Anticipated future warming of the climate system increases the need<br />for accurate climate projections. A central problem are the large<br />uncertainties associated with these model projections, and that uncertainty<br />estimates are often based on expert judgment rather than objective<br />quantitative methods. Further, important climate model parameters<br />are still given as poorly constrained ranges that are partly inconsistent<br />with the observed warming during the industrial period. Here we present<br />a neural network based climate model substitute that increases the<br />efficiency of large climate model ensembles by at least an order<br />of magnitude. Using the observed surface warming over the industrial<br />period and estimates of global ocean heat uptake as constraints for<br />the ensemble, this method estimates ranges for climate sensitivity<br />and radiative forcing that are consistent with observations. In particular,<br />negative values for the uncertain indirect aerosol forcing exceeding<br />-1.2 Wm(-2) can be excluded with high confidence. A parameterization<br />to account for the uncertainty in the future carbon cycle is introduced,<br />derived separately from a carbon cycle model. This allows us to quantify<br />the effect of the feedback between oceanic and terrestrial carbon<br />uptake and global warming on global temperature projections. Finally,<br />probability density functions for the surface warming until year<br />2100 for two illustrative emission scenarios are calculated, taking<br />into account uncertainties in the carbon cycle, radiative forcing,<br />climate sensitivity, model parameters and the observed temperature<br />records. We find that warming exceeds the surface warming range projected<br />by IPCC for almost half of the ensemble members. Projection uncertainties<br />are only consistent with IPCC if a model-derived upper limit of about<br />5 K is assumed for climate sensitivity.
Knutti, R., Stocker, T. F., Joos, F., & Plattner, G. K. (2003). Probabilistic climate change projections using neural networks. Climate Dynamics, 21(3–4), 257–272. https://doi.org/10.1007/s00382-003-0345-1