Statistically Constraining Uncertain Climate Model Parameters via Perturbed Parameter Ensemble (PPE) Approach: Application to the Radiative Effects of Biomass Burning Aerosols over Southern Africa

Fires over Southern Africa are the largest contributors of the biomass burning aerosols (BBA) budget in the atmosphere. These aerosols are important for the planet’s radiation budget due to their direct and indirect radiative effect. The radiative effects of these BBA contain significant uncertainties in the current state of climate models. In this work, we constrain the uncertainty in the modeled radiative effects of the seasonal biomass burning emissions over Southern Africa by calibrating the underlying climate model input parameters. These constraints were generated using the real-time measurements during the Observations of Aerosols above Clouds and their Interactions (ORACLES) campaign. We hypothesize that the uncertainty in the direct and indirect radiative effects of BBA can be reduced by constraining 12 uncertain parameters in the climate model.

A Perturbed Parameter Ensemble (PPE) is a technique used within climate modeling to study model uncertainties, wherein multiple model parameters are varied simultaneously to create multiple versions of the same model. This can be helpful for sampling a range of model uncertainties and understand the effects of different parameter combination on the climate projections on global and regional scales. In this work, a PPE was built, consisting of 121 runs of the atmosphere-only UK Earth System Model spanning our space of 12 uncertain parameters. The simulations were performed during the biomass burning season of 2017. The simulation outputs were written at a three-hourly time resolution and over the continental and maritime region around Southern Africa. The PPE outputs were interpolated over the 4-dimensional space-time domain defined by flight tracks during ORACLES. The interpolated PPE outputs were then used to train Gaussian Process Regression surrogate models for a dense sampling of the parameter space. Constraints on the climate model parameters obtained in this work were combined with those obtained using satellite measurements previously presented in a companion work (Sanchez et al.). Since radiative effect cannot be measured, we used other measurable variables (or, comparison variables, such as mass absorption cross-section and accumulation mode number concentration) and compared them with the climate model simulations to rule out parameter combinations that produce implausible results. We then calculated the radiative effects by comparing simulations with and without smoke emissions within the constrained parameter space.

To constrain uncertain model parameters, we built upon the frequentist approach developed in Carzon et al. (Environmental Data Science, 2023). This approach allowed us to quantify uncertainty in the parameters with a high level of confidence, improving on previous work that implemented the “history matching” method (whose theoretical guarantees are not clearly understood). We incorporated additional measures into the methods of Carzon et al. to broaden the applicability of the frequentist approach. Here we present an improved statistical methodology, confidence intervals on parameters, and the implications for BBA radiative effects.