|
Abstract:
|
State-of-the-science climate models are valuable tools for understanding past and present climates and are particularly vital for addressing otherwise intractable questions about future climate. Given the societal relevance, maintaining model confidence is critical. The complex processes characterizing the Earth System lead to inherently chaotic behavior, and while climate scientists typically use initial condition perturbations to create ensemble spread, similar effects result from seemingly minor changes to the hardware or software stack. This sensitivity makes defining “correctness” of model output separately from bit-reproducibility a practical necessity. To this end, we have developed a framework that utilizes an ensemble of established simulations and Principal Component Analysis to perform hypothesis testing for model consistency. This test is already implemented and provides valuable feedback to model developers. We are now exploring the impact of ensemble size, specifically as it relates to errors in estimating the unknown correlation matrix of the ensemble simulations. We will provide an overview of our findings and potential solutions such as matrix thresholding.
|
