The doors to our Whitianga and Paeroa offices will be closed for the summer break from 4pm on Friday, 20 December, while our Taupō and Hamilton offices will close for the summer break at 1pm on Tuesday, 24 December. All offices will reopen on Monday, 6 January 2025. To report air or water pollution, unsafe water activities in or on a river, lake or harbour, or make a general enquiry or information request during this time, call us 24/7 on 0800 800 401.

Close alert

A predictive model of fish distribution and index of biotic integrity (IBI) for wadeable streams in the Waikato region

TR 2006/07
  1. Services »
  2. Publications »
  3. TR2006/07

Report: TR06/07

Author: Mike Joy (Ecology Group & Centre for Freshwater Ecosystem Modeling and Management, Institute of Natural Resources, Massey University)

Abstract

A predictive model was developed for the wadeable streams (equal to or less than the fourth order) of the Waikato region using existing fish presence/absence data. The environmental variables used to generate the predictions came from two existing datasets that were collated in a GIS format from other existing databases. The predictive model was extensively evaluated and iteratively optimised to maximise predictive accuracy.

The levels of accuracy for the model were good to outstanding, and exceeded those from comparable North Island regional models. The predictions from the model were then expanded out over the entire regional stream network to give a predictive map of fish assemblages. These predicted fish assemblages were then used to create a predictive map of IBI scores for the wadeable streams over the entire region.

A Predictive Model of Fish Distribution and Index of Biotic Integrity (IBI) for Wadeable Streams in the Waikato Region [PDF, 861 KB]

Contents
Executive summary iii
1 Introduction 1
1.1 Background 1
1.2 Using artificial neural networks for spatial modelling 1
2 Methods 2
2.1 Predictive model construction 2
2.2 Data sources 4
2.2.1 Fish 4
2.2.2 Habitat 7
2.3 Model architecture and number of variables 8
2.4 Validation with independent data 8
2.5 Model evaluation 8
2.6 Quantifying predictor variable contributions 8
2.7 Sensitivity analysis 9
2.8 Predictive IBI 9
3 Results 9
3.1 Network architecture and variables 9
3.2 Model evaluation 9
3.2.1 Species comparison 9
3.2.2 Assemblage comparison 11
3.3 Predictor variable importance 11
3.4 Sensitivity analysis 14
4 Discussion 17
4 Waste in Waikato region 2003/05 20
4.1 Assemblage-environment relationships 17
4.2 Species-environment relationships 19
4.3 Limitations of the predictive model 19
4.4 Future data requirements 19
References 20
Appendix 1: Technical details on model construction validation and evaluation from Joy & Death, 2004 22
Appendix 2: FWENZ variables used in model construction 24