We evaluate the impact of adding two-way coupling between atmosphere and ocean to the Met Office deterministic global forecast model. As part of preoperational testing of this coupled NWP configuration we have three years of daily forecasts, run in parallel to the uncoupled operational forecasts. Skill in the middle and upper troposphere out to T + 168 h is generally increased compared to the uncoupled model. Improvements are strongest in the tropics and largely neutral in midlatitudes. We attribute the additional skill in the atmosphere to the ability of the coupled model to predict sea surface temperature (SST) variability in the (sub)tropics with greater skill than persisted SSTs as used in uncoupled forecasts. In the midlatitude, ocean skill for SST is currently marginally worse than persistence, possibly explaining why there is no additional skill for the atmosphere in midlatitudes. Sea ice is predicted more skillfully than persistence out to day 7 but the impact of this on skill in the atmosphere is difficult to verify. Two-way air–sea coupling benefits tropical cyclone forecasts by reducing median track and central pressure errors by around 5%, predominantly from T + 90 to T + 132 h. Benefits from coupling are largest for large cyclones, and for smaller storms coupling can be detrimental. In this study skill in forecasts of the Madden–Julian oscillation does not change with two-way air–sea coupling out to T + 168 h.
In many operational weather forecasts, interactions between atmosphere and ocean are simplified to acting in one direction only: the ocean can affect the state of the atmosphere but the ocean itself is insensitive to changes in the atmosphere. We investigate the impact on forecasts out to day 7 when coupled interactions are allowed in the Met Office global forecast model. Benefits of coupling are greatest in the tropics and give a 5% reduction in track error of tropical cyclones during days 4–7. In midlatitudes, the effect of coupling is neutral, probably because the resolution of the ocean model we use is insufficient. Our study shows that interactive air–sea coupling improves weather forecasts and suggests further ways to improve coupled forecasts.