Maximum-likelihood ordination — some improvements and further tests

A modification to a previously described maximum likelihood approach to non-linear ordination is presented. It involves weighting quantitative measures of species importance to ensure that changes in values along a gradient are given equal weight irrespective of their magnitude. The maximum likelihood method was tested with artificial data based on a model incorporating various types of non-linear species responses. Symmetrical and asymmetrical response surfaces, with and without central depression, were used and the method was shown to be reasonably robust. In tests on both real and artificial data it proved superior to other methods.