Comparison of Markovian matrix models of a primary successional plant community

Markov matrix models have been used to simulate a variety of dynamic ecological systems such as communities and landscapes. An important question is whether time-homogeneous Markov models are adequate for successional dynamics or whether non-homogeneous models are needed to reflect changes in species composition over time. We evaluate some alternative Markov model formulations, ranging from homogeneous to semi-homogeneous, to see which might be useful in replicating observed vegetation dynamics in a primary successional plant community at Mount St. Helens, Washington. We used a two-step multivariate process to classify vegetation in 1×1 m quadrats in a 12×14 m study plot which was surveyed annually from 1983 to 1994. Four different Markov models were derived to simulate successional dynamics at the quadrat level, ranging from a completely time-homogeneous model based on quadrat class transitions pooled across all survey years, to one with annual constraints on classes allowed and with annually-adjusted transition probabilities based on these allowed classes. Although none of these models were particularly successful in replicating the observed number of vegetation classes or the number of quadrats in each class across all the survey years, the models with more annual constraints produced better results than the completely homogeneous model. The generally poor showing of these models relates primarily to annual variations in species dynamics in this community. The substantial turnover in species each year results in rapid turnover of vegetation classes, so that transitions among classes vary greatly from year to year. It appears that annual influences, both biotic and abiotic, must be implemented in the models for accurate simulations in this and probably most other primary successional communities.