Skillful subseasonal-to-decadal (S2D) prediction depends critically on the quality of model initial conditions. This session invites contributions that explore how initialization strategies, including reanalysis-based approaches, forced ocean–sea ice simulations, data assimilation, leveraging emerging in-situ and remotely sensed land/ocean/atmosphere observations, and AI-based methods, shape S2D prediction skill and advance our understanding of Earth system predictability. We welcome studies that assess prediction skill for modes of variability and high-impact extreme events (e.g., hydroclimate, temperature extremes, tropical weather), as well as process-level investigations of how initial conditions drive improvements in prediction and forecasts of opportunity. Contributions addressing the predictability of regional extremes, the role of land and ocean memory and land-atmosphere-ocean interactions in sustaining prediction skill, and emerging AI or hybrid frameworks for improving initial conditions are also encouraged.