|Institution:||University of Otago|
|Keywords:||didymo; Species Distribution Modelling; Pfankuch; New Zealand; rock snot; potential distribution; niche modelling; didymo bloom formation; AICc; Maxent; Boosted Regression Trees; Logistic Regression; distribution model; invader; Didymosphenia geminata|
|Full text PDF:||http://hdl.handle.net/10523/5030|
The diatom Didymosphenia geminata (didymo) is recognised as a nuisance bloom-forming species in freshwater systems globally. In New Zealand, didymo has invaded many rivers in the South Island but, to date, has not established in the North Island. Research suggests that didymo bloom formation in rivers is controlled primarily by ambient dissolved reactive phosphorus concentration and flow regime. The purpose of this study was to provide a better understanding of the relationship between streambed stability and didymo bloom formation. In the first part of my thesis, I performed a survey at the reach scale (50 metres along the bank) in forty rivers across the South Island of New Zealand comparing didymo standing crop and cover proportion to the Pfankuch index, a qualitative bed stability assessment tool. Pfankuch values, dissolved reactive phosphorus concentrations and turbidity measurements were compared as predictors of didymo standing crop and cover using an information-theoretic approach. For the nine sites with visible didymo blooms, Pfankuch bed stability was the best predictor (in terms of model likelihoods) of didymo standing crop and cover. These results suggest that streambed stability – evaluated using the Pfankuch index – is an important environmental variable controlling the formation of didymo blooms. Previous attempts at modelling the distribution of didymo have focused on the potential for didymo cells to survive in waterways rather than the potential for blooms to form. In the second part of my thesis, I modelled the relative suitability of river segments in the New Zealand river network (mean length = 672m, SD=642m) for didymo bloom formation. I used three distinct distribution modelling algorithms (logistic regression, boosted regression trees and Maxent) and compared the results in terms of model structure, model performance and model predictions. I found that the choice of distribution modelling algorithm had a minor influence on the model predictions at a regional scale but a much greater influence at the river segment scale. The models produced reliable predictions for the South Island and indicated that highly susceptible river segments are concentrated in the Otago, Canterbury and West Coast regions. Additionally, the predictions suggest that a majority of river segments in the North Island are likely to remain free of didymo blooms. Any didymo management strategy should therefore consider the suitability of rivers for didymo bloom formation.