Finding reference: a comparison of modelling approaches for predicting macroinvertebrate community index benchmarks

Abstract

Reference benchmarks are needed to assess the contemporary status of rivers and to establish restoration targets. We developed predictive models to estimate site‐specific reference values for a macroinvertebrate community index (MCI), which is used to indicate a range of human impacts on wadeable streams. We compared three statistical modelling approaches – general linear, boosted regression tree (BRT) and random forest (RF) – and tested the effect of spatial scale on predictive accuracy by developing national and regional BRT models. Using fitted flexible models (BRT, RF) and resetting predictors to reflect natural state provided the most accurate predictions of reference condition. Variation in reference MCI predictions from national and regional models was within the range observed from methodological and temporal variability. The proportion of native vegetation in upstream catchments was the primary predictor of MCI scores in all models, while secondary predictors varied regionally.

Bibliographic Information