hence Δ r 2 provides a basis for rejecting climate model simulations as being inconsistent with the simulation with the minimum r 2 value. For a given level of significance we compute the appropriate F statistic that provides a cutoff value for r 2 . If r 2 is above this cutoff value we reject the hypothesis that the Δ r 2 is consistent with estimated climate noise.
For the surface and upper-air diagnostics the climate noise was estimated from the second-generation AOGCM developed at the Hadley Centre for Climate Prediction and Research (HadCM2) (37). For the ocean diagnostic the Geophysical Fluid Dynamics Laboratory AOGCM GFDL_R30_c (38) was used because estimates from HadCM2 were not readily available.
To compute the updated distributions for p (Π￨Δ T ) we first interpolate the r 2 values onto a finer grid [Δ S = 0.1 K Δ( K v 1/2 ) = 0.1 cm/s 1/2 ] over the range S = 0.5 to 10 K and K v = 0.2 to 64 cm 2 /s. After the data were interpolated in the S - K v plane we interpolated between the aerosol forcing levels to a resolution of Δ F aer = 0.05 W/m 2 over the range 0 to 1.5 W/m 2 . After the r 2 values were interpolated the probability of rejection was estimated (23) to generate the probability distribution on the finer grid. For all integral estimates of total probability or marginal probability distributions the finer grid spacing was used over the ranges specified.
The ranges explored for the model parameters set the limits of the uniform priors: S = 0.5 to 10 K K v = 0.2 to 64 cm 2 /s and F aer = −1.5 to 0 W/m 2 . We can assess the impact of these priors by examining the posterior distributions.
Marginal pdf(Π i ) ≡ ∫ p (Π) d Π j where Π i is the i th parameter and D j ≠ i is the parameter-space domain not including i.
U. Cubasch et al. in Climate Change 2001 The Scientific Basis J. T. Houghton et al. Eds. (Cambridge Univ. Press Cambridge 2001) pp. 525–582.
J. Hansen et al. J. Geophys. Res. in press.
We thank N. Gillett for streamlining the fingerprint detection code T. Delworth for GFDL_R30_c ocean data J. Antonov for deep-ocean temperature data the Hadley Centre for Climate Prediction and Research for HadCM2 data and A. Slinn for her graphics expertise. This work was supported by a National Oceanic and Atmospheric Administration Office of Global Programs grant NA06GP0061 (C.E.F.) and the European Commission QUARCC project (M.R.A.). The control run of HadCM2 was funded by the UK Department of the Environment Transport and Regions under contract number PECD 7/12/37.