Detecting potential structural errors in an aerosol-climate model using a perturbed parameter ensemble and a synthesis of observations

Abstract:

In this presentation we exploit the power of a very large ensemble of model simulations combined with a large dateset of aerosol measurements to detect potential structural errors in an aerosol-climate model. Structural errors (missing or poorly represented processes) are normally detected during model evaluation when a model fails to match observations even when key parameters are adjusted. However, this is not a very rigorous procedure because there are dozens of uncertain model parameters and a very large number of target observational metrics. Here we attempt to identify structural problems with a model by comparing a perturbed parameter ensemble of the UKESM climate model with a large suite of measurements collated within the Global Aerosol Synthesis and Science Project (GASSP). The ensemble perturbed over 50 model parameters related to aerosol emissions and processes in the aerosol, cloud, atmospheric and radiation components of the model. We then compare the resulting PDFs of model output with observations to detect conditions where the observations confidently lie outside the ensemble spread, or where agreement with different observations requires very different or inconsistent parameter combinations. We show that narrow 'tuning' of the model to match a particular observation without regard for the degraded agreement with other observations can lead to over-confident estimates of the radiative forcing.