Mapping habitat indices across river networks using spatial statistical modelling of River Habitat Survey data

• Published in: Ecological indicators 2016-07, Vol.66, p.20-29
• Main Authors: Naura, Marc, Clark, Mike J., Sear, David A., Atkinson, Peter M., Hornby, Duncan D., Kemp, Paul, England, Judy, Peirson, Graeme, Bromley, Chris, Carter, Matthew G.
• Format: Article
• Language: English
• Subjects: altitude; Channel substrate; correspondence analysis; experts; freshwater; freshwater ecosystems; Geostatistics; Habitat indices; Habitat mapping; habitats; kriging; laws and regulations; least squares; monitoring; prediction; Regression kriging; River Habitat Survey; rivers; statistical models; surveys
• ISSN: 1470-160X; 1872-7034
• DOI: 10.1016/j.ecolind.2016.01.019

•Maps of habitat distribution are needed to achieve environmental targets.•Existing monitoring data can be used to predict habitats using geostatistics.•A map of channel substrate is produced for all English and Welsh rivers. Freshwater ecosystems are declining faster than their terrestrial and marine counterparts because of physical pressures on habitats. European legislation requires member states to achieve ecological targets through the effective management of freshwater habitats. Maps of habitats across river networks would help diagnose environmental problems and plan for the delivery of improvement work. Existing habitat mapping methods are generally time consuming, require experts and are expensive to implement. Surveys based on sampling are cheaper but provide patchy representations of habitat distribution. In this study, we present a method for mapping habitat indices across networks using semi-quantitative data and a geostatistical technique called regression kriging. The method consists of the derivation of habitat indices using multivariate statistical techniques that are regressed on map-based covariates such as altitude, slope and geology. Regression kriging combines the Generalised Least Squares (GLS) regression technique with a spatial analysis of model residuals. Predictions from the GLS model are ‘corrected’ using weighted averages of model residuals following an analysis of spatial correlation. The method was applied to channel substrate data from the River Habitat Survey in Great Britain. A Channel Substrate Index (CSI) was derived using Correspondence Analysis and predicted using regression kriging. The model explained 74% of the main sample variability and 64% in a test sample. The model was applied to the English and Welsh river network and a map of CSI was produced. The proposed approach demonstrates how existing national monitoring data and geostatistical techniques can be used to produce continuous maps of habitat indices at the national scale.