The use of matrix models to detect natural and pollution-induced forest gradients

We sought to compare the efficacy of the stationary Markov model and conventional ordination techniques in describing compositional and structural changes in forest communities along natural and manmade spatial gradients at two scales, local and regional. Vegetation abundance and structure data are from six sites spanning a spatial gradient in the Great Lakes-St. Lawrence forests near Sudbury, Ontario, Canada. Ordination did not detect slope-related local gradients despite the general trend that, as distance from the pollution source increases, vegetation along the slopes begins to display Markovian spatial dynamics. We suggest that this is due to information loss resulting from static ordination analyses: information regarding transitions between observations along the natural ordering of quadrats is not maintained. Both ordination techniques and the Markov analyses detected strong regional pollution-induced gradients in abundance and structure.