PARENT SESSION
Oral Session 16: Statistics and Biometrics.
Presiding: E Garton and T Simons
Monday, August 2, 8:00 AM to 11:30 AM, Meeting Room D 139.

Multivariate techniques for monitoring, environmental impact assessment and ecological analysis.

Anderson, Marti^*,1, ¹ University of Auckland, Auckland, New Zealand

ABSTRACT- Ecological systems are complex. Detecting and characterising unusual changes of whole sets of species in response to experimental treatments, physical variables or environmental impacts poses a fundamental challenge to ecologists. Traditional multivariate statistical techniques tend to be inadequate for dealing with most ecological data sets, due to their lack of robustness to violations of their assumptions (such as normality or homoscedasticity). On the other hand, more modern non-parametric techniques (such as ANOSIM) cannot be applied to complex experimental designs, as they cannot be used to achieve a partitioning or test interaction terms. Recent developments in statistical ecological methods include the emergence of several new powerful tools which can be used for the simultaneous analysis of many species in an ecosystem, including in response to complex experimental designs. These use a distance-based approach, with no limits on the distance measure used (unlike CANOCO) and with tests provided by permutation, bootstrap or Monte Carlo resampling methods. The purpose of this talk will be to briefly describe the most recently developed techniques and, more specifically, to present their use and specific insights found by their application in: (a) the assessment of ecosystem health for estuaries across the Auckland Region, New Zealand (b) the monitoring of coral reef fish assemblages across the entire Great Barrier Reef, Australia and (c) the detection and characterisation of impacts on Mediterranean rocky reefs.

Key words: environmental impact, monitoring, multivariate analysis, experimental designs