Simplifying the cloud microphysics and aerosol representation of a global aerosol climate model

Abstract

Climate models are large and complex constructs. They are built and used with different purposes in mind, to generate realistic projections or aid in developing understanding. A common approach to model building is to try to represent all processes that are deemed important for a specific climate system compartment. Aerosols and cloud microphysics are two features of the climate system that influence the radiation balance as well as the hydrological cycle. Since the associated processes cannot be resolved explicitly on the coarse climate model grid, their effect on grid-scale variables needs to be parameterized. In developing such parameterizations, it is often assumed that greater detail is beneficial since it increases the representativeness of the model compared to the physical world. Thus, model complexity has become a norm in climate model development. However, model complexity also has negative effects. Among others, it hinders understanding and the interpretability of the model. Here I address model complexity by simplifying the aerosol and cloud microphysics scheme of the global climate model ECHAM-HAM. First I developed a method to assess the potential for simplifications in the cloud microphysics (CMPs) scheme. I implemented parameters for perturbing a processes' effect on model variables. Simulating many simultaneous perturbations I generated a perturbed parameter ensemble (PPE). Constructing a surrogate model from the PPE allows us to apply a quantitative sensitivity analysis. Indeed, I find that model sensitivities are dominated by one of the four investigated cloud microphysical processes, while two are negligible in comparison and thus could be simplified. I go on to apply this methodology to the whole two moment (2M) CMPs scheme of the aerosol climate model ECHAM-HAM. Perturbing 15 processes, I find that 8 have potential for simplification. Indeed, setting processes' effects constant or to a prescribed climatology or even removing some gives satisfying results for 7 of them. Importantly, the derived simplifications are robust in different climate states, preserving the models' fit for climate projection applications. Repeating the same analysis for the alternative P3 scheme, I see shared sensitivities. However, the process of ice crystal autoconversion, which dominates sensitivities in the 2M scheme is unnecessary in the P3 scheme and thus the latter scheme exhibits more balanced sensitivities. Third, I turned to the aerosols which impact clouds via their role as nucleation centers for cloud droplets or ice crystals. Since our scientific interest focuses on the CMPs, I attempted to drastically simplify the aerosol module. I developed two simplifications in the form of prescribed climatologies, one for potential cloud condensation nuclei (CCN) and one for aerosol mass and number concentrations. In terms of global cloud variables, the climatological approach proves promising. However, a mean climatology of CCN underestimates cloud droplet number concentrations in the Southern Ocean. This bias can be eliminated by incorporating a treatment of hygroscopic growth in the climatology. At the same time, the climatological treatment enables large computational savings. The sensitivity analysis and simplifications highlight scheme redundancies and peculiar model behaviour. Thereby the method of simplification is shown to generate new understanding, enable a new perspective, and open up promising avenues for model development. The results of this work call into question the complexity paradigm in climate modeling.