Forest canopies interact with the atmosphere by emitting heat and moisture fluxes, by dragging the flow and by forming obstacles to the flow. Forests are heterogeneous with structural features at a vast range of length scale. The atmospheric effects of micro-scale canopy structures, which describe differences between individual trees, have so far been poorly studied. Changes to turbulence, flow patterns, and fluxes in and above the canopy strongly affect the dispersal of seeds and its ecological consequences because they are strongly dependent on the far "tail" of the dispersal distribution.

The Regional Atmospheric Modeling System (RAMS) is further developed to operate as a large-eddy simulation (LES) at high resolution with 3D heterogeneous forest canopies. This RAMS-based Forest LES (RAFLES) represents the canopy through drag, volume restriction by stems, and heat and moisture fluxes in the canopy domain. The model incorporates explicit canopy descriptions, which can be obtained from observations, or from the virtual-canopy generator, which is developed here.

RAFLES is used to simulate noontime conditions for two days at the hardwood stand in the Duke Forest, representing two sets of atmospheric and canopy conditions. The results are evaluated against eddy-flux observations from these days. RAFLES compares well to the observed data. Comparison between artificial homogeneous cases and natural heterogeneous cases reveals that small-scale canopy heterogeneity affects the profiles of momentum and scalar fluxes, and modifies the spatial structure of the flow. Low areas in the canopy promote ejection events, which leads to a correlation between the canopy height and flow variables that extends up to four times the canopy height.

Seed dispersal kernels simulated with RAFLES closely match those measured in seed release experiments in a temperate forest. It is also used to examine potential biases resulting from simplifications in common dispersal models, such as planar-homogeneity in canopy morphology and the flow field, neglecting the effects of seed inertia. RAFLES shows that the statistical attributes of the dispersal kernels are affected by different processes operating from the seed scale to the scale of the atmospheric boundary layer. Seeds dispersed from these areas have higher probability of undergoing long-distance dispersal.