Donifan Barahona NASA Global Modeling and Assimilation Office Earth system assessments depend on historical data from models, satellite remote sensing, and in situ measurements. These sources help identify climate trends and refine models, especially for processes that are poorly understood or too computationally intensive to simulate directly. However, models simplify small-scale processes, and observations often carry significant uncertainties. These limitations introduce uncertainty into climate projections and forecasts. This seminar presents a new method that integrates high-resolution simulations, reanalysis datasets, and long-term observations to develop advanced atmospheric retrievals. In this approach, simulations are used to train feature extractors, which are then fine-tuned using observational data with generative techniques to reduce experimental errors. The method is applied to two key variables: vertical wind velocity and cloud droplet number concentration, both central to cloud evolution, turbulence, and long-term climate change. The results reveal important trends over recent decades with major implications for future climate projections.
Copyright. All Rights Reserved.