Learning about equilibrium climate sensitivity and transient climate response from observations
Abstract
One key uncertainty within the science of climate change pertains to the Equilibrium Climate Sensitivity (ECS). Several studies using instrumental observations to estimate ECS have found shifts in the Probability Distribution Function (PDF) towards lower values when including observations over the last decade. Identifying the reasons behind shifting PDFs - in the hope of making the best interpretation and use of them - points to the crucial and wider issue of how we should learn from observations, in particular observations accumulating over time. If our underlying geophysical model or our statistical model does not fully capture the climate variability and the warming signal we could expect periods when we appear to learn but when we do not (Oppenheimer et al, 2008). This highlights the importance of using several different model structures when estimating PDFs from observations and the importance of analyzing how the PDFs change as the observational time series get longer. By analyzing how the PDFs change over time and by offering explanations for those changes one should be better equipped to draw conclusions on what can be learnt from the analysis. In this presentation we analyze how the PDFs for ECS and Transient Climate Response (TCR) change as the observational record accumulates. We use a land-ocean resolved Upwelling Diffusion Energy Balance Model (UDEBM) together with a Bayesian Markov Chain Monte Carlo approach. Radiative forcing time series since 1765 are used together with land and ocean surface temperature observations (CRUTEM4 and HadSST3) as well as observations on ocean heat content to constrain parameters in the UDEBM. Our preliminary results show that with observations up to 1991 (starting in 1856) the PDFs is relatively well constrained in comparison with the prior assumptions. The most like values are 3.0°C and 1.9°C for ECS and TCR, respectively. The corresponding 5-95% intervals are 2.1-6.9°C and 1.5-2.9°C. This can be compared to the prior 5-95% intervals which are 0.9-11°C and 0.7-4.4°C for ECS and TCR, respectively. With observations up to 2001 the most likely value remains the same for TCR while it drops to 2.6°C for ECS. The 5-95% range both drops and shrinks for the ECS to 1.8-4.6°C, while for TCR the range is 1.4-2.8, i.e., the TCR range remains about the same. Adding observations between 2001 and 2011 has a strong effect on the 95% level for the ECS, but hardly any effect on the summary statistics for the TCR. The most likely values are 2.5°C and 1.9°C for ECS and TCR, respectively, while the 5-95% intervals are 2.0-3.6°C and 1.4-2.8°C, respectively. Hence, based on the summary statistics presented here the PDFs for ECS and TCR appear to have been effectively constrained downwards for several decades, while the last decade has had a strong effect on reducing the likelihood for high values (>95% level) on the ECS (>3.6°C) and the TCR (>2.4°C). References Oppenheimer M, O'Neill BC, Webster M, 2008, Negative learning. Climatic Change 89: 155-172
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2013
- Bibcode:
- 2013AGUFMGC31B1054J
- Keywords:
-
- 1600 GLOBAL CHANGE;
- 6309 POLICY SCIENCES Decision making under uncertainty;
- 1616 GLOBAL CHANGE Climate variability;
- 3270 MATHEMATICAL GEOPHYSICS Time series analysis