An approach to predicting future stand composition using standard ecological methods and multiple regression analysis is presented. A composition index value for each of 54 undisturbed, compositionally stable, mature stands was obtained by multiplying the importance value (relative basal area) of a species by its adaptation value and summing for all species in a stand. Adaptation values ranging from 1-10 were assigned based on the position of species maximum basal area values along a stand basal area gradient. Three multiple linear regression models were developed for predicting stand CI from effective soil depth, percent stone, available soil water, slope position, aspect and distance to the opposing slope. The models, which accounted for 90-97 % of the variation, were validated using data from seven compositionally stable stands not used in model development. The models closely predicted compositional change in a disturbed stand for which 48 years of permanent plot data were available. Successional trends were examined in six additional stands disturbed 50-70 years ago. In most stands, predicted stand composition closely coincided with the presence of established understory stems. The most substantial changes were found on N slopes and stream terraces where Quercus alba and Q rubra stands are being replaced by Acer saccharum and other mesophytic hardwoods. When applied to broad ridges that were once farmed, the models predicted that under presettlement site conditions (i.e., uneroded forest soil), stands were dominated primarily by Q. alba with small amounts of Q. stellata and Q. velutina. On sites where erosion has reduced soil depth to 25 and 15 cm, it was projected that stands would be dominated by Q stellata and Q. marilandica, and by Juniperus virginiana, respectively. A biological interpretation of the site-vegetation relationships, identified by the models is based on soil water storage and the influence of topographic factors on evapotranspiration loss. INTRODUCTION The successional process has been investigated for many years, yet there is no currently acceptable method for predicting potential forest composition given specific site conditions. Succession is defined here as the systematic temporal replacement of a forest community composed of relatively shade-intolerant species by one in which more shade-tolerant species dominate to form a compositionally stable community. Usually stands of early successional species develop after moderate to severe cutting, fire, wind, disease or insect attack which removes the overstory canopy and permits more lightdemanding species to invade and dominate sites where they normally are excluded. Considerable research has concentrated on community dynamics with a view toward predicting the species mix in compositionally stable (climax) forest. Stephens and Waggoner (1970, 1980) developed transition probability values that permitted anticipation of species change in forest composition, whereas Leak (1970) predicted successional change by birth and death simulation. In both studies, predictions were based upon systematic sampling over a long period. Stout et al. (1975) constructed multispecies models that indicated trends toward certain forest species (e.g, Acer saccharum). Goff (1968), Goff and Zedler (1972), Zedler and Goff (1973) and Auclair and Goff (1974) used several size association and stratification analyses to determine species succession vectors, replacement potential and forest trends. Botkin et al. (1972a, 1972b) developed a computer-generated model JABOWA) which simulated forest composition changes using Monte Carlo techniques to decide births and deaths over any selected period; this
Read full abstract