The uncertainty in the GENIE model

The evaluation of Regionnaly-Resolved Climate Change Uncertainty

In the SynsCOP15 project we have applied emulation to improve the quantification of uncertainty in climate predictions. Emulators are computationally cheap surrogates for models, here the Intermediate Complexity earth system models GENIE-1 (2D atmosphere, 3D ocean) and GENIE-2 (3D atmosphere, 3D ocean). Computationally-efficient, regionally-resolved calculations of climate change are required for the evaluation of uncertainty in many applications in integrated assessment. We have here applied emulation techniques to generate spatially-resolved estimates of climate change and associated uncertainty that are consistent with a range of possible COP-15 negotiation outcomes, driven by the CO₂ emissions output of TIAM.

1) GENIE-1: continental-scale warming and terrestrial vegetation

The first technique, building on Holden et al (2010), addressed the design and evaluation of ensembles of climate change. In order to better quantify model uncertainty, we precalibrated (Edwards et al, in press) GENIE-1, allowing 25 model parameters to each vary over the entire range of plausible input values. We required the resulting models to reproduce the main features of climate (though not precise observations), an approach which leads to a wide spread of large-scale feedback strengths, generally encompassing the range of GCM behaviour. However, in the design of such an ensemble it was not feasible to explore the entire input space with a naive Monte-Carlo approach; using such an approach only 10 from 1,000 ensemble members resulted in plausible climates. We thus built emulators of the preindustrial climate state of GENIE-1 and used them to perform a rejection sampling to derive a collection of 1,000 parameter sets which the emulators predict will be plausible; 894 of these were indeed found to provide plausible preindustrial climates in the GENIE-1, of which 480 simulations also produced plausible glacial climate states. These 480 model configurations are thus constrained to both reproduce preindustrial climate and to exhibit feedback strengths that are consistent with a well understood alternative climate state.

In the SynsCOP15 project, we extended this approach by applying uncertain future radiative forcing scenarios to the 480 plausible model configurations. This was achieved by decomposing a range of possible effective CO₂ concentration trajectories (2000 to 2100 AD) into Chebyshev polynomials and running the GENIE-1 simulations on from their preindustrial (1850 AD) equilibrium states to 2100, applying this uncertain future forcing together with known historical forcing (1850 to 2000). Emulators were then built from the output of these 480 simulations. The emulators describe the regionally averaged i) warming, ii) terrestrial vegetative carbon storage and iii) fractional vegetation coverage, in each case the change at 2100 AD relative to the present day (2000 AD). For each of these three variables, four emulators were built describing the average change over four continental-scale regions: i) Southern Hemisphere Asia/Africa/Australasia ii) Northern Hemisphere Eurasia/Africa, iii) South America and iv) North America. The inputs to these twelve regional emulators are the 25 model parameters and the three coefficients which describe the unknown future CO₂ pathway.

For each of the COP-15 negotiation outcomes considered in the SynsCOP15 project, the emissions output of TIAM was applied as forcing input to the emulators to generate an emulated ensemble for each of the twelve outputs, enabling us to provide estimates for each, together with an evaluation of associated uncertainty (through the standard deviation of the emulated ensemble).

2) GENIE-2: Low-resolution (about 1,000 km) maps of temperature and precipitation

The second technique developed for SynsCOP15 applied dimensional reduction to build emulators of 2D GENIE-2 output fields (Holden and Edwards, 2010). The approach was developed as an alternative to pattern scaling (Mitchell et al 1999), an alternative which enables us to capture non-linear spatio-temporal behaviour. For instance, as the spatial response under pattern scaling is assumed to be invariant, a region that is predicted to have a small change in precipitation will also be associated with low uncertainty (although it is possible to apply patterns from different models to capture alternative model responses). In contrast, our novel emulation approach naturally captures differences between the spatial structure of the simulated response and the spatial structure of the uncertainty. We note that whilst we have only emulated the five highest-order modes of variability, which together capture ~50% of the simulated variance in precipitation (~90% of the variance in surface warming), the approach will be extended to include additional low-order modes.

Analogously to the work with GENIE-1, we first used emulation to design a simulated ensemble of plausible preindustrial climate states and then applied known historical and uncertain future concentration profiles to generate a 245-member simulated ensemble of possible future climate trajectories. We then extended the GENIE-1 approach by applying principal component analysis to project 2D model output fields onto 1D space and then emulating the map from input space (19 model parameters and 3 forcing inputs) to the 1D output space. Emulators were built to generate 2D maps of surface warming and precipitation change (2100-2000 AD).

These emulators were applied to generate 2D maps of surface warming and precipitation, together with the spatially resolved uncertainty, in response to a range of future emissions profiles; effective CO₂ concentration profiles were generated by TIAM for each of the COP-15 negotiation outcomes and applied to the GENIE-2 emulators to produce consistent climate change fields.

The emulation of a 122-member ensemble takes ~0.06 s, in stark contrast to the ~1,000 hours required to perform the same ensemble with GENIE-2. This efficiency paves the way for incorporating improved calculations of climate change in coupled climate-economic integrated assessment models, including location-dependent estimates of uncertainty.

References

Edwards NR, Cameron D and Rougier J (in press). Precalibrating an intermediate complexity climate model. Climate Dynamics, doi: 10.1007/s00382-010-0921-0

Holden PB, Edwards NR, Oliver KIC, Lenton TM and Wilkinson RD (2010). A probabilistic calibration of climate sensitivity and terrestrial carbon change in GENIE-1. Climate Dynamics, 35, 785-806, doi 10.1007/s00382-009-0630-8

Holden PB and Edwards NR (2010). Dimensionally reduced emulation of an AOGCM for application to integrated assessment modelling. Geophysical Research letters, 37, L21707. doi:10.1029/2010GL045137

Mitchell JFB, Johns TC, Eagles M, Ingram WJ and Davis RA (1999). Towards the construction of climate change scenarios. Clim. Change 41 547-581, doi:10.1023/A:1005466909820

Rougier J (2007). Probabilistic inference for future climate using an ensemble of climate model evaluations. Clim. Change, 81, 247-264. doi: 10.1007/s10584-006-9156-9