AiBEDO: A hybrid AI model to capture the effects of cloud properties on global circulation and regional climate patterns

Clouds play a vital role in modulating Earth's radiation budget and shaping the coupled circulation of the atmosphere and ocean, driving regional changes in temperature and precipitation. Clouds are, therefore, opportune for climate intervention techniques such as Marine Cloud Brightening. However, these techniques may have unintended consequences on regional climate patterns, which must be studied thoroughly. AiBEDO is a pioneering effort in developing a hybrid AI model framework to resolve the weaknesses of Earth System Models by generating rapid and robust multi-decadal climate projections. We follow an approach rooted in an innovative application of the fluctuation-dissipation theorem that states that the forced response of the climate system mirrors the response due to internal variability. We will demonstrate its utility using marine cloud brightening scenarios—to avoid climate tipping points and the consequences of ill-planned interventions. Our hybrid model consists of spherically sampled deep learning neural network (spatial) and multi-timescale (temporal) modules bolstered by physics constraints. For better consideration of spherical spatial component of our data to the model, we used deep neural network architecture based on Chebyshev-polynomial graph convolution from spherically resampled. For considering temporal component, we augmented temporal dimension by concatenating along the variable vector axis of input data. Our model is trained on 19 CMIP6 model ensemble data and validated against observational reanalysis data. Our results thus far show that our hybrid AI model can accurately reproduce the short-term regional climatic patterns due to the natural variability of cloud radiations while retaining the overall physics of the earth system. Our next steps include extending the model for targeted cloud perturbation experiments and incorporating long-term decadal trend analysis.