Multidecadal climate to within a millikelvin

We analyze and compare the monthly global land-sea surface temperature datasets HADCRUT3 and HADCRUT4 for 1850-2010 by subtracting two analytically modeled components and demonstrating with a suitable low-pass filter that the residue contains no significant fluctuations with periods longer than the 22-year Hale cycle. It follows that the two components are essentially all there is to recent multidecadal climate thus understood. With either dataset, the model forecasts a 4 C rise for 2100. The first component, AGW, models the impact of increasing CO2 based only on Arrhenius's 1896 logarithmic dependence of surface temperature on CO2, Hofmann's 2009 exponential model of anthropogenic CO2 emissions superimposed on a preindustrial base, and Hansen's 1997 pipeline modeled as a fixed delay between increasing radiative forcing and resulting increasing surface temperature. The delay is the time required to heat the ocean in its role as planetary heatsink; neglecting it creates two problems for our analysis, namely a one-degree-per-doubling difference between theoretically predicted and empirically observed *modern* climate sensitivity, and an apparent absence of parameters leading to anywhere near as good a fit to the data. The best fit is obtained with a pipeline of 14.5 years and a climate sensitivity of 2.8 C/doubling, which would be 1.8 without the pipeline. The second component, SAW, is an inverse sawtooth wave filtered to attenuate and phase-shift its first five harmonics and neglect the rest. The second and third harmonics correlate well with the principal ocean oscillations, e.g. AMO; they are passed through largely unfiltered while the other three are more sharply attenuated. One possible physical origin would be two similarly sized seismic events at the core-mantle boundary in respectively the mid-18th and early 20th centuries. Each event would briefly bring heat from the lower mantle to the crust, felt most strongly in the oceans because the crust there is only 20% the thickness of the continents, with the crust providing the filter. The parameters for AGW and SAW were determined by joint least squares fits to the respective datasets along with the Keeling curve and the CDIAC datasets for CO2 changes attributable to burning fossil fuel, cement production and land use changes. We used a purpose-designed low-pass filter to extract the long-term fluctuations from the residue. These vary during the 20th century by 10 millikelvins for HADCRUT3 and 1 millikelvin for HADCRUT4, making our model a particularly good fit to the latter. All other long-term components besides AGW and SAW are invisible either because their amplitude is less than a millikelvin or they are too well correlated with AGW or SAW to permit separating them (e.g. H2O and CO2). This analysis is downloadable as a separate Excel spreadsheet for each dataset from http://clim.stanford.edu. Controls are provided for experimenting with alternative parameters.