Forests cover nearly a third of the Earth's land area and exchange mass, momentum, and energy with the atmosphere. Most studies of these exchanges, particularly using numerical models, consider forests whose structure has been heavily simplified. In many landscapes, these simplifications are unrealistic. Inhomogeneous landscapes and unsteady weather conditions generate fluid dynamical features that cause observations to be inaccurately interpreted, biased, or over-generalized. In Part I, we discuss experimental, theoretical, and numerical progress in the understanding of turbulent exchange over realistic forests. Scalar transport does not necessarily follow the flow in realistic settings, meaning scalar quantities are rarely at equilibrium around patchy forests, and significant scalar fluxes may form in the lee of forested hills. Gaps and patchiness generate significant spatial fluxes that current models and observations neglect. Atmospheric instability increases the distance over which fluxes adjust at forest edges. In deciduous forests, the effects of patchiness differ between seasons; counter intuitively, eddies reach further into leafy canopies (because they are rougher aerodynamically). Air parcel residence times are likely much lower in patchy forests than homogeneous ones, especially around edges. In Part II, we set out practical ways to make numerical models of forest-atmosphere more realistic, including by accounting for reconfiguration and realistic canopy structure and beginning to include more chemical and physical processes in turbulence resolving models. Future challenges include: (a) customizing numerical models to real study sites, (b) connecting space and time scales, and (c) incorporating a greater range of weather conditions in numerical models.