Modelling seed dispersal to predict seedling recruitment: Recolonization dynamics in a plantation forest

Abstract

The extend to which seed dispersal influences seedling recruitment is of major importance for forest dynamics. In non-equilibrium situations, seed dispersal might be of even further importance as it is a major determinant of the rate and composition of secondary succession. We modelled primary seed dispersal in a planted pine–spruce stand intensively recolonized by Abies alba and Fagus sylvatica, an ecological situation commonly found throughout southern Europe. We then evaluated the role of primary seed dispersal in seedling emergence and recruited seedling density. Using a seed trap experiment and inverse modelling methods, we calibrated short- and medium-distance seed dispersal functions. The relationship between the density of dispersed seed, the density of emerging seedlings and microsite characteristics were assessed using generalized linear models. The cumulative distribution function of seedling-to-nearest-adult distances made it possible to test the concordance between seed rain patterns and seedling spatial distributions. Seed production was highly variable between years for A. alba and F. sylvatica. Seed dispersal was shown to be locally restricted (median dispersal distance of 13.2–19.2 m for A. alba depending on the year and 6.49 m for F. sylvatica). Model prediction was considerably increased when seed production was directly assessed (and not indirectly estimated using diameter, for example). The number of A. alba seedlings that emerged in 1999 was positively correlated with the number of seeds dispersed in 1998 and with the local density of Picea abies adults, and negatively correlated with grass cover. The spatial distribution of seedlings was less aggregated around adult trees than expected from seed dispersal models, but significantly different from random beyond 6 m to the nearest adult for F. sylvatica and 26 m for A. alba. Thus, seed rain patterns are only partially responsible for recruitment dynamics in our model forest. Our study demonstrates that inverse modelling methods are well suited for the study of seed dispersal at the local scale, especially when a direct count of seed production can be made, and are therefore of particularly high interest in forests where several successional stages are present.