Over the last a few decades tropical cyclone (TC) track forecasts have improved significantly, whereas very little progress made in TC intensity forecasts. The lack of the skill in the intensity forecasts may be attributed to deficiencies in the current prediction models: insufficient horizontal resolution, inadequate surface and boundary layer formulations, and no full coupling to the ocean. The extreme high winds, intense rainfall, large ocean waves, and copious sea spray push the surface-exchange parameters for temperature, water vapor, and momentum into untested new regimes. To resolve the TC eyewall structure, crucial in intensity forecasting, the horizontal resolution need to be at ~1-2 km. The air-sea interaction in the eyewall region is largely unknown with very little observations. While TCs draw energy from the ocean surface, they cool the ocean by wind-induced surface fluxes and vertical mixing. The enthalpy and momentum exchange coefficients under the high-wind conditions are difficult to determine. The stress is supported mainly by waves in the wavelength range of 0.1-10 m, which are unresolved by wave models. Rapid increase in computer power and recent advance in technology in observations have made it possible for us to develop a strategy for the next generation of high-resolution TC prediction models. We begin by examing key parameterizations including effects of the wave spectral tail on drag coefficients, the source term for sea spray, and subgrid-scale turbulence property at 1-2 km resolution. The components of the coupled model system are the PSU/NCAR MM5, WAVEWATCHIII, and the University of Miami HYCOM. Model simulation of the Hurricane Floyd (1999) is compared observations of surface wave spectra, surface fluxes, and vertical profiles of atmospheric boundary layer and ocean mixed layer from recent hurricane field programs. Remotely sensed SST, surface winds, and rainfall are used to evaluate and validate model simulations, develop parameterizations of the air-sea interface, and initialize the models over the open ocean in a data assimilation mode.