A field test of inverse modeling of seed dispersal.

Premise of the Study: Seed dispersal distance-a key process in plant population dynamics-remains poorly understood because of the difficulty of finding a source plant so well isolated from conspecifics that seeds or seedlings can be unambiguously attributed to it. Inverse modeling (IM) of seed dispersal, a simple statistical technique for parameterizing dispersal kernels, has been widely used since 1992; surprisingly, however, this approach has never been verified in the field.
Methods: We released from 20 nearby trees the winged seeds of a liana species, Entada polystachya, near the coast in a tropical, dry forest in Jalisco, Mexico.
Key Results: With a two-parameter log-normal function, we found that IM predicted both the shape and scale parameters well as long as we used the entire data set. When, however, we subsampled (thus simulating the use of transects for seedlings or an array of seed traps), the estimates of the scale and shape parameters were often more than double the real values. The problem was due to the marked anisotropy (directional bias; in this case, in the direction of the diurnal sea breeze) of the individual dispersal curves. When we randomized the direction of dispersal of individual seeds from the trees (keeping dispersal distances unchanged), predictions of parameter values were excellent.
Conclusions: Inverse modeling must include directional parameters when dealing with areas where strong anisotropy is to be expected, e.g., for wind dispersal of seeds near coasts or pollination by any vector where a plant species is limited to a strongly linear habitat such as river banks.