Robust Estimates of Earth System Predictability of the First Kind Using the CESM2 Multiyear Prediction System (CESM2-MP)

Abstract Here, we present a new seasonal-to-multiyear Earth prediction system, Community Earth System Model, version 2, multiyear prediction system (CESM2-MP), based on the CESM2. A 20-member ensemble that assimilates oceanic temperature and salinity anomalies provides the initial conditions for 5-yr predictions from 1960 to 2020. We analyze skills using pairwise ensemble statistics, calculated among individual members (IMs), and compare the results with those obtained from the more commonly used ensemble-mean (EM) approach. This comparison is motivated by the fact that an EM of a nonlinear dynamical system generates, unlike reality, a heavily smoothed trajectory, akin to the evolution of a slow manifold. However, for most autonomous nonlinear systems, the EM does not even represent a solution of the underlying physical equations, and it should therefore not be used as an estimate of the expected trajectory. The IM-based approach is less sensitive to ensemble size than EM-based skill computations, and its estimates of attainable prediction skills are closer to the actual skills. Using IM-based statistics helps to unravel the physics of predicted patterns in the CESM2-MP and their relationship to ocean–atmosphere–land interactions and climate modes of variability. Furthermore, the IM-based method emphasizes predictability of the first kind, which is associated with initial error sensitivity. In contrast, the EM-based method is more sensitive to the predictability of the second kind, which is associated with the external forcing and time-varying boundary conditions. Calculating IM-based skills for the CESM2-MP provides new insights into the sources of predictability originating from ocean initial conditions, helping to delineate and quantify the forecast limits of internal climate variability. Significance Statement The purpose of this study is to introduce a new seasonal-to-multiyear Earth prediction system and present a robust method for determining predictable patterns associated with ocean initial conditions and internal climate variability. The conventional ensemble-mean-based approach utilizes smoothed information and captures a larger portion of the predictability related to external forcing and time-varying boundary conditions. We introduce the individual-member-based approach, which uses unfiltered information and emphasizes the predictability originating from the initial conditions. Therefore, the individual-member-based predictability provides more systematic and confident information associated with the ocean–atmosphere–land interaction. Our results suggest a method for quantifying forecast limits related to internal climate variability for the prediction community. Additionally, our work offers practical and reliable guidance for end users in their decision-making.