Given the relatively slow evolution of the ocean, it likely holds the key to North American climate predictions on sub-decadal and longer time scales. We plan to use empirical models trained on multiple variables (SST, thermocline depth, MOC strength, etc.) from ocean assimilation products to forecast the global ocean and its impact on North America. The forecast system will also include surface air temperature and winds, both to improve ocean forecasts and to predict societally relevant quantities. The same approach will also be applied to coupled climate model simulations to identify model errors, determine the processes responsible for predictability, and investigate the extent to which global climate change influences the predictability of the oceans. Our primary forecast method will be linear inverse models (LIMs), which are currently used operationally to predict SSTs in the tropical oceans. We have recently extended the LIM prediction system to include thermocline depth and surface winds, which has improved ENSO predictions at longer leads and encouraged us to explore predictions at decadal time scales. Forecasts will be made on a seasonal basis for at least two years ahead, and on an annual basis for at least five years ahead. In addition to providing skillful forecasts, LIM also allows exploration of important aspects of the dynamical system, including processes that give rise to rapidly growing and/or persistent anomalies and limits to predictability. This information is particularly useful for decadal prediction, since not only does it help determine the construction of an initial ensemble for climate model runs, but it also helps show where observations are most needed to reduce forecast error growth. We will also evaluate the dependence of North American forecasts on information from different regions, investigating linkages between the Atlantic and the Pacific, and tropics and the extratropics.