Subseasonal prediction fills the gap between weather forecasts and seasonal outlooks. There is a general understanding that the predictability on the subseasonal timescale comes from the atmospheric, land and ocean initial conditions, with predictability from land initial conditions being related primarily to slowly varying changes in soil moisture and snow pack, and ocean variability such as El Nino Southern Oscillation and the Madden-Julian Oscillation providing potential predictability. Here we use a unique suite of subseasonal reforecast experiments to quantify the role of atmospheric, land, and ocean initial conditions on subseasonal predictability. We find that in a global average the majority of predictive skill in this window for surface temperature comes from the atmosphere, and the land becomes the largest contributor after week 3. The contribution from the ocean initial state is very small all the way through the week 6 forecast. These findings however vary by region, with surface temperature predictability in South America coming primarily from the land initial state throughout the entire 45 day forecast period. Precipitation prediction skill however comes primarily from the atmospheric initial state, except for few regions for which the ocean initial condition becomes important.