Bayesian analysis of Earth's climate sensitivity: past, present and future

The biggest cause of uncertainty in predicting the magnitude of future global warming, for a given pattern of CO2 emissions, lies in Earth's 'climate sensitivity' (the increase in average surface temperature following a sustained doubling of atmospheric carbon dioxide). Earth's climate sensitivity is currently evaluated using many separate lines of evidence (Knutti et al., 2017), including palaeo-climate archives; complex climate models; observations of Earth's energy imbalance and warming since the industrial revolution (e.g. see Goodwin, 2018); and from the observed climate responses to volcanic activity.

However, these different lines of evidence produce different estimates of climate sensitivity leading to large uncertainty in future warming. The disagreement in climate sensitivity estimates may arise because climate sensitivity evolves over different response timescales (Goodwin, 2018), for instance due to the ocean's large heat capacity delaying the response to a change in greenhouse gas concentrations. Disagreements also arise because climate sensitivity may be dependent on the background state of the climate system (such that increased warming may itself increase the climate sensitivity and lead to additional warming).

For this study, the student will apply statistical techniques to produce a probabilistic assessment of Earth's climate sensitivity from multiple lines of evidence. This will include, but not be limited to evidence from: contemporary observations, historical reconstructions, geological archives of past climate change, and complex climate model simulations.

This study will use a range of observational and model data to constrain climate sensitivity over different timescales, and assess possible dependency of climate sensitivity on background climate state.

Bayesian statistical approaches will be employed to build a probabilistic estimate of climate sensitivity from multiple independent lines of evidence. A computationally efficient Earth System Model (Goodwin, 2018) will be extended to include both a background state-dependence of climate sensitivity, and long-timescale feedbacks that alter climate sensitivity over century timescales and longer. This Earth System Model will be used to generate large ensemble simulations with prior climate sensitivity characteristics reflecting particular lines of evidence (for example palaeo-archives). These large prior ensembles will then be assessed against constraints from other lines of evidence. This will generate posterior climate model ensembles with climate sensitivity characteristics reflecting multiple lines of evidence, building on a pilot studies by Goodwin et al. (2018) and Goodwin (2018).