Machine Learning to Decrease Uncertainty in Climate Forcing by Aerosols


The largest uncertainty in climate forcing is associated with the forcing by atmospheric aerosols. Aerosols cause climate forcing through their direct scattering and absorption of solar radiation and through their indirect formation of cloud droplets on aerosol cloud condensation nuclei (CCN). The uncertainty in aerosol climate forcing is estimated from the spread in forcing estimates in global models and/or observations. The uncertainties in these estimates have remained large primarily because there are no direct estimates of forcing based solely on observations, and the many model processes required to estimate forcing are treated differently in different models. As a result, a given model may fit some types of data, while other models fit other data. So far, it has not been possible to understand the causes of the differences in models and thereby to decrease the spread in forcing estimates, even though a large network of modelers have run intercomparison studies.

In this work we propose to build a feedforward neural-network emulator that uses inputs from the aerosol/climate model from Penner’s group that are adjusted (in the neural network) to better fit available satellite-observations of aerosol optical depth. This will allow us to determine which aspects of the aerosol/climate model need to be improved. After demonstrating the ability to use and develop a neural network for this purpose, a follow-on project proposal (to NSF, NASA, and/or DOE) would extend the neural network to treat the formation of cloud droplets by CCN in the model and identify which processes in the model need to be adjusted to fit the satellite-observed droplet number concentrations. In addition, we would gather process data from other models within the climate modeling community, build a neural-network emulator specific to each model, and compare differences between the models. By identifying those models that are outliers, we can reduce the spread in climate forcing by aerosols. Moreover, we can identify which processes are most important for determining the spread in model predictions. With this information, the models can be improved so that they all can better fit the observations. This would also lead to a decrease in the spread of aerosol climate forcing.

The techniques we would use have been pioneered in the study by Nicely et al. (Atmos. Chem. Phys., 2020). They examined which processes in atmospheric chemistry models lead to differences in their predicted hydroxyl radical concentrations. Thus, our proposed approach for aerosols is likely to succeed and enable a decrease in the uncertainty in aerosol climate forcing.

People

Joyce
Penner

CLaSP
Engineering

Xianglei
Huang

CLaSP
Engineering

Yang
Chen

Statistics
LSA


Funding: $30K (2022)
Goal: Our project will build a feedforward neural-network emulator that uses inputs from an aerosol/climate model that are adjusted (in the neural network) to better fit available satellite-observations of aerosol optical depth.
Token Investors: Joyce Penner, Xianglei Huang


Project ID: 1013