Towards Maximum Feasible Reduction in Aerosol Forcing Uncertainty (Aerosol-MFR)

For several decades, large uncertainties have persistently affected model predictions of how the atmosphere and climate behaves, evolves and responds to changes, severely limiting confidence in climate projections. In particular, estimates on how aerosols (tiny particles in the atmosphere) have affected the Earth's energy balance since pre-industrial times, the 'aerosol radiative forcing', are notably uncertain. The Intergovernmental Panel on Climate Change (IPCC) has repeatedly flagged aerosols as the largest source of uncertainty in simulations of global temperature change. Reducing aerosol radiative forcing uncertainty in climate models is a significant challenge. This project, Aerosol-MFR, will develop innovative statistical approaches to combine observations and large climate model ensembles to determine how best to reduce this large and persistent uncertainty.

Aerosol-climate models, like complex models from many disciplines, are highly uncertain for several reasons: the equations in the model have many uncertain inputs (parameters), they contain many simplifications of the real-world processes they represent, and they have structural deficiencies (poor or missing representations of important physical and chemical processes). Because there are so many sources of uncertainty in a climate model, it is difficult to define the most realistic set-up of a model that produces best agreement with a wide range of observations. It is also difficult to tell whether poor model agreement with observations is related to structural deficiencies in the model or just inappropriate settings of the uncertain model input parameters. Without being able to tell what is causing the poor agreement with observations, we will not know how to improve the model and reduce its uncertainty.

Comparison of models with observations is vital for reducing the uncertainty in the simulated aerosol forcing. However, previous research shows that reduction in uncertainty is strongly limited by two problems: (1) Multiple model errors can cancel each other out, so good agreement of the model with observations can hide the errors, which then become apparent again when the model is used to make predictions (e.g., of the aerosol forcing). (2) Model structural deficiencies mean that no amount of adjustment of parameter values will produce a model that can make reliable projections. The problem is that the effects of structural deficiencies and choices of parameter values cannot easily be distinguished when a model is compared against observations, but this separation is necessary if we are to improve models.

This project will use the Met Office's UK Earth System Model (UKESM1) and new statistical methods to investigate the relationships between observations, model output variables and aerosol radiative forcing, and will identify combinations of observations that can be used to reduce the model uncertainty and reveal model structural deficiencies. The causes of the structural deficiencies will be investigated and several will be corrected through model development activities. This will produce a new model version that will then be evaluated against observations to assess the reduction in aerosol forcing uncertainty that results from our approach.

The outcomes from this project will be a reduction in climate model uncertainty, a more realistic climate model, information to prioritise future model developments, and statistical approaches that will be transferable to other models and other disciplines.

Aerosol-MFR is at the frontier of model analysis and improvement. In particular, this will be the first time that the two major causes of climate model uncertainty (structural model deficiencies and uncertain model parameters) will be tackled together. It is also the first time that the effectiveness of observations to reduce model uncertainty will be tested and optimised using advanced statistical methods.