Quantifying and Observationally Constraining Uncertainty in Aerosol-Cloud Radiative Forcing (Invited Presentation)

Understanding and reducing uncertainty in the aerosol effective radiative forcing (ERF) is challenging because of the many diverse sources of uncertainty and the computational cost of running climate models, which makes it difficult to fully sample the uncertain model processes. Here we aim to quantify which atmospheric and aerosol processes contribute most to the uncertainty in aerosol-cloud ERF and then to explore which observations will help to constrain the model spread.  

Our approach is in two stages: we first develop a way to fully sample a large number of climate model uncertainties to generate a spread of model ERFs and associated cloud, aerosol and radiation metrics. We then use observations to narrow the model spread by rejecting implausible model behaviour. Our definition of “model uncertainty” is the range of predictions of ERF using all the model variants that are deemed plausible against observations.

To quantify the model uncertainty we created a large perturbed parameter ensemble of the HadGEM3 climate model. We perturbed 27 parameters in the physical atmosphere model known to influence the properties and distribution of clouds and humidity (atmospheric parameters) in combination with aerosol emission, deposition and process parameters (aerosol parameters). We then used model emulation and Monte Carlo sampling to fully sample model output across multi-dimensional parameter space.

Our results show that most of the uncertainty in present-day radiative flux is caused by uncertainty in the physical atmosphere model, particularly parameters related to clouds. However, over half of the uncertainty in ERF due to aerosol-cloud interactions (ERFaci) over the industrial period is caused by the representation of aerosols in the model.

In an attempt to reduce the ERFaci uncertainty we constrained the model using CERES observations of the present-day ToA flux. This would appear to be an ideal candidate observation for constraining ERFaci because our results show that much of the uncertainty in ToA flux is caused by aerosol and cloud processes that also determine the ERF uncertainty. However, we find that tight observational constraint of present-day ToA flux has almost no effect on reducing aerosol ERF uncertainty. The cause of the weak constraint is model “equifinality” – in a complex model with many compensating uncertainties there are many model variants that simulate approximately the same ToA flux but these variants predict very different values of ERFaci.

Our results have important implications for how observations will help to reduce uncertainty in ERFaci. It seems obvious to use observations of variables that are sensitive to the same model uncertainties (as is the case for ERFaci and ToA flux). However, constraining the model using these observations may not affect ERFaci uncertainty very much unless we simultaneously constrain all the other uncertain processes. Tuning a model to match particular observations will result in an observationally consistent value of ERFaci, but if the full uncertainty space of the model is not explored, then the resulting ERFaci will only be one of many plausible values. Our methodology, of comprehensively sampling model uncertainties in multiple dimensions provides a way to relate various observations to their effect on reducing ERFaci uncertainty.