Characterizing climate predictability and model response variability from multiple initial condition and multi-model ensembles

Abstract: Climate models are thought to solve boundary value problems unlike numerical weather prediction, which is an initial value problem. However, climate internal variability (CIV) is thought to be relatively important at near-term (0-30 year) prediction horizons, especially at higher resolutions. The recent availability of significant numbers of multi-model (MME) and multi-initial condition (MICE) ensembles allows for the first time a direct sensitivity analysis of CIV versus model response variability (MRV). Understanding the relative agreement and variability of MME and MICE ensembles for multiple regions, resolutions, and projection horizons is critical for focusing model improvements, diagnostics, and prognosis, as well as impacts, adaptation, and vulnerability studies. Here we find that CIV (MICE agreement) is lower (higher) than MRV (MME agreement) across all spatial resolutions and projection time horizons for both temperature and precipitation. However, CIV dominates MRV over higher latitudes generally and in specific regions. Furthermore, CIV is considerably larger than MRV for precipitation compared to temperature across all horizontal and projection scales and seasons. Precipitation exhibits larger uncertainties, sharper decay of MICE agreement compared to MME, and relatively greater dominance of CIV over MRV at higher latitudes. The findings are crucial for climate predictability and adaptation strategies at stakeholder-relevant scales.