Insights on Internal Variability and Uncertainty from a Downscaled Climate Large Ensemble (invited)

Although CMIP6 single model initial-condition large ensembles (SMILEs) are increasingly used to investigate the role of internal variability at coarse GCM scales, there remains a significant gap in understanding how internal variability evolves at higher resolutions (e.g., approximately 10 kilometers or finer). Additionally, the impact of downscaling on internal variability remains underexplored. Federal and state agencies are increasingly relying on and disseminating downscaled data for planning, yet few publicly available downscaled datasets incorporate multiple ensemble members from the same GCM, which is crucial for accurately assessing the inherent variability of future climate projections. In this work, we analyze the newly released Generalized Analog Regression Downscaling method Large Ensemble (GARD-LENS) dataset to enhance our understanding of internal variability, uncertainty, and signal-to-noise ratios in high-resolution, downscaled climate projections. By leveraging GARD-LENS, which includes 200 statistically downscaled ensemble members from 3 CMIP6 SMILEs, we quantify the contributions of model uncertainty, structural uncertainty, internal variability, and downscaling to overall projection uncertainty. Our results reveal that downscaling significantly alters the signal-to-noise ratio and time of emergence, especially in mountainous regions. Internal variability appears to contribute more to the overall uncertainty in high resolution precipitation projections than in coarse GCM projections. This study underscores the need to account for internal variability in high-resolution climate projections to better inform planning for future communities and infrastructure.